Home
cover of episode 260: CPUs Have Been Juicing for Too Long

260: CPUs Have Been Juicing for Too Long

2024/11/10
logo of podcast Brad & Will Made a Tech Pod.

Brad & Will Made a Tech Pod.

Chapters

The episode discusses the latest advancements in CPU design, focusing on AMD's Zen 5 and Intel's Arrow Lake architectures. It explains the concept of disaggregated CPUs and how they differ from traditional monolithic designs.
  • Disaggregated CPUs use multiple smaller chips connected together, allowing for different manufacturing processes and increased core counts.
  • Intel's Arrow Lake uses TSMC's N3 process for its chiplets, marking a significant shift from Intel's traditional in-house fabrication.
  • AMD's Zen 5 continues the trend of disaggregated designs, improving upon previous generations with better performance and efficiency.

Shownotes Transcript

Translations:
中文

Okay, as mentioned last week, there was one thing the way forgot to touch on about my time back home, and that .

was A B Q sauce barbecue is, yeah, I mean, near a muster y guides. Forgive a bommel at a pass on this case.

That was a very poor attempt of .

the child.

Barcus say, barker.

So, yeah.

OK a power tools, power tools.

I love power tools. Power tools .

Operated by barbecue. E I mean, I mean.

batteries seems like a big power out of battery based power tools. Probably not. The ideal was after the .

power came back OK for the most part but and it's not my first time like like they have Better Operated long and i've cut grass before, so it's not my first time touching all this.

So really okay.

but battery or leaf lower and this was the big one. In fact, this is really most of ly, i'm bringing this up. Battery powered chains all wait. Really yes.

sounds nice. Does that just make a good worry?

Cutting sound and this no, like, yeah, it's it's still like later than you might think OK, but it's not like that gas engine like extreme wildness. Uh, I don't know. We were we just got a couple lens down over the driveway and I cut up. So like I wasn't using IT. I bet I ran IT total like ten, fifteen minutes.

That's Better than I would have expected.

I guess i'd actually maybe the cool thing about all this stuff is, and I guess like my parents, like one of the way maybe to make this happen, but the batteries are all basically yes.

If you buy everything from the same company, you can make that happen. I I my electronic stuff is all handles because we don't. We pay somebody .

to model on yeah and of course, I still have a dump a bunch of oil and into the chains .

off to like I I have to assume this isn't like a logging style chains. This is like a trimmed up stuff around your house chains.

It's a handheld. It's got the same form factor as you would think of a chance of having. But yeah, it's like it's like so what side on the smaller side? Like it's probably you're not going to cut the whole street down with IT.

probably. So we do have an electron mower, but it's plug in, which is why I pay people to model on because having at two hundred food extension court is kind of pain the ass yeah .

it's plugin while Operating. I can only runs off well.

yes, so I basically I put the plug and I chock the court right over my shoulder and I just dull out cord and roll the cord back as i'm going.

Seems like the risk of running over the cord with philon might make things complicated.

I think that's prime mistake .

you'd only make once. Yeah, fair. Well, yeah. Think you've using long or again after that.

So how was your chainsaw good?

I like I, I can use the chains all that much in my life, but they're terrifying.

Yes.

partially because i'm afraid of them. And I like, Frankly, I was a little afraid of this one.

but there's a guy down the street for me that I see he is a bunch of like, scrub trees. And he's often up there with the chain saw working on on the scrub trees. And he does IT wearing shorts. And i'm just like, man, that is, that is a level of of lack of fear of the tool that I am. I do not understand.

but I I was also .

wearing shorts OK.

I guess that enough gentlemen might make a difference. Some guy was going to say if the chains off slips badly enough.

I don't know. The layer of cloth is leather s on ite into into the blade make plate mail airport they make you can get like you can get like like cvo sh gloves and something like that. But um yeah i'm i'm interested in the electric chain. So that's an interesting idea is that one of those ones, it's like also you can put on the end of a long stick and do like conversion on like trip, trimming branches and stop.

Now this this was like I said, regular like the handles are in the place you expect them to be. Okay, it's just looks like a chain or maybe a little smaller. Oh, okay, yeah, it's fun. I'm kind, kind to IT.

I mean, look, you get a chance how you go cut down tree. You you live near enough the park, you can cut down three anytime you want, right?

Welcome to brand. We'll made to check. But I will. Oh, I D I just.

I fixed.

I worked on my compressor of the other day so I can, I can do that. Can yet wild up in here .

in a lot of live. I was just looking right before this. I think i'm finally going to buy that. What do we want? Call A A channel trip a preprocessing.

It's a magic thing that lets you yell into your microphone and nuclide that's .

I blow out audio and make people hate the way you sound.

Does that have a gate?

也是 的, has a nice day. It's got a compressor. It's a um but it's got A D S R on its oh like one .

other thing.

It's A D B X. It's one the people the discord recommended talking about have an analog ico phone like an excel. Think we've talked about this before. But yeah, you can run that, mike, through a thing that kind of cleaning up and massage is the audio before IT comes in to the computer.

which is days when you do the live stuff and turn out.

yes, well, actually. So ironically, you've got a bunch of O B S has got some pretty good life. Filters are a lot of the stuff were talk non. So actually, my O B S.

when I stream, is like fairly well cleaned up now. But you gotta have the CPU power run stuff, just a box IT mac.

but also it's a safeguard for these recordings, because I die clip on our recordings way more than I would like well.

and also having the noise get, like when i'm doing podcast at this or something. And it's and we have somebody who has a noise gate. First is somebody doesn't have a noise gate.

IT is a noticeable difference here. It's funny to me that there is not mean IT. Well, okay, your voice call stuff typically handles this.

So exam and google meat and teams and all that stuff kind of handles this kind of processing automatically. But IT also makes your idea sound of look craps. You don't want to record that for podcast purposes.

We want to sound of nice for all. No um but the i'm surprised that there's no like software stack that works like the O B S virtual camera that you can put between you man. I mean, I guess the video broadcast as this, but IT also has some downstream consequences. Um so anyway, maybe this something out there we just know that.

But I actually well, I like I was looking for like real time recording plugins, like VS style plugins, like like I just don't think that exists or at least you need way quality .

soft using or or I mean, you voice matter does some of that stuff to, but voice matter has a so unique set of problems. Now just check a box in line .

and do the here's a .

bunch in that and jobs get turning bread. Speaking of the utilization and .

turning dials and knobs and .

turning dials and knobs, it's it's finally, the last of of this year's CPU models are updated. Intel, an AMD, both have released entirely well updated CPU lines, updated architecture. Intel's were not entirely new architecture this year. Everything's this aggregated now.

So if you don't know what that means, that means that inside your CPU package that you plugged in your mother board, there's A A, A chip that has a bunch of other chips, glue to IT smaller chips and those chips can be made on different processes. They handle different parts of of the CPU job. And um it's kind of changed the way CPU work over the last five, six years. Yeah traditionally they .

would fabricate one ship, and that one ship have literally everything that the CPU did on IT.

Yes, so so that traditionally have a way for r. It's a big piece of silicon, special silicon. They carve some stuff into IT. They do some batches. And all sports are really toxic chemicals and gold and business.

And then there's a bunch of little chips on there and they cut each of those out of that of that big wafer and each of those little tips gets gets glued onto this package. And then IT becomes the C. P. Of your computer.

Now so in the past, he was just the one ship getting fabricated, the everything. But now, now, now they're fabricated bunch of different ships, like you said, sometimes are different sizes because not everything on the CPU. And that needs to be tiny necessarily. And then they get, yes, they get aggregated together to a package.

Yes, so aimed. He's been doing this for a long time now. This is their fifth generation of those. They started with the rise and years and years ago, and IT was a little hinky in the early glad I didn't you .

read stories about the first and second generation ons that sounded like a real mess.

Well in the first, first, second generation ons part of IT is that things that hadn't been a problem before. We're certain ly a problem like late and see between CPU A A B U course um and and amy did a really good job of reaching out to software developers and in OS Operating steam providers and a bunch of other people to kind of work around and work through those kinks until we're a pretty good place with that. Now in the end, d decide that is probably worth pointing out.

I mean, people might be asking, like what's the benefit of this stuff? Like I think this is a big part of what gotten too much higher core counts on c to use because previously, previously having a ton, of course, was the domain of like a big workstation ship and server parts and .

big clock speeds too. Like this is why clock speeds continue to improve this because the term al management at different, like you don't have to build an enormous eight billion transistor CPU all on the most expensive process in the world. You can do the parts that are not as expensive to build on process in this case is the is the kind of resolution of the chip matching. So how small the the little traces and stuff are on in the cp.

This also on the manufacturing side, this increases because some chips that come out of the fab or bad. And if the entire chip is one ship, then you have throw the entire chip away. But if it's a bunch of different ships, you can therefore, ally, you have more good chips to plug together. Now do they pass those savings along to the consumer? Maybe I mean.

the chips keep getting faster is the savings that we get as consumers for risk for now. So okay. So on the AMD side, the new chips are those. And five, this is the the rise in nine thousand series chips. And there's ability, sixteen core, eight core and six core versions of that on the, on the intel side.

So do they do a .

twin twelve? So what will talk about another? But on the intellect, this is arrow lake. Uh, these are called the core all tress s series two.

Those where they they also have dumped the I five, seven, nine, nine branding is totally gone .

well but they yes, but the seven in the I guess the seven and nine are on, yeah they have and they're using the same branding as on mobile now basically. So the difference is that the desktop parts have an s in the name where the mobile parts don't OK and then there's other other letters come up in the name. Like if IT has A K IT means that it's over clock able but has an f IT means there's no G P U.

Do they always use the word ultra that what I was getting at is effectively that drop the eye? I could still a core ultra and nine.

two ty five k for the top on part, but it's gone. The people just referred to these as the five fifty five, two forty five.

And what I was wondering know is does the ultra label to note anything in particular? Like does that mean like IT has graphics? Or does this or that just marketing?

I think core ultra means is the disagreed one core has around for a long time OK. Um I would have to look to be sure but like ultimately the brain doesn't just know this.

Consume your branding is always confusing. So you have to kind of like read some tea leaves to make sure you know what they are talking about.

Yeah, that so. And then the other wild card on all this is that A M. D. Does this thing with an enormous cash that they glew on to the chips to make them really fast for games. Basically, they're fine for other things in the past.

These are the X V D parts they just rolled out of the first of the xd parts for this generation that rise in ninety eight hundred. Um the tie. So we're going to unrated, unrated the union and peel the onion this week, and we're going to give you everybody the high level here first, and then you can draw down further if you want.

I don't know, we could just core straight to the center of the onion right now really quickly and just say, if you want a gaming PC, just get in nineteen eight hundred and three.

that's the answer. And if you care about a lot, a lot, of course, then the do a lot of, like workloads, stuff. T, that requires a lot of CPU cores.

The intel stuff is okay, but you might be Better off going with last year. You gotta look at the benchMarks a little bit and see what what the things that you do matter for. yeah.

And then the higher big core risks also are fine. yeah. So now why is all that the case?

Why is all that the case?

Well, let's start all all of these launches head problems is a well with the exception of the x depart, the rise and nine thousand when they came out a months and a half ago um they messed up some testing stuff that caused them as a cause. Everybody who's doing the benchMarks to get lower than expected results.

what when you say expecting you mean lower than AMD was expecting? Because like they provide number of testers and say, hey, we got this what we think you should be in yeah .

that we think of those is kind of sanity check check numbers so that when you're doing testing, if you have something that's way off of the like, for example, benchMarks like a cna bench, which is a heavily multiscreen ted benchmark, are really repeatable and an easy way to tell if there's something mess up with your file settings or your windows configuration or something like that. And we were seeing numbers that were like nine percent lower than the provided seven chic numbers, which is pretty unheard of.

IT was IT was bad. I mean, like you know, if you get into the slightly more contentious regions of youtube hardware reviews, like you restraints up seeing video titles just like and five sucks like rise in one thousand fail.

So there were couple of reasons that happened. Um one is that AMD was running their tests as the as the administrator account versus an administrator account which had never mattered before. But something changed with this release that caused the administrator account to have a pretty significant performance advantage over just running as an administrator.

I don't know why that is. They've stopped doing their sanity checks as the administrator. You shouldn't run as the administrator account in windows.

Please don't do that. What we're talking about like the like not a administrator enable the user account .

at all is some kind of talking about people named .

administrate which permission account yeah.

it's kind of like winning windows as root, but you shouldn't do that, right, right? It's not exactly the same anyway. Um so we we at PC world, we were doing our benchMarks and we saw numbers that were off significantly enough that we thought we had missed up our our testing process um and as a result, we held we ended up doing a live stream the day the day of the embargo we did not release our review at in parter tanks.

We thought we had met something and then we look at every else's numbers and they got the exact same results as we were like, well, we Better get this going. The upshot is ammi did oppose a couple weeks later explaining what what had happened exactly. They went through IT. There's also been some windows updates for twenty three h two and twenty four two that change the way branch prediction works. Um like they're not you should if you have one of these processors, including them for dual C C D chips, will talk about the minute ah you you'll get up pretty significant programs boost by running this stuff.

Yeah and I want to say I don't have IT in front of me, but I think didn't the backdoor this fix to even like I believe on those ten, like I think IT was pervasive enough they will make this. Shouldn't something we hold back for modern one? Do I think .

it's just twenty three, two, twenty four, five.

Maybe you be maybe you want to be on one on and then if you have one of these, but if if you care, branch prediction is the thing. Where do you know do you know much about modern CPU design? Is the thing in the pipeline where C, P, U, kind of free compute multiple outcomes for the instruction they are running come like in a control flow kind of way.

Like if then sort of thing, like like C P U, looks at what what instructions are being punched through IT and kind of goes ahead and does both versions of the math or whatever, and then switch back to the that and that, right, and then just takes whatever discards the unneeded result and goes the one that was correct. It's part of, it's part of hell. You know, performance has continued to increase over time.

Anyway, that's my, that's my lake, not a computer scientists understanding of IT, but but apparently something about ten, five, actually, i've got the quote here from a md five architecture, incorporate a wider branch prediction capacity and priorities and generations. And then they go on to basically say, like that's what they are optimizing around. And and yes, with this updates so IT is like a IT is a legit harbor n improvement. This is not some smoke and mirrors type .

of thing yeah no and if you install the updates for windows, then then you'll see the the same similar impact um IT and also worth munching IT impacted the chips with two compute dies, the dool ccd chips like the one thousand nine hundred and fifty and the ninety nine hundred more than the single C C D. Although he does show improvements in the single C D C C D chips when you when you installed anyway that's the till they are a now like you said, there were a lot of people saying as A M D screwed at the end of this um part of the issue I think with this is also that they locked the the single die chips to sixty five watts tdp, which is pretty low for a desktop part. I don't think that's ultimately a bad thing.

but that's music to my years, honestly.

Yeah it's uh the the downside is that the numbers were not with between these two things, between the fifty five war and the performance problems related to the to the brand prediction stuff and the admin account. They were coming in like nine percent under what they expected and those numbers looked exactly the same as says seven thousand series or somewhat worse in some cases since the release aimed is command said, hey, if you want to run the P B O mode, which is their precision boost overdrive, which is basically like building over clocking, it's a little more configurable than than like it's not as it's not.

And we open this up and turn the multiplier up or when we open this up and turn the bus up, it's I want to to lock you like run as fast as you can and keep the CPU under A D salza us, right? Like which if you have a big cooler attached, so IT means you can get some really nice performance boost, uh, without any danger to the chipper. Whatever they wrote and said, we're onto on a aranca for people who have uh flipped P B O on and cracked up their T D P on the ninety six and ninety six hundred to one hundred five watts, which is what the last gen parts shipped .

at yeah or actually was one hundred five. The eco de on four.

want to say for the high party, for the low and I IT was what a ship at the point is.

even one hundred five is not that high by the standards of modern C. P. U. So the fact that you can get very good performance at sixty five or hundred and five lots like a pretty damn huge improvement. And efficiency.

yeah, if if you care about efficiency, everything that we're going to talk about today is good. If you care about performance, yeah, I would yeah.

but you will get to tomorrow when we kind of wrap this up. But I think I think I think a general reset on efficiency to performance ratios was needed well.

especially after intel was burning the world down with directly .

and indeed briefly did that was then for as well. The point is, like everybody was playing too fast and loose with voltages and and power consumption to try to eat out like point .

one percent more benchmark. I, I, I compared this to the steroids, air and baseball the other day. Okay, well, you know, after you used to mark, require barry bonds and send me so so just tear them off constantly for years and years and years.

You know, maybe it's going to be a little more we you have more pitchy games for a few years here when I see the CPU have been ducking for too long, it's what i'm saying. So okay. So overall then five rising nine thousand is a kind of incremental advancement to then for they have the branch prediction uh enhancement. They add full of x five twelve support, which is a new instruction, said.

not in third and four head nav, x five twelve. Well, well, were getting run, getting out over my skis here. I think, I think he was like, maybe limited in some way.

IT was not a full implementation by understanding, I should say. I mean, this doesn't matter to consumers typically. But like when you say and five is an incremental improvement like on the benchmark s that's true. Understanding is that is actually is the first like new from the ground up architectures since we've moved to rise in though like then one, three, four was all like iterate vely improved on from the same design and in five I believe was I was a pretty much like a brand new microarrays cure so so it's the foundation .

for .

the next the next several several years of .

horizons is the point yeah yeah um so they um double cash, ben, with four thousand and four. They rearranged how things connect on the trip in some ways that make that make that possible. Um it's a foreign animal processes and all of these chips have twenty eight lanes of P I E, five out of the CPU, which is a .

lot of P C, I, E five lanes on four twenty comes to mind and much of that right. I had this in four for some months and I want as I feel like I was like twenty years on the twenty fifty I had, but I can't remember anyway. That's a lot of lanes.

That's that's a good, generous number of lanes. Now with only mother board makers won't devote all of them to end up to, i'll be happy. Well, I mean.

I think the point is you can get them up two and a sixteen lane PCI express slot and still have some stuff. Let maybe have two m tools that five uh P C I five or whatever. Um the the rising nine dull C C designs also get stuff that was limited to the X D V uh chips in the past, really.

So there's this is this is one of those hey, this is kind of good, maybe not for everyone features where if you have the right stuff installed on windows, when you do something that needs to to run on one. So okay, we should take a set back when you have these dl C C P designs where they have two eight core computer ze right? So you can get sixteen CPU cores with thirty two threads total between them because the hyper threats is enabled on these chips. Then you can uh if say a game is running part of its threads on one of these guys and part of the threats on the other, you see pretty significant performance impacts because of the cross dye communication being a slower be than intrade die communication, right.

which is an important understand this because when we get to until .

they're doing things differently here, yeah um with the extra chips because they put the cash on the dull C C D chips only on one of the C C. They would just turn off the die weren't using and a crank up the the, the. So the game has to run .

on the C, C, D, H. One of, like I said, and and I never ended this problem like this, never really seemed like an issue to me. But some games, the other, the other ccd would get on park and threats would kind of migrate over anyway. Like that was kind of something people frustrated with was like, okay, not not every game actually knows how to do this correctly and you're still ligation problems. But I I never really saw an issue with IT.

So what I said this is like these these aggregated designs are hinky and have required a fair amount of work on the part of I S, S and OS vendors to make everything work the way you'd expect. This is a perfect example of that, right? In order to get this to work, you have to have the right bias, the right chip set driver, you have to have windows game mode turned on, and the new version of the xbox APP.

And if that all happens, the x box APP says, hey, man, you're run in a game. I'm putting game mode. Game mode is OK if you're run in a game and we're in game old and we got to do this and IT turns off the the the slower C, C D.

critically theodoret ally, that entire stack will make that work. Like I said, I would see games. I would see the other C, C, D get on part in the whole games. Um sometimes but yeah the thing I never saw was I never got like significant hitching, which is the type of thing you would see when, yes, we're misbehaving and then stuff. So like you.

and part of the chAllenge is that windows windows in just fundamentally doesn't exactly understand what's going on in these chips, right? Windows at its core, things that all CPU are created equal here because that's so what they've been for a really long time scheme is hard.

Yeah, I turned out it's a problem in .

this in this design is also a problem on the entire design because they have the big course in the little course. And that's complicated for other to a minute.

Yeah like I I saw like i've got a i've gotten until chip with big and little cores and and my server yeah and because I am this kind of nerd, like I watched some of the like cynics kernal mAiling list development chatter because like intel has a ton of engineers that contribute to the links channel, for example.

Like every hard work company contribute to the link, cornel, that you still to this day see until engineers talking about tweak regressions like new X, Y and z things they are introducing, like they are still even on the linux s side, like working on kind of had a original scheduling between type of cars. Like it's a hard chAllenge to solve her problem to solve yeah and and getting the O S to know like which which applications need more priority, which ones can run. Like on other words, like Bobby.

I like it's it's not a it's not an easy thing. And it's to the point that like intel on the server side for data centers, partly for power reasons but partly just because IT makes some applications easier. Our building chips that are all p cores are all e cores, all big cores are all little cores for different applications. Sometimes is you just want something that has a whole .

shit load of threads yeah don't don't have some server part that's nothing but like a bajillion equi, a horrifying number of that's mostly .

to go in old data centers that have lower power requirements for each for each rack. But anyway, um okay, so two new chip sets are announced when they roll out these. They just came out a couple of years ago.

That's crazy. I know know AMD does the thing until never does, which is although I did do with seven hundred psych.

but I think that was intentional for .

th I think I was more were forced into IT. The A M D does the thing where you multiple generations of CPU can run on the same board. But I don't know that ever was seen IT where they didn't put out new motherboards at launch of the CPU IT was like two, three months of you had to buy an old last gen motherboard n to even use these things .

yeah so so A M five is the socket um it's gonna around, theyve said. We're going to support this until twenty twenty seven at least. I think twenty twenty seven plus is what this slide .

say that at least one more generation after these and five parts yet two maybe.

I mean, for what it's worth that they announced new aim for CPU this year. yeah. So like .

announce is three. I knew in three parts last month. I don't know about that. That's crazy. I think I saw that I be wrong.

But anyway, uh, the X, Y and x, each chip heads are out now. They all have us before they're specifying the us. Before chips, I believe, that are allowed to be used by the mother .

board venture for at least some subset of the boards in third part.

yeah they had a list of hear the controllers that are can be supported on their deck.

interesting. okay. Because the USB compatible was probably the biggest was the third voting attention to making that Better and more compatible.

That's very good. I think this is one of those situations where until builds a lot of their own, you go p chip sets and he does not.

Yeah I think that's probably the case because I ve ve had infinite Better experience with USB preferable on this intel machine.

Yeah I think like my board has intelligence es p PS. So I think h anyway um and PCI g five for graphics plus N V M E is standard across, like good for multiple and b me salts and the and the main graphic slot, anything five seven support as .

we we think about on the wifi seven episodes few weeks ago. But every x eight, seventy board, I look up wifi. I wonder what the market spread is on people using wifi on desktops.

It's surprisingly high.

would be my guess. Run 4。 yeah.

Faster memory is supported on the new chip sets. Well up to six, four meg transfers per second, four hundred days, four a transfer second be very low, 4 but four hundred and that's that's supported right?

That's like out of the box spec .

2 hundred that is jeet。

Like way out of spec. That's kind of over clock territories .

well and you have to turn some ratios down. So you're doing like the number transfers a run single channel. But this guy stuff you do that happened there.

sixty four hundred supported out of the box, seems like it's get getting there .

like officially rocks solids is the thing. There's four parts out now. I'm sure they'll feel out the stack more later as time goes on. Uh, the high end is the ninety and fifty x, which is a four point three digger hurts to five point having gig hts depending on boost and how many cause are winning all that for six hundred box it's sixteen cores with hypersonic ing something thirty two threads on two ccd two dies um and it's one hundred and seventy watch total tdp the ninety nine one hundred x which is four point four giger hearts to five point six year hurts is has been dropped to four hundred hundred and thirty dollars already don't know that's permanent or not. But that was the this one is under performing compared to the other people's park Price twelve course with type threat on two days.

The ninety seven hundred x is a single ccd three point eight to five point five digger hurts for three and thirty bugs um and that shipped at sixty five. Watch G D P. But that's one of the ones you can overcome.

Ck, one of five using P, B O and the honor the warranty on. And then the ninety six hundred x is a six core version of that same thing. Three point nine gigas to five point four gig hurts the two hundred and fifty box with high per threading.

Uh, sixty five 8。 T, D, P, again. Overcook able to one five. yeah. This time.

I know this is just branding, but IT seems like they have possibly now reserved the eight hundred label purely for the extra chips, because he used to be the core was A N eight hundred years he was a ah 48 hundred x seventy eight hundred x those were the non v cash uh eight core parts。 But IT looks like they've moved to seven hundred. I just just to make IT more obvious which one is the good gaming .

so we can talk about that the nine eight hundred now if you want xd now if you want because they kind of do a surprise, right? As until was rolling out their their arrow like CPU, they were like, hey, by the way, guys, we got next three d come but .

it's going to be here week after next.

almost like they planned at that way.

A suspicious timing is the aim. These stuff now, even though were kind of the progressing, the intel stuff, which will get back to the other thing I was to say work quick is that I still find the twelve core part of the weird here.

I like, hey, we have some guys that fail. Oh.

for sure. It's absolutely. I think it's a core C, C, D that they put in that thing, right?

I believe I they don't specify a six core or if they are only fabric .

one aid core C C D now and like they're the ones that have bad cores, they're just dropping down to six.

My guess is that they have ones that have bad course that failed that they dropped out of.

But I don't know. I'm sure that people have dead these things and at the maybe figure that out so much. Yeah, I know this is a matter to the average user. We're not we're not average users for huge erds, and I just am curious about this kind of thing.

But but the the bigger point though with the the twice core one is you're still you're still having the two ccd, so you still have worry about threads migrating across and had taken performance yeah but especially for games like councils are all eight cores now. So like I always feel like a six core, the six core died devoted to gaming and maybe sure changing you a little bit. I mean that these are not gaming parts.

To be fair. I think that the six core stuff made a lot more sense before the current and councils were out, right? Like I think I think that as we're looking at games that like hit all eight cores, if you like, there are a pretty substantial number of games that are hy that are multiple read enough that you you actually see A C P U cores and use now. And um I I would not probably buy one of those six course machines. If you're looking at playing games now, if you're just doing multi thread stuff, like it's a it's a pretty substantial Price drop from the .

ninety sixteen. Yes, if you're do and like video coding, a photoshop or like productivity stuff should be fine because those are way less late and see dependent on games.

Yeah like it's like but even like video encoding is less of a, less of a hey, I need a lot of CPU course. Thanks for these days because most people are doing that on the GPU. So IT feels like it's like rendering and stuff like that. Well.

if you are different doing like extreamly, you're on the GPU. If you're if you're encoding for a maximum quality, you're probably still on software codes.

I think be surprised.

suggest there are still the hardware n codes .

are much Better but yeah um anyway so that they need x 3d is a four hundred and eighty dollar M S R P。 It's thirty box more than the seventy eight hundred 3d at launch。 The seventy seventy is basically unable possible to find. Now they .

stop making them. No.

it's a new three D V cash design. The first theyve done that since .

they launched the three D V cash on.

So previously, the three D V cash was on the top of the chip, which meant IT was basically like an insulating layer between the hot part of the of the of the um disagreed ted chip, the people buried .

underneath the extra cash, which is a strange design choice.

Well, as a result, you couldn't run those quite as high o'clock. So if you had a single C, C, D design, like a rising seventy one hundred, for example, then you would just know that you going to sacrifice some some cock cock performance, some upper upper bounds of that, because the massive of amount of all three cash on this and the speed cash gave you enormous performance and and stuff. It's like memory latency sensitive, like games and things like that. Now if you had a dull cc design, you were in a weird situation where windows didn't know where to put stuff because you're non V H cores would be wave ter dramatically .

stuff because that's what I had. You was like six eight hundred megahit difference, I think, between the non v cash die and the one .

with yeah and and as a result, if windows put your your threads on the right cores, stuff was great. And if they didn't, you were running at the non optimal .

speed and and in fact, what you would see, I because I wash, you know test manager and stuff allied like you would see, I believe they parked the through the v cash die in windows. Morally, yes, because you want, you want all your windows tasks on the the die with a higher oxby. Yes, so exactly. Only you were only running on the a course of the faster the higher oxby die, unless I need to the extra eight cores like he was a weird we are split thing and worth pointing out here, the xd parts on and four, they launched eight, twelve and sixteen core parts all at once.

yes. And IT was six months later.

five months later. Well, was time later, right? But but the point is they have only launched the a core part now .

that people want a day call part. That's exactly .

i'm i'm wondering if i'm onderstand the SONY. Her outsold the twelve and sixteen core parts so much that I want to even bother this time.

Also a the the thing the thing about the eight core part is that the eight core part is faster not to skip to the chase. The eight core seven eight hundred x 3d from the four generation is faster than pretty much anything intel launch this time and anything AMD launched to except for the new X V D part in games。 Ah so it's an incredible the v cash is incredible for games. Is is the kind of killed the r IT .

just designed. Just googling around IT, looks like there are benchMarks leaking for the sixteen core next morning part so that that probably is happening.

Um the the so the new design is that they put the v cash on the bottom of the chip so it's no longer insulating the eyes, which means you get a base clock that's five hundred megger heads faster than before and it's unlocked for over clock, which is new. Um so you is really, really, really good for games like it's the first of all of the cp s that have come out so far this year.

It's the only one that the that the enthusiasts gaming channels have been like. This is a real good CPU after you get this. And the overstocking .

results for IT are kind of crazy call cash the button a huge yeah I think they .

pitched as hey, this is the first time we've redesigned this since we started doing the b cash yeah the .

three D B cash stuff um work pointing out that not every game is going to see massive games from the extra cash. Like it's very dependent on the type of game and special game engines, even like like strategy games, anything that does a turn off, like math constantly, you Victoria, or like civil ation types similarly for x games and stuff like just like huge, huge, huge gains and a lot of a lot of three d action gains as well also get a huge boost. But there are some other games that see next to no improvement on the x pix games or even like I to think far, cry cry games, for example, like you be soft engine, doesn't see much of a lift like this is a highly game dependent. But IT is a lot of games to .

get a huge boost. So the till D R is, if you just want a kind of general all purpose CPU and you play games that x rd at four hundred and fifty five hundred dollars is a pretty good deal. They are sold out everywhere .

right now, anywhere.

because there. So three days? Okay, two days? Yeah, sense. yeah.

But I, I, I assume we'll be making a lot of these because they are going to sell a lot of these yeah.

this is this is what how they gain market share against intel enthusiastic aces, well, part of the market, but it's one that they seem to care about quite a bit. So that's what A M S got going. Um I like these chips are pretty good.

The power efficiency like I actually really like that they ship them at sixty five watts per two, one hundred and five. Um because default, this is a situation where defaults matter. And the vast majority people who buy a ninety six hundred and ninety seven hundred. Don't care that it's only the same speed as last generation part a and like it'll be nice that their rooms are less warm and their tastings, much electricity doing .

the same exact work they not rising in the world are yeah both .

companies .

have had examples in the last two years of C, P, U. Committing suicide because of the amount you, there ild different scales. IT happened, I know, yes, but they shouldn't happening at all, anywhere, ever. That's true. So for them to take steps to elevate the problem is, is extremely welcome.

Yeah ah okay. Next up is intel's the arrow lake A K the core ultra s series two. They skipped the one hundred series for reasons that I don't understand.

Wonder is that because media lake, he was technically the first generation where they did dissection gated, but that was only in mobile.

And that top it's possible this is this is technically like lure lake is the mobile part that came out earlier this year. Um this is actually A I can follow on to the media lake more than the lunar lake. There is a really different thing.

Arrow lake is depended. IT comes from me. The chip lets are built by T S M C and a variety processing .

four and n hts meters. R yes, which I think I I think one, the just terminology. L til calls their little chips tiles. They go on their created packages. AMD calls them chip lets or release chip let is they're a clock y alterna. Um but I think some of others like an I O tile and a memory controller and more, I think I think I think it's like between three and six hundred .

meters yeah that's probably right. And tsc makes all of those, which is the first time that ever happened .

for an intelligence top part. Yes, except I get up the best tile, which is basically the connect. All the actual tile sit on made by the and I realized literally the entire rest of the ship is made by tmc. Like that's that's like big for my historic perspective and a business perspective because like intel has always made all the own stuff and their fabrication traditionally has been extremely capable and has high capacity, but they have fAllen so far behind on that front.

Like doctor and cultures over tech, tech potato has a really good video about why this is a big deal because they they until doesn't just make their own stuff, they also have a completely different process for like how the designs are transferred to the to the machines and the problems and all stuff IT. As it's over, it's impossible to over emphasize what a big deal IT is.

Like like it's of like they are of eating cw, right? Like it's sort of like this sort of an admission of maybe failures is too strong a word, but like they have really fAllen behind the market to have to go to t sm c to make this thing.

But they were able to be iconic class for a long time. And finally, as all often happens with people, with companies that are iconic class in industry at the scale that they caught up with them, yeah, there's a whole story .

there about know who was running until for the last decade and how complacent may have gotten. But but also like they .

had they had a fab process that this was supposed to roll out on at least part of IT that flopped and they had to cancel and then they're moving future stuff to a different, different process. They is it's it's difficult like I don't think we need to get into why intel l having problems right now, but intel is having problems right now that are ready .

to stuff of the last decay. Think it's not they are having problems. But like they got this new process spinning out that supposed to be pretty competitive with T S M C. They're also moving, not moving, but they're building a ton of fabrication in the united states.

which is taking longer than expected, which is used them a bunch of money, which is anyway.

they got they got big on the chips act, which got knows the fetal survive the incoming administration or not. But I think I think you tell us, us us be getting like three billion from the government for to spend up all this new fabrication in the united states. Yeah um the point is like they have their own process that is was supposedly going to be competitive.

But IT may not beyond line for another year or two. It's taking time. And my understanding is the little bit i've read like the thing you're talking about with designs or transfer to kind of industry standard inner workings of the fab process like they are, they are getting more standard is .

part of thing that's changing. Well, they want to be to get other people's business as as typically, intel built intel chips on intel fabs and nothing else was built on intel fabs because they are cooling wasn't set up to accept files from other and it's like they were using word star still that everybody else has moved .

on to google documents like all to my mind, that's actually pretty important. Get into the national security conversation about having fab in different countries. But even just like the the thing you always here is that apple and in video or or just completely dominating T, S M C output. No, because the phones, like apple mobile staff in the video, A I stuff is so in demand that everybody else just has to quite fight over the tablets, raps of whatever capacity .

is left well. And you're in a situation when you're buying capacity years and years in advance, right? Like this stuff is scheduled for that, like when you by the time the fb opens its doors, it's it's first five, five plus years of time are probably booked.

right? And like i'm not going to place the bet that and tells a new stuff is going to be and like directly competitive on day one. But like more more next generation fab capacity in more places just seems good for a demand standard point of nothing .

else from me like a natural disaster. And right this like you said, there's taiwan is chAllenging for lot reasons. Um okay. So this is the first whole new design crowd up soup to nuts redesign from entel on desktop since uh like the halm which .

was twenty the early twenty .

ten was two CPU on the same chip that talked through memory or north bridges. I can get specifics.

but basically from a basic level, this is the biggest ant change design change until has made they went to multi core use .

since they went to real multi core yeah um there's new p uh so this uses like I said, it's just aggregated, just like the reasons have been for ages. There's a feros is that there are kind of underlying three d packaging tech that theyve been talking about for years. And this is the first thing ships on IT for just up. Um there's the tiles connect to the povero s layer um or the inter poser layer based tile. They got four different things in different places.

I ve I ve the term interpos'd. Yes it's hard to go um .

but there is a system on a chip tile which has like the basic housekeeping staff, there's a compute tile. There's an I O tile which has the memory controller in that stuff, I believe PC express lanes of that um and a GPU tile is optional. And then sometimes they also put a filter tile and that makes the whole thing nice and square. I don't understand why that's necessary, but I guess that is .

um so this is the thing I mentioned earlier where they differ significantly for M D at least for now all their C P U dies, so other C P U chores are still on one .

die single day. It's the three nine meter until uh three M C, right? So could I believe so?

So I did for now, at least for as long as going to do that. I guess that of the problem. And d has of different dies with CPU title or god. Sorry, a lot of terminology, different CPU cores on different dies, trying to talk each other of ray slow or interconnect, like for the moment at least, does not have that problem because every CPU core is still in one ship.

Well, because the thing to remember always is that as you get further, like once you when you're inside the same piece of silicon, you can talk an incredibly high speed across across CPU course right? So for example on the p cores that are in this in this um in this new package, you have uh thirty three magazines l two cash each right? And if one of those p cores wanted to talk to another p core, IT goes out to the l three cash, which is shared with all of the two cases in the system.

Thirty six magas of that with the equals, the equals all share four megabits of our two cash in each of these four CPU core clusters. So that means those four equals can talk each other very, very quickly across this l two connection, which is a lot faster than the l three connection that the p cores are using. Talk ture, this all. Here's a thing. The upshot is its way faster to talk through the l three cash between these cores no matter which cause you're talking to then IT is to go out to that overs later and talk to another chip on the eye on the same CPU.

on the same CPU, on the side of, on the infinity fabric, me into between ships.

But an instance fabric is slower than talking to cross .

l two cache inside the just anyone.

So these are sky mod e cores, which are supposed to be significantly faster than previous genie chores. And the thing about the equals in the p res, the equals are smaller. They're less capable in that they sometimes don't have the same instruction set accessible to them as the p cores do.

They have different cash configurations. You're mostly there to to run on benchMarks where you benefit from having lots of CPU course on on test workloads that you benefit from nothing, lots of CPU cds. So your video and codes, you're rendering stuff like that.

Every everything i've seen is only for whatever other problems until is having like there there the group that doing the equals is just going ham.

I think I like the equals.

The sky on equals are like not very good um and that is um i'm trying to think I was keep in track at this a walk for a while because like I got i've got this older lake for example, as of older lake, I can't remember which mounted was the Grace mand.

Anyway, the equals in the older lakes, which are twelve gen, were said to be about equivalent to a sky lake top part which was was sixty seven k yeah so basically the course even even a generation or two ago, the equals were roughly equivalent to, and I am desktop part of what? Eight years ago, yeah, the fifteen, nine, nine years ago. Anyway, the point is, the point is the equals are getting like very capable themselves and want to play games on them, to be clear. but.

I think you probably be fine depending like what kind of resolutions you're playing on. So like that, the point is the equals this year, suppose ly thirty percent faster than previous years.

which is because again, they were getting pretty good. And part the reason parking on this is because there is think people probably are getting familiar with this. There's this this whole scene of like lower and famous x ty six boxes yeah coming out of like all express and stuff, like things that people are building routers out of and like appliance type stuff, literally like a low power enough and not even need fans that are pass of cool. I am extremely eager for for those guy on course to start migrating to those things.

Yeah, that's the kind of thing that come over time like like intel loges six four five CPU six CPU and then theyll roll out a hundred different variants in all to fill every available market dish um okay so big deal this year. No more hyper threats.

Yeah I I wish I wish I understood CPU design Better enough to understand why this is apparently a good thing and not a bad thing.

Well, IT used to be that the p cores had hythloday. The e cores didn't a hyper threading that you basically, you segment off different parts of the CPU and do desperate tasks on the same CPU as if they are there. Two virtual CPU basically is the lay the the easy analogy um the upshot is the overhead of high per threading makes the CPU a lot bigger and they were able to pack more. And if there is my understanding of the .

thing OK .

and also with the whole shetler of equals, so just to be clear, the high end part here as A P cores and six teen equals the over. My guess is that the overhead of having hyper threats was less efficient than having just more equals jammed into the place.

And that's twenty four reads like that's not really anything to score that yeah .

well because the thing is if you're using the same if you're doing the same kind of math on the on with high per threatning on IT can't use different parts of like hyperthermia works. If you're using different tasks, different kinds of math on the same parts of of the CPU, if you're doing the same math, IT doesn't scale and a lot of the heavily multiple ratted workloads are going to do the same math, just lots of them. The A I stuff in the and things like that, uh, integrated graphics are the X E L pg architecture, which is like a mini architect from their best top parts.

Bear from the cards, the G P U, stronger, stronger push and graphic cards last couple years, which like sounds like that's kind of paying off, like they're still not super competitive with video and A M D.

but they are getting Better. I mean, I think the baseline for integrated graphics this way up from what I was four years ago, it's not the battle mage stuff that in winter lake though, of mentioning I I I .

want to say I thought us all people are talking like these other graphics are getting into like and ten theories and video performance levels.

Now I had ten sixty years something .

which like eight year old video card, seems about right.

yeah. I mean, we can play fortnight and switch if you want. It's a, it's a two G P, U course. It's I would describe IT as fairly underwhelming.

I mean, you know, obviously anybody who's serious about games still want to graphic s card, but great for embedded situations or use cases where you don't want a portola phh c card in.

And the real reason you buy a chip with A G P U. On IT is because you want .

quickly support, which has one, you know, how do one code, their video code on here, which is going to be huge for streaming and bleak and all kinds .

of stuff yeah if you also this is worth mentioning, dmd parts don't include N P use this time really uh this does include uh the arrow lake n pu which isn't past enough to get the copy plus features turned out. No kidding, it's only thirteen tops, but IT is in there. So congratulations. I think probably they they specifically said if you care about that kind of performance you program put a discreetly graphics garden um and IT wasn't worth taking up dice space on NPU core stuff .

to get in to the forty forty tops that compile plus requires can you run all the copilot the graphic know why .

that's the uh to give you a sense of scale here, this is a thirteen top M P U tops N P U forty sixty is two hundred and forty eh tops of perf. So that's the lowest end in video of four series card that's available today.

Do you have any idea why you can use a graphic card for the features on the saying I want .

to use this soft, just arbitrary string.

i'm going to probable OK hauld again doesn't seem to make sense why you would lock that off.

It's if yep, I don't I don't understand them that's .

that we are choice by microsoft. I mean, you know, he goes back to like I I wondered about things like this because you remember the days of which directives do they lock up to as visitor was an aid? No eight ten, no eight dix. And anyway, the point is microsoft has a history of, I think IT was, I think that was the extent that you had, because I forget which halo IT was that .

you had to get halo three.

right. I forget there are some anyway, been a while. But microsoft has a history of locking arbitrarily locking features behind new windows versions, which just may be wonder if there's something yeah .

directly tent was was this to only because you needed W D D M graphic drivers, which is that kind of like that's a choice but whatever anyway um my guess is that this is more we just haven't built the apps for this to work with the gp s .

that that that also was believable.

And also the kind of things that are copy plus features or so would like low power require that it's massive over killed to put the forty, forty, ninety years on that anyway. Um the new chip set because it's intel. Yes, new chips at new cat intel move is always new chips at new new C, P, U. socket. They have done for the last releases of core I nine core eyharts yeah well twelve thirteen .

th and four eighth gentle work, the same motherboard which is extremely out of character for .

them that is literally the longest run i've ever had in tel board. Last.

assume they make a decent t of money on the chip sets.

I think that they just want to keep selling new motherboards every year when they sell. Yeah they're .

making money selling chips to mother board .

makers yeah so the new chips has the the eight ninety, the new sockets L G eighteen fifty one. I would assume that you're never going to get another generation of CPU ago es in the soccer or this chip set when you .

buy these boards. Probably notes.

which is fine. I mean that's how most people that that's how computers .

work for most people, right? Um the parts .

are the two eighty five k which is a five point seven gar hertz max at seven five hundred ninety box unavailable everywhere. It's ap cres six cores, four G P U cores .

and the NPU. Wait, you're saying it's sold out really weird because something we're getting into in a minute, but considering be driving this thing took in the press and making you can make a lot of them. I think they put them all to .

the o EMS k um the two sixty five k and K F parts K F parties or reminder are are unlocked, but with no GPU are five point five digger hurts max. Three hundred and ninety four for the k three hundred and seventy nine for the K F so you save like fifteen bucks, not getting an integrated GPU and probably a little bit thermal overhead. Ap cres sixteen cars and four GPU cres except the f one o and IT has the pores save cause that's interesting, locked down a little bit OK just to sch and then the two forty five k in kf, again, no graphics on the kf five point two giggling urts max and at six p cost six eight e cres four G P U cores and then P U and the the k is uh three twenty and three three hundred for the no graphics of .

party of A F all over call locked .

and overlook and again, this is a situation with the fill out more CPU that most of the parties good. O ms, but these are like the enthusiastic x parts.

yes. IT was the two forty five k what you would think of as an I seven? No, because that's six cords. Like typically the I seven is like the same number of performance cores and just somewhat smaller H E chores like they're still a room for a nine, seven equivalent in the lineup.

There are some weird holes here also. They used to do things with hyper threats and not hyper threading. And that would say like I trees to hyper threats or something like that, it's a brave new world. We're going to talk about what's weird here and where they are big perf holes. I think that they Prices these based on perf would be my guess and for the previous generation OK. Um so okay, they're pushing power consumption there like all their marketing for these was about power consumption because the power consumption bits real good and perf is kind of eh, especially if you compare IT to ninety nine thousand series rising, it's pretty mir if you compared which was already made and if you compared to uh rap last generation and rapper lake, it's it's real double man that's so damping .

like like not being great against your competitors. Contemporaneous s released is one thing ah. Not being great against your own previous generation is really hard as swell now.

okay. So there's there's A H risks here, right? It's real good and workloads that usual boat loads of threads, right? So like cna bench, hand break, video CPU transcoding, things like premiere does real well. It's not good in games because the clocks in the and the thermos are a little bit constrained and compared to the fourteenth gen intel um and get asked ked by the three D V cash M D party thing like that.

the rise on the new .

rising stopped to until the ninety was out.

Um yeah I I had thought about trading up to early until us all the game performance of and that's just a little hard to swell.

I mean, now the thing that's worth mentioning is when you're when you're getting fourteen, nineteen k performance out of the thing is drawing like thirty percent power in a lot of cases OK.

So like like you're basically getting this is basically the the intel l equivalent of what we saw with then four last around five in terms of power consumption, right? And when you compare the place is interesting when you're doing the power test, is the idle draw for intel is way, way, way lower than the idle draw for AMD? No, we're honestly we're working on ways to like show this in benchMarks because it's a little bit tRicky. Um but like if you're thinking about a Normal workload where you're using your computer, it's idle most of the .

time for most people, right? Eighty and ninety percent. The ship two yeah and like the .

idle draw on these intel chips was like ninety watts. This is one hundred and twenty eight hundred twenty five for the for the comparable reasons. So like it's it's a significant about the power over a long period time.

If you're talking about like not turning football all like you're running football all the time, they amb each chips have a little little advantage. If you're running Normal human computing workloads, recent in front computer and doing some zoos and doing some emails and then play a game, probably it's gonna be a wash at best and probably favor the intelligent. No, the gaming perf problems here probably do to memory lencs is what the consensus was amongst reviews.

The memory controller is on its own chip let. It's not integrated in the computer. Chip let, which was like the assumption. My assumption there is that that, uh, design choice to get more computer, more C, P, U, course on the same, sorry, more compute wafer chips out of each wafer, because those computer chips are the expensive wafers from T, S, M, C.

right right now. I have not lived in this world for long, like you've been in IT for thirty years or whatever. But like the hardware the iterated hardware view kind of but even I have seen enough like slight decks and then enough presentations to suspects, like maybe the next generation of this thing, like that's a prime candidate for like we have improved the memory controller. Suspects know twenty percent, but you know like that's one of those things it'll be like or more broadly, that's one of those optimization points where there's something they didn't do as well as I could have been done, but that they will you know improve an iterate on in future generations.

So the thing to remember with all of the CPU stuff and GPU stuff to is the pipelines for design and building. These things are years, years deep, five, seven years in some cases. So they aren't going to realize that this, this is a bottle neck, until they started seeing the first runs of chips off the actual flood because you can't actually sym these anymore because they too complicated like they get the first ones back.

And by the time they get those first chips back off the off of the line, it's too late to make any significant changes. So they're not going to realize that oh yeah okay a memory band with this going to be late is jacky. They also had some problems will talk about a minute te um that seem to be windows and or BIOS related issues.

Um so anyway you to your point, yes, this is I would assume, if not the next generation, the one after that, i'll see a significant change in the way the memory memory agency works and we may even see some substantial improvements in the in the interm. Don't buy stuff based on future promises. This is always the role.

The other things that worth mentioning here is that all of the chip lets on the articles on their disagreed ted design, almost all of them are overlooker independently, including the base layer. You you can over clock the I O tile. You can over clock the, the, the poos connection layer like go ham, they put all the dials into this thing is crazy yeah so have to write board yeah yeah um so we have to test that any of that stuff yet before closing .

the base layer, which IT does nothing. Fact, communications between tiles is hilarious. Well.

but but if you think about IT, if it's memory lencs the memory the problem, then overtalk that actually might have a significant totally, don't over clocked like this is let other people figure this out up with your cp.

S I never ever in a million years do something like this but it's pretty funny that is .

not possible yeah um and and we talk about this before but like this is the most complicated, biggest change in the way these cp user built for a long time so I would have been shocked if this thing came out as this is the fastest thing we've ever seen. When risen came out, the thing that they had going from this, they had a blower to CPU CPU cars and a whole bunch of weird issues. When you were doing things that that needed like communication between dies and stuff like that expected.

it's actually and I feel I feel like i'm not saying you should go and buy one of these because it's like at the end of the day, performance is what IT is. But IT seems somewhat impressive that IT actually did come in at some like reasonable level of performance and like so far, not a tone of real deal breaking issues.

I mean, and that's the thing is like if you're in a Normal user and just do an computing stuff and encoding some video on stuff, this you're not going to notice this. But like this is going to feel fat if you're up creating from a from eleven gene in tel to this is going to feel fast as hell, right? It's only when you're comparing about, again, everything else is on the market that IT pull low. If you're spending money, you care that you should buy the thing. It's fastest.

I do yeah really right. Again, if you if you want to build a gaming machine, get an idea once you can get them. That seems right. But again, you are just just conceptually or international laying foundation for the future, like the fact that this thing works generally is a good sign.

It's very exciting, especially given the problems ahead with rap lake right? Um so we are testing a PC world. We ve found some issues with power settings, other people according orated this so you know that setting the you haven windows, you can say most efficient, baLanced or most performance are a little little filter.

If you're not on most performance, weird shit happens. So for example, of the baLance settings, we saw like a significant double digit performance drop across prety, much every CPU benchmark when you're on the most efficient settings, IT just ran everything at idle for single thread benchMarks. So IT never clock up the the boost clocks on the CPU core. This is stuff t that he'll fix over time for sure.

That seems like a pretty agreeable overside.

I mean, this this is again, this is the kind of thing that happens. Part part of this is that windows twenty four, each windows 7, twenty four eight two, which is what I really tested on, came in a little hot and rework a lot of how this stuff works. Um but yeah, if I was shocked that we had these problems, I wouldn't expected this at all and like I wouldn't buy these until they resolve that um the edge case oh this is this is your question.

This is so I guess you just talk about this now let me um what i'm curious about and again thought about trading up to one of these from a 4402 k until I saw some of the the benchMarks。 My question is I am limiting the power on my 4 hundred k i'm not running in for bar because i'm cooler。

Okay, so i've like, for example, boost up to two hundred and fifty three watts, but I ve ve got to kept up one monday just because basically you need a liquid cooling to actually handle two, 3, three o so i'm wondering how my power limited fourteen hundred or k would compare to to eighty five k stock since IT runs Better at lower, lower voltage or higher efficiently. Know you would mean like you're getting more performance poor wab on the new report. So i'm like, obviously, pounds for pound IT doesn't measure up very well, but if i'm running the two ty five k at its full capacity versus a fourteen k that is limited by twenty percent or more, I wonder if I would actually be coming out ahead.

I will show the max. okay. So first off, I don't think I can answer this without actually doing test this.

Finally, not A I do. This is more just a general fought experiment, conversation .

starter, the thought experiment.

Is that the reason you doing this because you're on an air school? Yeah was a one was in constantly in games. Max paradol.

yes. So you're not going to you're going to have the same thermal limit and the two to five k does go up to twitter. Fifty watts, I think is the max T P foot for boost k.

yes. So you're going to hit a limit somewhere in there. okay. Um IT doesn't hit boost a whole lot like that. We resting IT IT was I don't know if we put in on on really answered your questions that you need a big a big water.

I O yeah I know I know yeah .

like a cooling is not the right thing .

for top line CPU these days.

Sorry though in some ways in other way, yeah, i've never look at my peel one, peel two, five. But my guess is you can end up you'll do more at that, at that turmoil, at the same limit then you would on the, on the new chips mean the new chips will you will get more work out of the same amount of whats right on the new chips.

Is IT enough more to warn the up probable? No.

absolutely probable. Here, the real till the art of all this is that if you're upgrading your CPU every generation, you're just kind of wasting a lot of money is like the seven hundred x three can end up being the same level of performance as the as the all of these chips, except for the nineteen eight hundred x 3。 I think I think .

even isn't the fifty one hundred x thirty. So like pretty competitive.

the fifty eight hundred if you have fifty eight, eight hundred, fifty you doing great. Yeah yeah. So okay. So the other thing that comes with this new chip set in this new CPU is is C U dim support. Um this is a new memory standard.

IT basically adds this clock driver chip to the dims that generates the timing clocks for the RAM on the rams stick itself instead of taking IT from the rest of the CPU. Uh the the practical upshot is must be to run much, much higher clock speed on the on the RAM. And I like the detection limit is like sixty four hundred mega transfers per second.

I think the hey, if you want to do some crazy shit, you can get all the way up to eighty four hundred. There are some big heavy y ots. And that you have to do.

You have to adjust some ratios. The number transfers cycle, go down and stop like that. You might have to run them in single channel mode, which means your effective your clock speed will go, but your effective band with will go down. Um so it's early. This is what I want to say because i'd like this sounds .

like a good improvement in the long term, like obviously new stuff like this never quite where you wanted to be right out of the gates. The problems we've been having with D R five stability generally since D R five came out by, clearly this was necessary if they take the step. So this this seems like a good thing for the future.

to be clear. Ddr five stability, eject speed, no, 是 G D D, D R five stability at seventy two hundred mea transfers per second. Not great. yes.

You know, if you go back and look at the D D 4, like what was the jac was like, twenty one, thirty three may be twenty four hundred, four hundred or four out of the gate, but people were routinely getting up thirty six hundred and pushing four thousand without issue. So like the point is there is a lot less had room on d or five until they took .

a step like this IT seems and and this is the path, right? Like I was working ninety hundred k machine in the other day and i'd looked was in the bio sounds like, oh, to put the X M P on for this and see what thirty six hundred mega transfers per second it's not it's entirely respectable machine still but yeah we ve this is the path to hire memory close speed. And my assumption is will see Better support and performance in the future.

Now if it's in the existing slot, if you plug in one of these chips into a BIOS that understands that these chips can be on the dims, it'll say, hey, we're going to turn off the C, U, dim chip, the clock driver chip. So is just Normal. D, D, R, five little run at jex beds.

So it's kind of the BIOS has to understand IT rise right now. You're knock like there's only a handful BIOS. Is that understand? We plugged into a bunch of machines that were in the lab and IT was weird every time. interesting.

okay. That's yes.

So um you but but like you don't have to make this choice and do IT now when the kings get worked out, you can do IT later if you have if you're up getting memory at some point in the future or something.

Are these widely available?

No, there's like one kid in the world.

okay. So is that kind of I don't .

even know they're selling IT. They might have .

just sent us kids. Interesting because I don't know with pricing, I want to how much might be with an extra chip on there.

But I guess I say, I do you right now and see by C. U. Dims, I think when I searched this another day on the full nerd, IT looked out, IT came up with.

so dims, because I thought I was typing something wrong. A crucial.

crucial OK. Three, two.

Good by in sixteen, by two hundred or five, sixty four hundred to you, ten hundred and seventy dollars.

Okay, here's a forty eight gig by kick from kit from manta v color manta expand ity forty eight gigabits C U dims eight eight hundred mega transfer for four hundred box.

Well, well, but that's eighty four hundred though that is that that is extreme bleeding edge grand speed.

the eighty two hundred or two sixty nine. I wouldn't. Here's a thing. This is going to be jane for a while, right? Like this is, this is a let, let everybody else figure this out situation, not a you should do this today, such way.

So, okay, this is what I was trying to find, an apple's, the apples comparison. And I have found IT. Okay, here. Here is a similar, crucial ddr five, sixty four hundred. Kid, thirty two bites for one hundred and fifty dollars, okay, on their side.

So one hundred nineteen for the regular dim or U D M C U what size uh for is is sixty by two? okay? Uh, the exact name sixty by two sixty four hundred and A C U down is one hundred and seventy right now. So there is actually .

A T pretty hefty Price bump .

yeah big premium at the moment .

yeah so um anyway you d like let them figure that out. Don't work right of the six hundred. And um so intel has said this isn't the performance we want out of this chip er we can expect out of this chip. We think that there is something wrong with BIOS and that were working with vendors .

to solve I just found this this morning thing is so this is reported by video cards dot com with the e yeah um they linked to this is interview that supposedly robot holic until vice president did with hot hardware okay, the odd thing is, is a youtube interview and when I click through the video has been pull weird. But I do have the quote they pulled from the video here. If you need to read IT get part even though the video is longer up.

And i'm not sure what's up of that. But here's the quote they pull out of the video. I think people, I think what people have been interested here is what happened.

I can go into all the details yet, but we identify a series of multi factor issues at the O S level, at the BIOS level. And I will say that the performance we saw in reviews through no falter reviewers is not what we expected and not what we intended. IT launch just didn't as planned. That has been a humbling lesson for all of us, inspiring a fairly large response internally to get to the bottom of what happened and to fix IT.

So look, what they should do is they should set the micro code to keep increasing the voltage of every time is more demand until IT is just infinity.

It's just crazy enough to work. Yeah so I don't know I I don't know if who who or what caused that video to go down, but I assume we can still take that quoted face value. We heard that he got hold.

No, they upshot is A M, D had the same thing where they had to put out the windows of right to improve performance. And now IT looks like until might be doing both U. S.

And bio updates to try to help things here. I just I just the world we live in now like is IT just like on one level, IT seems insane to me. The companies as bigger until and microsoft and AMD would not all be collaborating enough hind scenes to get things working properly before they come out. But is like if we just reached a level of complexity where that's not realistic anymore.

I think that's the problem. I so before it's worth, there's another video that's a live stream that has Robert lo talking to hot everywhere.

okay? Maybe they took down my life stream archive .

and put up version up. okay. Anyway, I think I think to your point the harbor is really complicated like the these disease gated designs are complicated in a way that is that windows and and software aren't necessarily equip to deal with right now um both because of the differences in the P N E chores on intel and because of the like the communication across eyes on the AMD side like these are all difficult things and a lot of the underpinning of windows are based on, hey, all CPU cores are the same and that's .

no longer the case. CPU got weird. CPU are weird. CPU is got weird. And it's all it's a reaction to the units of silicon rain and how much you can just increase close speed like this is these are soldiers. Further effect of we have to find other ways to make CPU faster.

Well, I mean, you say that, but also the process text to keeps still getting smaller even though we thought we walls years ago.

But but but it's pretty minor. We're slowly egg up past five aged reliably.

yes.

I mean, I think I think the the thing i'm curious about these days is how much performance can we get out of like how much work are you getting for each? What that goes into the CPU now, right? And and folks like Steve over gas nexis and other people have been doing um some really interesting tests by isolated the rails, like the power going into the rails of the CPU uses to actually see what how much what is each each benchmark run costs. Um and I think I think for me the fact that our solution isn't just well, you can get sixteen or eight one hundred watts out of one hundred and twenty four fifteen nine circuit and we can just keep dumping that into the CPU and y'll keep getting faster isn't a sustainable solution and it's good that everybody is moving in that direction extremely.

These CPU even even whatever issues they have individually, just generally IT looks like the race to the top on access footage and trying to come out point one percent ahead on dench Marks is finally ending.

Now the bad news is that the G, P, U, side hasn't reached this conclusion yet. So whatever savings we get out of C P U is just onna get absorbed by the G P S. Until I think one day, you know, the gensler vision, vision of the computer is just a giant in video GPU that you drop A C P U in the rain into a anyway rough launches.

They are rough because of their complicated products. I think not because like there's no look, the internet let's see gordon over PC world likes to pull out the jump to conclusions, matt, and get everybody on their outrage ponies and get really mad about stuff. Um I think that this is just a testing. This stuff is increasingly difficult like I did, but I spent a lot of time winning benchMarks of the last couple months.

And it's not you know the old days you you could run a bench market and you could change the front side bus multiple er or the the the front side bus speed or the multiplier and kind of test how things were running to test your your assumptions about about what was going on inside these chips. And there are so many things going on now that that's not as much of the case anymore. You just really can do IT. Yes, there are way more knobs, more knobs um to the point that I think given sometimes internally like between the OEM, the om making, the motherboards and the CPU manufacturer, these even missed communications about how stuff should be set up as we saw last generation on both sides.

I think I think that maybe too amends the thing I just said about the fact that these jane companies aren't getting products out working as well as they could be. I think that the more surprising thing to me, not that they not shipping them at the an optimal state, they don't even seem to know that they're not an optimal state until the reviews had and they like, oh, we don't even quite know what we shift all the is an out of but I guess. That's just the nature of both the complexity in all of the time lines of the launches are working on, like who knows how much internal time they even have to fully test them and get familiar with everything .

before they hit the market well and and the timelines for the reviews are, I don't want to say, compressed, but it's like if you have five days with upset at two or three CPU, you're really lucky. And typically, you get one mother board. So if you do as you test sweet, that takes twelve hours to run like it's chAllenging yeah mean.

this just strengthens my growing feeling that maybe wait like six months after launch before buying a new CPU.

These I look I think that's save advice for literally everything unless you really like in video of founders edition cards, which sometimes are no longer available after six months. There there yeah I think I think that .

good of places I need to rap IT up, right? I've been pretty bumped out at the state of CPU for the last few years. Like part of IT was part of this.

Like I said, everybody, he's just burning the things out, trying to trying to make them look Better on paper. And also like the covered disruptions, seem like they cause issues with design and testing. And Bobby aba, like IT, feels like feels like all of that stuff is finally settling down. And there, both companies are laying good groundwork for the future, for things to get Better.

I like I yeah, I agreed. I think the the stuff that's going on with the this rise inside is really, really exciting. I I am super curious about the x 3d versions of the dull C C D chips now, because the delay and clock speed on those dull C C D ones between the v cash one and the not v cash one should be a lot lower, which I think solve a lot of the threat, parking and weird problems that we had with the last generation.

Um but and and also that ninety of other actually is just a dope. Part of people say games is so frequent fast yeah um the the the intel l stuff whether they work out the kinks this generation or next like I mean, I know they're going to continue to work on this generation until he gets to the point the point where they're happy with that, right? Don't buy IT based on that. But but like they spend a lot of money building these chips and .

they need them to move. And you know with those comments about BIOS dates and windows updates like maybe and six months like the area equals more.

Well, the says the other thing is I think that the the basic premise of, hey, we're going to do these low power chips and we're going to get good performance out of them is painting up like the sea.

The the server side of this reviewed much Better than the best top is right? And the lunar lake is is killer like on the mobile side, the T, L, D, R is that unless you're running corporate stuff and and really don't want third party of here to be compatible, don't buy qualm, right? Basically three chips sound the mobile right now.

The quarter based uh elude x luter lake for antil and the rising A I for for AMD. The reasons are a little bit more performance but use a lot more power, not a lot more power, but significantly more power. The luter lakes are basically as as power efficient as the armed processors run all yours software on our hell of us.

Ah that's trade like literally sounds killer if you have the .

next eight six laptop yeah so like IT, it's where a good place. It's black Green teeing Peter territory here. Yes, yes.

I think this condition, I thought I had seen this rumor and sure if I have pull up, I mean, it's still a rumor, but until least being reported by reputable sites, I love a good rumor on the the to the two C, C, D extra departs. So the hosting and twelve core, the rumor is both of those C, C, D are going to have V, H.

That I would be shocked if that is the case.

but that's what's going around.

I would be shocked if that is the case. Well.

but at least the collect speed disparity .

will no longer be a problem.

I would be shocked if that is the case. I fully and I would be more expensive.

I would be shocked.

Leave there. Leave at there. yeah.

On and and speaking of living things there, I guess this is the time. And we thank everybody for listening to the show and supporting the show bread. We will made a technology, a listener supported show.

We wouldn't here be here without you all. I know the listeners so much. If you want to find out how you can support the show, you can go to patron 点 com sih teck pot again。 It's patron dot com sash tech board. And i'm so to say three times is what paton tells us. They say, say they called action three times of the people, make sure they get that you are all so when they are in the car, when they in front computer, that next time they think, oh, I should go to patron, compile checks pod and give bread wheel five box so that they keep making the show because we don't run ads here, this all listeners supported. We only we do this because of you, the listeners who who we appreciate for for taking good care of us.

And and you get a whole extra podcast a month.

You get one extra podcast a month. You get access to the discord here. You can like, look with, maybe we need more community than ever now it's a good place to get into the good community of good people, yes.

people. And you need distraction, you know? yes. Also, if you want a place wear, for example, if you're thinking about getting a virtual private server and running a proxy to your health to run quake servers and relaxed after through, yeah, i'm just i'm throwing that out.

There is an example. The point is if you have a thought that you're like, I wonder if I could do X, Y and z project. I wonder if this would work. Almost guarantee you that no matter the domain of the project you're considering, there are people in there who can like answer to the question .

or have advice or have been in had some more thoughts can talk through with. So you got a patron dog size checked. You give us five bux a month to get access to the is going.

You get the patron exclude episode every month. And if you're an executive producer, you get thinked every single episode, just like Andrew s. Lucy co.

Jordan lipid, bunny fined, twinkle, twinkle vid, Allan James chemic and petite makers of the H. S. Three high speed 3 winter。 Thank you all. So, so much.

No.

thank you. And we will back next week with another addition of the tech pod. Thanks for listening, really. We will see you all then.