Question AMD Ryzen Threadripper 9000 “Shimada Peak” spotted with 32 and 64 Zen5 cores

[eek2121]

The author of Rapydmark is a much more accomplished developer than I could ever be and he didn't implement this so it's bothersome enough to be called trickery.

From the docs:

Starting with Windows 11 and Windows Server 2022, it is no longer the case that applications are constrained by default to a single processor group. Instead, processes and their threads have processor affinities that by default span all processors in the system, across multiple groups on machines with more than 64 processors.

 

[igor_kavinski]

Starting with Windows 11 and Windows Server 2022, it is no longer the case that applications are constrained by default to a single processor group. Instead, processes and their threads have processor affinities that by default span all processors in the system, across multiple groups on machines with more than 64 processors.

The docs are mad because I have not been able to run Rapydmark or my own benchmark on all 128 threads even when Process Lasso showed that there's a single NUMA domain of 128 cores. Tried Win11 and Win Server 2025.

 

[StefanR5R]

@igor_kavinski, the snippet which @eek2121 provided only pertains to how the kernel schedules threads of a process. When an application program sets out to spawn a bunch of threads, it first needs to find out how many logical CPUs there are. But to find out also about logical CPUs which exist outside of the group¹ on which the application program is running when it asks this, it needs to access an extended API. If it doesn't do so, the application program won't know of the rest of the CPUs and won't create as many threads in the first place as it would take to utilize the whole machine.

There are a several APIs to query hardware information such as the count of logical CPUs. The majority of them show only the number of logical CPUs in the current CPU group. One of the APIs which are limited like this is GetLogicalProcessorInformation. This one has received a counterpart API which also reveals logical CPUs outside of the currently used CPU group: GetLogicalProcessorInformationEx. If your application program does not access the latter API (or a similarly extended one if there is any), then the change of default kernel behavior which @eek2121 quoted does not matter (much).

________
¹) Edit: Processor groups are defined by the Windows operating system. It never puts more than 64 logical CPUs into one processor group.

 

[igor_kavinski]

Thanks, @StefanR5R ! Need to play with that extended API call then.

 

[soresu]

Consoles use GDDR6 so definitely using some sort of latency hiding mechanism.

That's consoles though - different animal altogether from PC gaming.

Having ultra limited potential hw configs to code for vastly increases relevant coding expertise in the job market and software optimisation over time.

Games made at the end of a console life cycle (like The Last of Us 1 or GTA5 for PS3) can get far more out of the hw at release time than any PC or Android game/app will ever manage on equivalent spec hw.

Game dev tools/internal code libraries + console SDK/proprietary APIs + 3rd party middleware are optimised continually for the 5+ years of the hw life cycle.

I can only imagine how badly modern Switch games would perform on a nVidia Switch TV, or equivalent Android hw from 2015 having to account for so many possible different Android hw configurations.

This is why DICE and AMD originally teamed up to create Mantle which started the low level gfx API ball rolling in the industry, to (at least partially) address this imbalance.

 

[iamgenius]

It is official now:

https://www.amd.com/en/products/processors/workstations/ryzen-threadripper.html#tabs-705187c2a6-item-ff624515f5-tab

AMD Introduces New Radeon Graphics Cards and Ryzen Threadripper Processors at COMPUTEX 2025

– Radeon RX 9060 XT brings next-gen gaming experiences and FSR 4 ML-powered upscaling to mainstream users – – Threadripper 9000 Series…...

ir.amd.com

AMD Announces Ryzen Threadripper PRO 9000 "Shimada Peak" HEDT Processors

AMD at Computex 2025 announced the Ryzen Threadripper PRO 9000 series processors targeting high-end desktops (HEDT) and workstations. These processors are codenamed "Shimada Peak," and are based on the "Zen 5" microarchitecture. "Shimada Peak" is a variation of the EPYC "Turin" MCM, designed for...

www.techpowerup.com

Can't wait to see the reviews and overclocking results. Will be waiting for July.

 

[igor_kavinski]

Can't wait to see the reviews and overclocking results. Will be waiting for July.

I'll only be mildly interested because no one dares to run game benchmarks on these. What a waste of a really expensive machine.

 

[Thibsie]

Well, it makes a really really expensive gaming machine.

 

[igor_kavinski]

Well, it makes a really really expensive gaming machine.

True, that may become available on the used market five years into the future

 

[iamgenius]

I'll only be mildly interested because no one dares to run game benchmarks on these. What a waste of a really expensive machine.

Well, you buy them for other purposes but you can still game with them no problem. I myself run older titles, so with a good GPU, it won't be an issue at all.

 

[igor_kavinski]

I myself run older titles, so with a good GPU, it won't be an issue at all.

Someone tell that to the reviewers. I bet they run all sorts of personal stuff on the review sample machines and get all giddy like kids but don't share the results.

 

[iamgenius]

I'll only be mildly interested because no one dares to run game benchmarks on these. What a waste of a really expensive machine.

These are some gaming benchmarks I could find for the 7000 series.

AMD Ryzen Threadripper 7980X and 7970X Review

It's not every day that we get to review a $5,000 CPU, so this should be fun. The Ryzen Threadripper 7980X and 7970X pack 64 and 32...

www.techspot.com

AMD Ryzen Threadripper 7980X & 7970X Review: Many-Core Desktop Supremacy - Page 5

AMD returns to the HEDT market with its most powerful, many-core processors yet, the Ryzen Threadripper 7000 series. - Page 5

hothardware.com

AMD Ryzen Threadripper 7980X review (Page 22)

We review the all-new Ryzen Threadripper 7000 series processors. The many-core beasts are paired with a TRX50 motherboard. We tested the Ryzen Threadripper 7980X with is 64-core based on Zen4 (that's 128-threads by the way). And then we also will place a focus on the Ryzen Threadripper 7970 with...

www.guru3d.com

 

[StefanR5R]

AMD made some Threadripper 9000 slides for their "Advancing AI 2025" event:
via computerbase.de
via techpowerup.com

The slides show the changes relative to Threadripper 7000.
Existing sTR5 (TRX50, WRX90) mainboards are forward compatible by a BIOS update.

 

[iamgenius]

Some more info:

https://videocardz.com/newz/amd-ryzen-threadripper-9000-set-to-launch-in-july-up-to-26-faster-than-tr-7000-series

 

[adamge]

You can buy AM5 with 256GB, at least this month G.Skill started to advertise their 256GB AM5 kit.

My work finally gave me something I asked for in our VM environment. I'm referring to the RAM kit. I definitely did not ask for the Intel CPU.

Threadripper-Who?

 

[gdansk]

My work finally gave me something I asked for in our VM environment. I'm referring to the RAM kit. I definitely did not ask for the Intel CPU.

Threadripper-Who?

View attachment 126813

ECC? I saw one kit for W880 but it was $1400...

 

[adamge]

ECC? I saw one kit but W880 but it was $1400...

Not ECC.

 

[igor_kavinski]

View attachment 126813

Running at 5600 MT/s? What timings?

 

[Thunder 57]

My work finally gave me something I asked for in our VM environment. I'm referring to the RAM kit. I definitely did not ask for the Intel CPU.

Threadripper-Who?

View attachment 126813

LOL! A 265K? No ECC? Whoever decided on that should be fired for incompetence.

 

[adamge]

Running at 5600 MT/s? What timings?

Regarding the ECC, we've been running a VMware environment internally for our lab use for probably 15 years - the last 6 have been on plain consumer workstation hardware like AMD Zen3 and without ECC. I've never noticed it being a problem. No random crashes or corruption. And if we did have one VM failure per year or month, it's not very expensive internally for us to correct it and move on.

We didn't pay attention to timing or speed config in the BIOS. This is how it default booted. I would have paid attention but I didn't do the install.

Here's the Linux diagnostic readout.

Handle 0x000F, DMI type 17, 92 bytes
Memory Device
Array Handle: 0x000E
Error Information Handle: Not Provided
Total Width: 64 bits
Data Width: 64 bits
Size: 64 GB
Form Factor: DIMM
Set: None
Locator: Controller0-ChannelA-DIMM0
Bank Locator: BANK 0
Type: DDR5
Type Detail: Synchronous
Speed: 5600 MT/s
Manufacturer: Kingston
Serial Number: 80192D1A
Asset Tag: 9876543210
Part Number: KF556C36-64
Rank: 2
Configured Memory Speed: 5600 MT/s
Minimum Voltage: 1.1 V
Maximum Voltage: 1.999 V
Configured Voltage: 1.1 V
Memory Technology: DRAM
Memory Operating Mode Capability: Volatile memory
Firmware Version: Not Specified
Module Manufacturer ID: Bank 2, Hex 0x98
Module Product ID: Unknown
Memory Subsystem Controller Manufacturer ID: Unknown
Memory Subsystem Controller Product ID: Unknown
Non-Volatile Size: None
Volatile Size: 64 GB
Cache Size: None
Logical Size: None

 

[igor_kavinski]

LOL! A 265K? No ECC? Whoever decided on that should be fired for incompetence.

He would get fired anyway if he told them how much an ECC workstation would cost

 

[Doug S]

Regarding the ECC, we've been running a VMware environment internally for our lab use for probably 15 years - the last 6 have been on plain consumer workstation hardware like AMD Zen3 and without ECC. I've never noticed it being a problem. No random crashes or corruption. And if we did have one VM failure per year or month, it's not very expensive internally for us to correct it and move on.

Modern DRAM is very reliable, and DDR5 includes internal ECC. There was a lot of negative speculation at the time that was announced that DRAM OEMs would use it as a way of utilizing "known bad" DRAM chips, but their processes yield nearly perfectly thanks to post manufacturing testing / ability to map in spare rows before they're assembled into DIMMs/stacks/etc.

I can't remember the last time I tested brand new RAM using memtest86+ and actually found errors. Maybe when DDR1 was brand new? Their testing/remediation is just so much better now. However I have seen the occasional error when testing several year old RAM - that's what the DDR5 built in ECC will hopefully (silently, but you can't have everything) address. Over time DRAMs that were "perfect" can become less so, and that's not something they can catch during post manufacturing testing, but is what ECC DIMMs address (plus stuff unrelated to manufacturing issues like cosmic rays)

It sounds like in your situation you can tolerate the occasional crash/corruption which means the added cost may not be worth it to your or your management. It is when you cannot tolerate any such issues or need/wish to reduce the problem space when diagnosing issues by immediately ruling out bad RAM as a cause that the added cost of ECC is justified.

Clearly, issues are statistically more likely when you have a lot of RAM. If you have 4 TB of RAM in a server you're already 1000x more likely to see problems than with 4 GB of RAM, and that's before you consider that if you max out capacity the DIMMs you buy will be using the latest and most aggressive DRAM process to deliver the densest DRAM chips.

 

[aigomorla]

Being a Threadripper 7960X owner:



i can say you can not trust AMD with HEDT, and at its value point your probably better off getting a EPYC and a regular Ryzen 9 X3D desktop. If you can afford one, this is probably a much better solution, and use both on a dedicated KVM to flip between PC's.

Thread ripper has no place right now... its like half of a EPYC and Ryzen 9 X3D, trying to fit in 1 package.

Until Threadripper PRO can at least compete against a X3D and still cost less then a 2/3rds of a single socket prowess of a EPYC and NOT kill your warranty when overclocking, i do not recommend it at all.

Just don't overclock, hence why EPYC is a better solution, and game on a dedicated X3D Ryzen 9 with overclocking.
OH and did i mention if you enable EXPO only u still get that NAG warning that you will VOID your warrenty.



AMD says they will look at it with a case by case, but i believe you still blow that overclocking fuse that the new Threadrippers have.

He would get fired anyway if he told them how much an ECC workstation would cost

Then what would you call me to use a ECC hybrid Gaming / production PC? :X
I needed it for the PCI-E Lanes... If Ryzen 9 X3D had double the current PC lanes, i would not have gotten Threadripper.


My ECC are the Gskill 192GB 6400mhz DDR5


I have no issues in windows 11, with EXPO on even, but i think i blew my overclocking warrenty fuse by doing so.
So i hope i never need to RMA my CPU and find out if AMD will honor it just for enabling EXPO.

I also use an ASUS board, so i know ASUS will be absolutely WORTHLESS if i ever need to RMA it.

 

[iamgenius]

Being a Threadripper 7960X owner:

View attachment 127053

i can say you can not trust AMD with HEDT, and at its value point your probably better off getting a EPYC and a regular Ryzen 9 X3D desktop. If you can afford one, this is probably a much better solution, and use both on a dedicated KVM to flip between PC's.

Thread ripper has no place right now... its like half of a EPYC and Ryzen 9 X3D, trying to fit in 1 package.

Until Threadripper PRO can at least compete against a X3D and still cost less then a 2/3rds of a single socket prowess of a EPYC and NOT kill your warranty when overclocking, i do not recommend it at all.

Just don't overclock, hence why EPYC is a better solution, and game on a dedicated X3D Ryzen 9 with overclocking.
OH and did i mention if you enable EXPO only u still get that NAG warning that you will VOID your warrenty.

View attachment 127054

AMD says they will look at it with a case by case, but i believe you still blow that overclocking fuse that the new Threadrippers have.




Then what would you call me to use a ECC hybrid Gaming / production PC? :X
I needed it for the PCI-E Lanes... If Ryzen 9 X3D had double the current PC lanes, i would not have gotten Threadripper.


My ECC are the Gskill 192GB 6400mhz DDR5
View attachment 127052

I have no issues in windows 11, with EXPO on even, but i think i blew my overclocking warrenty fuse by doing so.
So i hope i never need to RMA my CPU and find out if AMD will honor it just for enabling EXPO.

I also use an ASUS board, so i know ASUS will be absolutely WORTHLESS if i ever need to RMA it.

Okay, this is certainly bad. We will see if they confirm this overclocking fuse thingie when reviewing 9000 series threadrippers.

 

[igor_kavinski]

I have no issues in windows 11, with EXPO on even, but i think i blew my overclocking warrenty fuse by doing so.
So i hope i never need to RMA my CPU and find out if AMD will honor it just for enabling EXPO.

You are being too paranoid, methinks, probably coz of the sticker shock you suffered.

I violated and tortured my 9950X3D in ways that no decent CPU should have to endure but I was always able to revive it successfully one way or the other. It's been off for ages so no idea if it will turn on NOW if I try it, especially since the mobo is ASROCK with stock BIOS and I did not update to the recommended BIOS.

My Zen 2 Epyc also was running fine the last time I tried it, despite it being used almost 24/7 by the previous owner.

I think you are giving AMD too little credit by thinking that you may need to RMA your TR CPU. If it dies, it's probably due to ASUS's crappy BIOS validation and not due to something you do to it. In that event, give ASUS hell and get them to replace your TR CPU or use everything in your power (your industry contacts whatever) to give them a lesson they won't soon forget.

At least from my personal experience, it's not easy to kill an AMD CPU.