Hi, this is Wayne again with a topic “The Return of HEDT? AMD’s NEW Threadripper CPUs Revealed!”.
Uh, this is a priority transmission uh, two back, not just Pro and there’s a lot. We got to talk about. Yeah AMD has broken cover on thread Ripper Pro 7000 WX series. I’Ve got the inside scoop, but today is really just about the Press day and usually I don’t do videos for just just like the announcement, because you can read the announcement anywhere and and it’s fine, but I’ve seen these firsthand there’s not a lot of details that I can share yet, but I’ve seen these firsthand. What it looks like AMD is putting together with the launch of the 7,000 series. Zen 4 based thread. Ripper is something that will uh submit thread Ripper as a product for the ages, a workstation product for the ages, but they’re also going back to their high-end desktop Roots.
Why well, maybe a little bit because Intel did that with sapphire Rapids. The market demands it. Maybe because the supply is a little bit better than with the 5000 series thread, Ripper parts I mean there was only thread Ripper Pro 5000 series. There wasn’t thread Ripper 5000, which meant that you had to have a an expensive motherboard and the uh.
The pricing sure got a lot more expensive if you’re doing software and Sciences, media and entertainment, design and Manu facturing architecture, engineering and construction. You probably are aware of thread Ripper and now with AI when we’re talking about inferencing 40 %, 50 % of inferencing tasks in AI still done on the CPU, not the GPU and the CPU can make a lot of sense for other AI tasks. Amd’S got some pretty good slides where they’re, comparing their 7995 WX with 96 scores to the W9 34 95x with 56 scores. I think the better comparison is going to be 64 cores versus 64 cores and then know that you can go actually beyond that things like the the uh chromium compile and the after effects. It’S like. Ah, it’s going to be a little bit faster for compositing. It’S going to be a little bit faster for whatever listen with after effects. If you run multiple jobs in parallel on After Effects, because you can, I guarantee you, the 96 core is going to be way faster, same deal with chromium. It’S like! Oh, you know, chromium is 28 % faster. Listen, your bottleneck is not going to be your CPUs. In that case, your bottleneck is probably going to be your storage. Are you? Are you doing the comp from a ram dis? You need to be doing the compile from a ram dis. It is 50 % faster and Unreal Engine, which you know larger projects. They do parallelize a little bit better, at least according to AMD slides.
So this time around there’s the pro platform and there’s a high-end desktop platform. The pro platform is remote manageability, that’s the out of band management that we saw on workstation boards and all thread rep Pro motherboards, pretty much there’s also some other Pro manageability features. Eight channels of memory, it’s 128 pcie Lanes the extra PCI Lanes Beyond 128 – that you see report orted in some places has to do with chipset IO, but there is actually eight extra lanes that are just quality of life improvements uh, so you know it’s 128 Lanes Is not technically accurate because there are a couple of Lane. Well, it’s like eight lanes that you have extra for doing other types of interfaces stuff, so their thing says 148 to 144, with up to 128 Gen 5 Lanes. So that’s! What’S going on there! It’S like! Oh, these are Gen 5 Lanes, both the pro and the HDT platform support overclocking, I witnessed overclocking, that’s all I can say the other differentiating thing is. You can use the pro processors in the HDT platform if you’re in a weird situation where you need 96 cores, but you don’t need a lot of memory.
Bandwidth and four memory channels will do it for you. You can run 96 cores in the HDT platform with a pro processor. Now there are going to be three processors in the non-pro lineup, the 7980x, with 64 cores, the 7970 X x with 32 cores and a 7960x 24 cores.
Actually oems are probably also going to have access to a 12 core model. That’S that’s possibly Pro only we don’t know, don’t know how that’s going to shake out, but this is what’s going to be in the retail Channel, meaning you can go to Micro Center. Some new egg somewhere like that and buy these.
The 79 7x is what I’ve got my eyes peeled for it’s going to be a good balance of workstation platform gaming platform, turbo up to 5.3 gahz 350 W TDP at the floor. You can have significantly increased power consumption with the overclock. I imagine probably just saying AMD provides a graph of performance of the 7980x with the w34 95x. It’S 56 core configuration versus the 64 core, and you see some pretty significant uplift from software Sciences, media and entertainment, design and Manufacturing architecture, etc, etc.
Can’T wait to do my own tests. This is not me doing my own tests or anything like that, but be aware this is coming high-end desktop high-end desktop is, is maybe coming back. Maybe something interesting. Both the HDT and the Pro system are going to registered error.
Correcting memory only R dims ddr5, the reason for that is twofold one. The ddr5 standards now are not physically compatible in the ddr4 generation. With thread rer Pro, you could use error, correcting memory, you could use um, Regular, desktop memory or you could use server memory registered error, correcting memory, that’s because they were all physically compatible in terms of reporting, ddr4 and all the Myriad different configurations uh. That was a bit of a headache and an engineering challenge because of the engineering challenges of supporting the insane.
Ddr5 memory speeds we’re in a world where, unless you’ve only got two dims or two memory channels – and you really really want to cut the costs of the platform – there’s registered airor correcting ddr5 the memory. The officially supported memory. Speed is ddr5 5200 officially, but you’re going to be able to overclock and really push that to the moon on the pro platform, as well as the high-end desktop platform.
Now, if you’re thinking well, this isn’t you know the same as epic, Genoa, it’s not 12 memory channels. What gives I want my four memory channels to you. I would say well keep in mind this isn’t 4800, it’s 5200 and Beyond. One imagines that 6,000 might be possible or doable and when you’re running ddr5 memory that much faster, do you really need 12 channels? I mean eight channels is pretty awesome. The other thing here is one dim per channel, so that takes out a lot of the engineering Kur fuffle.
Now, if you’re thinking, I would much rather have uh desktop memory on the highend desktop platform. Maybe we have registered air cracking memory only on the pro platform. No, I agree with amd’s decision here, because, if you’re buying something like this, you really should have airror correcting memory across the board and like a ddr5, ECC, udm type of thing doesn’t really make sense from an engineering perspective. Also from an engineering perspective, it doesn’t really make sense to try to bring up synchronizing, four or eight memory channels for uh.
Non-Registered error correcting like the memory companies are doing the work for you to do the engineering to bring up ddr5. They always bring up ddr5. First, in a server context, not a desktop context, the desktop context is, can you take what we this awesome thing that we did for server and just make it work somehow in desktop and as anybody who has had any ddr5 platform from any company, be it’s team Red or team blue, I wasn’t a lot of fun in the beginning, whereas server ddr5 and registered air cracking ddr5 has been perfectly reasonable, like actually a good experience to deploy that with six or eight or 12 memory channels. To give you an idea of how they split up the the lanes and everything else, take a look at the platform view slide. They provided. I always appreciate this level of documentation because it’s like, oh, I can see how my Lanes work now, if only I could get a diagram of how the iommu groups are going to be. Is this going to be the ideal platform for passing through a GPU to a virtual machine? I sure hope, so they also give us a platform view slide, which gives us a breakdown of how they’re going to split it up between workstation and HT, eight ccds and four channels of ddr5 and 48 Lanes of pci5 for HDT, which doesn’t seem unreasonable, especially when We’Re talking about 24 and 32 core configurations AMD also threw in a slide for power management. We see you, we know this situation, we know the score on power management. You don’t have to remind us good job 73 % lower power per core.
Oh, they went there. Both the HDT and pro processors are going to be available on shelf November 21st. I asked explicitly and said: hey.
Is it going to be a weird situation where the pro processors were only available at the big three oems, and the answer was no now I happen to be in the neighborhood for other reasons, and I got to take a close look at systems from Dell HP And Lenovo Dell and HP are allowing us to show their systems now no performance, of course, but the Dell system was shockingly quiet and it’s obvious. The Dell’s put a lot of engineering into this workstation class platform. They opted for a hardware.
Raid controller looks like, and we get a pretty smart internal case layout, also notice the arrangement of the pcie slids pretty interesting stuff. Similarly, the HP workstation looks to be a pretty thoughtful configuration and again running configuration doing stuff stuff sites to behold. Again, it’s sort of a split slot design, eight memory channels, obviously a pro platform there’s a lot you could discern from looking at the pictures here.
I don’t have any idea about board pricing board complexity, everything that goes into that. Will we see standard ATX size, HT boards that are at a reasonable price point? I think it’s a different world in 2023 than it was when we saw the launch of the original thread Ripper. I really do economically everything, and I think that that is, you know, sort of the unfortunate world that we live in.
But I am very excited to see The Return of HDT and I hope that macroeconomic conditions, notwithstanding that people are able to make that Mak sense. For them, the pro processors in the creative industry, you know any uh. Any gains that Intel might have had with the launch of sapphire Rapids are looking mighty, suspect.
Well, good competition is good, it’s good for you, you it’s good. For me, it’s good to have a lot of compute able to do things, I’m old level, one I’m signing out. You can find me in the level one forums.
Oh, I can’t wait for this to get here. Woo, woo, .