To get down to a MATX, If Intel has support for the W Class, in the smaller formats, why not threadripper. Yes, I know this is a deep micro atx too!

HBSoundTo get down to a MATX, If Intel has support for the W Class, in the smaller formats, why not threadripper. Yes, I know this is a deep micro atx too!

Little in the past Taichi x399m

kapone32Yep the 2920x inspired me to only go with 12 cores. What would be even better would be a chip like the 1900X. That chip was like $30 more than a 1700X with more than double the PCIE lanes. Today's $500+ MBs are just porn. I still have my X399 and will be using it as a Virtual hub for some Racing rigs.

Yeah, ideally, starting at 8 cores would be nice. There are a lot of power users who needs lots of IO and memory and good core performance for productive tasks that are not giant batch jobs. (Not to mention it would double as a great gaming rig).
AMD (and Intel) are really screwing up by not satisfying the "HEDT market". Just imagine them selling cherry-picked 8 core ($500), 12 core ($650) and 16 core($800) with slightly higher clocks and higher TDP than mainstream, along with non-gimmicky motherboards at $400-500 (many well featured mainstream motherboards cost this anyways).

kapone32If X3D was a gimmick there would be lots of stock.

It mostly affects low resolution gaming and very select edge case workloads, while most heavy workloads see no real benefit if not a disadvantage from heat or throttling. It makes no sense for a workstation CPU.

I’m looking forward to a single CCD CPU with 12c/24t accompanied with an advanced high frequency I/O die. When something like that surfaces it would be hard for me to resist…
If X3D still makes sense at the time, even better.

Personally I’m ok with the existing PCI-E lane and memory channel count. I do not need more of them just as I don’t need more board cost.
Also I think DDR5 has a lot more to offer still.

efikkanAMD (and Intel) are really screwing up by not satisfying the "HEDT market"

What kind of workload do you have that isn't yet optimized for a lot of cores but does need a ton of PCIE lanes/extreme memory capacities?

Zach_01I’m looking forward to a single CCD CPU with 12c/24t accompanied with an advanced high frequency I/O die

Would be interesting to see for sure, but I just don't think it will happen for a few years? Maybe if Intel keeps pushing with their E-cores. Games won't bother to push for more than 8 coress until at least next generation of consoles, provided the latter happen to have more than 8 cores...

Zach_01Personally I’m ok with the existing PCI-E lane and memory channel count. I do not need more of them just as I don’t need more board cost

Same-ish. The grand majority of people won't plug anything other than two RAM sticks, a GPU card, maybe two NVME drives, tops. Maybe the odd SATA drive. So I don't think the current somewhat limited PCIE lane amount will change much either.

windwhirlWould be interesting to see for sure, but I just don't think it will happen for a few years? Maybe if Intel keeps pushing with their E-cores. Games won't bother to push for more than 8 coress until at least next generation of consoles, provided the latter happen to have more than 8 cores...

I believe that some of those if not all my expectations could be fulfilled with Zen6 on AM5.

TumbleGeorgeLittle in the past Taichi x399m

Exactly, why did they not keep running with this board and layout.
Perfect platorm! I reached out to Asrock to see they could upgrade the 3 PCIe 3.0 x16 to 4.0 lanes, and etc. They said not, but if I said I needed a thousand of them, they possible would have considered.

Zach_01I’m looking forward to a single CCD CPU with 12c/24t accompanied with an advanced high frequency I/O die. When something like that surfaces it would be hard for me to resist…
If X3D still makes sense at the time, even better.

Personally I’m ok with the existing PCI-E lane and memory channel count. I do not need more of them just as I don’t need more board cost.
Also I think DDR5 has a lot more to offer still.

That sounds like a job Arrow Lake could do! The 285K has all of its 24 threads (cores, even) on a single die, which is paired with an I/O-die that is certainly advanced and presumably high-frequency (Are you talking about memory clock?). It works well enough without extra cache (though more of it could have helped for sure) and all the rest of the platform you’ll likely find to be similarly robust.

So, there you go! Available for days short of a full three months so far, the future is now, it can be yours whenever you want.:rockout:

Name	Graphics Model	# of CPU Cores	# of Threads	Max. Boost Clock	Base Clock	Default TDP
AMD Ryzen™ Threadripper™ PRO 5995WX	Discrete Graphics Card Required	64	128	Up to 4.5 GHz	2.7 GHz	280W
AMD Ryzen™ Threadripper™ PRO 5975WX	Discrete Graphics Card Required	32	64	Up to 4.5 GHz	3.6 GHz	280W
AMD Ryzen™ Threadripper™ PRO 5965WX	Discrete Graphics Card Required	24	48	Up to 4.5 GHz	3.8 GHz	280W
AMD Ryzen™ Threadripper™ PRO 5955WX	Discrete Graphics Card Required	16	32	Up to 4.5 GHz	4 GHz	280W
AMD Ryzen™ Threadripper™ PRO 5945WX	Discrete Graphics Card Required	12	24	Up to 4.5 GHz	4.1 GHz	280W

AMD Ryzen™ Threadripper™ PRO 7945WX

Discrete Graphics Card Required

12

24

Up to 5.3 GHz

4.7 GHz

Not Included

350W

It would be great to see this generation of 12 cores and 24 threads reach a maximum clock speed of 6.0 GHz. The previous generation was 1.2 GHz slower (7945wx to 5945WX), so let's hope they achieve the 6.0 GHz mark.

windwhirlWhat kind of workload do you have that isn't yet optimized for a lot of cores but does need a ton of PCIE lanes/extreme memory capacities?

Pretty much any interactive "professional" workload has some kind of sweet-spot where more (slower) cores are outperformed by fewer faster cores etc. This includes development, CAD, 3D-modelling, graphics and multimedia. The workloads that do scale are async workloads, usually large batch workloads that are not interactive, like large simulations, build jobs, video transcoding etc. And what do all of these have in common? The end user usually don't sit and watch it happen, and if it's big enough the user will offload this to some kind of server. It is unfortunate that "workstation" benchmarks focus on these big (often synthetic) batch workloads which doesn't accurately represent workstation use.

In reality, the typical workstation user switches between multiple medium to heavy workloads that are mostly interactive. Responsiveness, productivity and stability is key, so balancing many aspects incl. core speed, core count, CPU performance consistency, storage IO, GPU performance, and memory bandwidth and capacity all may play a role.

Especially for "prosumers" who combine "work" and play, the old HEDT systems was excellent; flexibility, expandability and very consistent performance. Nowadays pretty much any heavy workload will benefit 2-4 high performance SSDs (1 OS, and however many for workloads, VMs, etc.), and running those through the chipset will result in terrible performance. Not to mention this will be competing for bandwidth with network, USB and SATA devices too, so there's not much use in PCIe gen 5 devices then.

If you want anything beyond a very basic setup for gaming or light office use, you'll have to do serious compromises. And with expensive "top" mainstream motherboards for both AMD and Intel being more limited than before, you'll have to live with your choice of limitations, or keep upgrading all the time. Unfortunately the mainstream platforms are becoming "increasingly useless" by trying to be a Jack of all trades, basically threading into "HEDT territory" with slightly more IO and cores, and especially pricing, but with heavy throttling/inconsistent performance and severe limitations. I'm not arguing that everyone needs a "HEDT" system, just like not everyone needs a pickup and a chainsaw, but carving out a proper HEDT segment again and moving the high-power CPUs there and let them have some breathing room would be great for everyone so the mainstream can be cheaper for gamers and basic office use.

HBSoundIt would be great to see this generation of 12 cores and 24 threads reach a maximum clock speed of 6.0 GHz. The previous generation was 1.2 GHz slower (7945wx to 5945WX), so let's hope they achieve the 6.0 GHz mark.

The AM5 CPUs will probably be a good indicator of what kind of clock ranges to expect, but I hope they do better cherry-picking and manage to push them a little further. Nevertheless, the higher TDP headroom and easier cooling will at the very least mean less throttling, so don't get too fixated on specs alone.

But keep in mind you're talking about Threadripper PRO though, the much more expensive siblings in the Threadripper family with 8 channels of memory and 128 PCIe lanes. It's kind of strange that the regular Threadripper (4 channel, 48+24 PCIe lanes) starts at 24 cores, yet the PRO offers 12 and 16 core variants. If there is a professional demand for all that IO on mere 12 and 16 cores, there should be no reason not to offer those for the regular Threadrippers too (or better yet 8 cores too).

To make things worse, Threadripper PRO isn't widely available, and have usually been very late to market.

eidairaman1Now if AMD would just merge the TR and Epyc Socket into 1 universal, that would lower their development costs and increase their revenue

I HIGHLY doubt that... Money is the core principle to 99% of businesses, if the data showed AMD that consolidation of those chips would lower production cost and increase profit they would of done that LONG ago... I mean what would there reason be NOT to do that then?

efikkanPretty much any interactive "professional" workload has some kind of sweet-spot where more (slower) cores are outperformed by fewer faster cores etc. This includes development, CAD, 3D-modelling, graphics and multimedia. The workloads that do scale are async workloads, usually large batch workloads that are not interactive, like large simulations, build jobs, video transcoding etc. And what do all of these have in common? The end user usually don't sit and watch it happen, and if it's big enough the user will offload this to some kind of server. It is unfortunate that "workstation" benchmarks focus on these big (often synthetic) batch workloads which doesn't accurately represent workstation use.

In reality, the typical workstation user switches between multiple medium to heavy workloads that are mostly interactive. Responsiveness, productivity and stability is key, so balancing many aspects incl. core speed, core count, CPU performance consistency, storage IO, GPU performance, and memory bandwidth and capacity all may play a role.

Especially for "prosumers" who combine "work" and play, the old HEDT systems was excellent; flexibility, expandability and very consistent performance. Nowadays pretty much any heavy workload will benefit 2-4 high performance SSDs (1 OS, and however many for workloads, VMs, etc.), and running those through the chipset will result in terrible performance. Not to mention this will be competing for bandwidth with network, USB and SATA devices too, so there's not much use in PCIe gen 5 devices then.

If you want anything beyond a very basic setup for gaming or light office use, you'll have to do serious compromises. And with expensive "top" mainstream motherboards for both AMD and Intel being more limited than before, you'll have to live with your choice of limitations, or keep upgrading all the time. Unfortunately the mainstream platforms are becoming "increasingly useless" by trying to be a Jack of all trades, basically threading into "HEDT territory" with slightly more IO and cores, and especially pricing, but with heavy throttling/inconsistent performance and severe limitations. I'm not arguing that everyone needs a "HEDT" system, just like not everyone needs a pickup and a chainsaw, but carving out a proper HEDT segment again and moving the high-power CPUs there and let them have some breathing room would be great for everyone so the mainstream can be cheaper for gamers and basic office use.


The AM5 CPUs will probably be a good indicator of what kind of clock ranges to expect, but I hope they do better cherry-picking and manage to push them a little further. Nevertheless, the higher TDP headroom and easier cooling will at the very least mean less throttling, so don't get too fixated on specs alone.

But keep in mind you're talking about Threadripper PRO though, the much more expensive siblings in the Threadripper family with 8 channels of memory and 128 PCIe lanes. It's kind of strange that the regular Threadripper (4 channel, 48+24 PCIe lanes) starts at 24 cores, yet the PRO offers 12 and 16 core variants. If there is a professional demand for all that IO on mere 12 and 16 cores, there should be no reason not to offer those for the regular Threadrippers too (or better yet 8 cores too).

To make things worse, Threadripper PRO isn't widely available, and have usually been very late to market.

As a 3D Professional I can 2nd this great post, i absolutely hate when power use becomes a driving factor over performance in workstation chips and HEDT chips... I want all the cores, all the watts.... Maya can NEVER have too much compute..

efikkanNot to mention this will be competing for bandwidth with network, USB and SATA devices too, so there's not much use in PCIe gen 5 devices then.

Tentatively, I’ve given you a thumbs up, although you might have to give me numbers on that, since to me, a dirty consumer peasant, bottom feeding nonetheless, those seem utterly insignificant. Or are these really the kinds of problems you solve? :wtf:

Bit RageAs a 3D Professional I can 2nd this great post, i absolutely hate when power use becomes a driving factor over performance in workstation chips and HEDT chips... I want all the cores, all the watts.... Maya can NEVER have too much compute..

In a world where most enthusiasts are more fixated on rated specs. and synthetic benchmarks or edge cases irrelevant to their usage, very few get to experience the difference between a "normal" computer and a proper workstation. Even in GamersNexus' Threadripper review mentions how hard it is to accurately measure how much better the user experience is. And as someone who has used/owned Sandy Bridge-E and Skylake-X and compared to their mainstream counterparts, there is no question that workstations are much more responsive and satisfying to work on, especially with mixed workloads. (Not to mention workstations are much more reliable and are built to handle sustained load.)

If I were paying people, whether it was development, CAD, or creative works, I would certainly get them the optimal rig for that workload, which in most such cases would mean high-end workstations of some sort.

I'm looking forward to see Zen 5 let it rip, with some proper headroom in terms of power. Then we'll see whether Threadripper scales better than the mainstream parts. Zen 5 did in fact offer a fair bit of substantial improvements on paper, and we know AVX-512 works well from server workloads, yet I would expect more thanks to large front-end improvements and integer performance.

As for specific details, I think it's a safe bet to assume the new Threadrippers will run at least DDR5-5600, hopefully DDR5-6400 like Epyc does. (Xeon 6 supports 8800 MHz MRMDIMMs, which would be great). Hopefully this means a refresh of motherboards, and hopefully some cost-reduced versions similarly like Asrock did with W790 WS R2.0 for Xeon W. But most importantly; motherboards with the socket in the correct orientation so we can put some good air coolers on top of it.

NoLoihiTentatively, I’ve given you a thumbs up, although you might have to give me numbers on that, since to me, a dirty consumer peasant, bottom feeding nonetheless, those seem utterly insignificant. Or are these really the kinds of problems you solve? :wtf:

And what's the question?
If you read the sentence before it, I was comparing mainstream platforms to high-end workstation/HEDT platforms, as the limited amount of PCIe lanes from the CPU means basically anything beyond 1 GPU and 1-2 SSDs must be connected through the chipset without major sacrifices. AMD currently offers x4 4.0 lanes through the chipset, while Intel offers x8 4.0 lanes. It should be obvious that anything connected through the chipset will then share that bandwidth in some fashion, and connecting a PCIe 5.0 device wouldn't get much speed though there as stated.
If you need more specifics I can provide more details.

Any idea when zen 5 threadripper CPUs will be announced?

mtosevAny idea when zen 5 threadripper CPUs will be announced?

Later in 2025 rumours.

mtosevAny idea when zen 5 threadripper CPUs will be announced?

Well, this whole thread is all about it! The leaked info present in this thread is the latest thing we got. There is no known date at which they will be announced. The leaked info shows AMD is still planning to launch them, and hopefully this will be true.