• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD announces Ryzen Threadripper 9000 series: 9955WX with 96 Zen5 cores and 5.4 GHz boost

The "enthusiast" keyword on a TR announcement caught my eye, since they've been non-forthcoming about Lenovo-locking the sub-$10K SKU's in past Threadripper launches. For anyone unfamiliar, that meant not only could you not buy the CPUs standalone, but they made them exclusive to Lenovo workstations - going as far as BIOS-locking them to Lenovo MB's.

I don't trust Videocardz' often wildly wrong slop, so went straight to the source press release to see if the "enthusiast" and "DIY" language was actually used - surprisingly it was.

So either it's "just kidding we didn't abandon DIY/HEDT/Enthusiast after all", or it may be more Lenovo timed exclusivity and we won't find out until later.
 
Last edited:
The "enthusiast" buzzword on a TR announcement caught my eye, since they've been so deceptive about Lenovo-locking on the sub-$10K SKU's for past Threadripper launches. I don't trust Videocardz' often wildly wrong assumptions, so went straight to the source press release to see if the "enthusiast" and "DIY" language was actually used. Which surprisingly it was.

So either it's "just kidding we didn't abandon DIY/HEDT/Enthusiast after all", or it'll be more Lenovo lock fuckery and we won't find out until later.
It could also be a case of a time lock on the CPU’s. Or there are so few MoBo’s that support the chip that it might as well be a Lenovo exclusive.
 
  • Like
Reactions: DPI
like this
It could also be a case of a time lock on the CPU’s. Or there are so few MoBo’s that support the chip that it might as well be a Lenovo exclusive.
Yeah I was assuming time lock as a possibility, but we wouldn't know because they don't disclose it. Like when I bought my 5965WX, it was after a year of Lenovo exclusivity, they announced the next product cycle a week later, and AIB support for WRX80 fell off a cliff. I decided I'd skip buying scraps thrown out at the end of the product cycle if given the chance again.

The other Threadripper gotcha at the time was not finding out until post-sale that memory bandwidth was gimped on lower core count SKU's due the CCX/IMC interdependency. So if you had let's say a 16-core 5955WX and all four memory channels populated, you were only getting 1/4 the memory throughout of the 64-Core SKU since only two CCX's were active. AMD never disclosed this, and none of the reviews picked up that mem bandwidth scaled with core count - maybe because they were all sampled the 64-core. It was an obscure Phoronix article that finally told the tale.

So I'm most curious if memory throughput will be tied to core count / pricepoint again, and if not, the degree of improvement.
 
Last edited:
Interesting news. Hopefully there will be DIY/HEDT.
 
What would an enthusiast do with 96 cores?

I totally agree with the idea of not wanting them vendor locked, but calling it an enthusiast part seems gimmicky. High end workstation? Absolutely.
 
What would an enthusiast do with 96 cores?

I totally agree with the idea of not wanting them vendor locked, but calling it an enthusiast part seems gimmicky. High end workstation? Absolutely.
Threadripper really is a graphic studio platform more than anything. You save on the per unit cost without going to EPYC which is now considerably more once you add in the platform cost.
 
I lament the loss of "enthusiast mixed use HEDT" from both Intel and especially AMD. I'd love to see the latter bring it back, especially in an era where even high end Ryzen X870E boards are still limited to dual channel RAM and are so stingy when it comes to PCI-E lanes; you can't even get 2x PCI-E 5.0 x16 lanes to both function at full speed, not to mention how even mid-level gaming desktops are making use of PCI-E lanes thanks to NVMe SSDs! Years ago when they broke the 8-core threshold on Ryzen and brought 16c/32t CPUs on the mainstream platform (along with the 'new' direction for Threadripper) I thought perhaps the rest of the platform would have variants to expand other features on certain CPUs and chipsets, but it has not been the case.

If Threadripper 9000 was capable of consistent single/few core performance and boosting/OC at least parallel to what one can expect from Ryzen 9000 X3D CPUs, kept to the rest of the announced features like quad channel RAM and 128 PCI-E 5.0 lanes, didn't have the kind of gotchas that DPI mentioned above, ensured that motherboards based on the updated chipset were widely available like inthe old days (Bring back a a ROG Zenith release etc, they could have a significant enthusiast userbase uptake if they priced the thing even vaguely reasonably. Come to think of it, I'm curious about cache accessibility vs something like the Ryzen 9000 X3D chips. In theory, wouldn't the 32c version (or definitely the 64) for example have access to similar amounts of cache as the 9800X3D / 9950X3D but without the asymmetrical arrangement?
 
Interesting news. Hopefully there will be DIY/HEDT.
What would an enthusiast do with 96 cores?

I totally agree with the idea of not wanting them vendor locked, but calling it an enthusiast part seems gimmicky. High end workstation? Absolutely.
There will be HEDT, the text says, 9970X(32 core) and 9960X(24 core).
With these versions enthusiasts will for example play the new GTA 6, paired with the new RTX 6090 which should also be available by the time the GTA 6 PC version is released.
 
How do you think AMD will price them? I hope they will be little less than the previous generation. Hopefully something like $2400 for the 9970X.
 
What would an enthusiast do with 96 cores?

Compiling FreeBSD or the Linux kernel several times a day. Compiling Rust or Chromium several times as week,

I am not that amused that it takes a full year from the consumer chips to the Threadrippers. They are outpaced by the next consumer line (in per-core speed) very quickly.
 
There will be HEDT, the text says, 9970X(32 core) and 9960X(24 core).
With these versions enthusiasts will for example play the new GTA 6, paired with the new RTX 6090 which should also be available by the time the GTA 6 PC version is released.
Not unless they have X3D versions...

Fuck, could you imagine 24X3D Cores?
 
Maybe AMD will surprise us later with an X3D Threadripper :)
 
Last edited:
Last edited:
It was always puzzling that they offered the big cache on EPYC but not on Treadripper. Maybe because of clock speed concerns?
 
What would an enthusiast do with 96 cores?

I like to mess with Unreal Engine. It can peg my 5950x at 100% for a solid 15 mins compiling shaders opening up a new project. (much different than compiling shaders for a packaged game)
I also develop some of my own multithreaded apps, which can also peg my 5950x doing tons of work.

If I had the money, I would buy one no question...
I'm sure they are many people with more budget than I have that would also.
 
Impressive in term of wattage AIO can do on that big of a die, but that only 15% higher score than a 7995WX air cooled with a manual OC. (the regular PBO score could be more impressive here depending of the cooling)
 
Last edited:
And the fact that a motherboard can deliver 947w at 1.2v... what the hell, my Asus Sabertooth board can get toasty with a 300w FX chip, how do they pull this off at 900w???

Motherboards have come a long ways in power delivery, mostly thanks to Intel due to their high power demands overclocked. But that's pretty close to the max wattage I would want to put through a socket for a short workload.
 
it is a $900+ motherboard that they seem to use:
https://www.newegg.com/asus-pro-ws-...c6V2Ec4OGIBpHpngGV3tzMwzDtj_EED-LHtA2WtLLyZDA
Our ASUS Pro WS TRX50-Sage Wi-Fi is based on an 18×2 (vcore) 3 (vdd11) 4 (vddio) 4-phase (vsoc ) power supply. This is the motherboard with the largest power supply stages we’ve ever had! The part in charge of the CPU is made up of 36 phases, but actually uses a new organization called “Teamed"
 
And the fact that a motherboard can deliver 947w at 1.2v... what the hell, my Asus Sabertooth board can get toasty with a 300w FX chip, how do they pull this off at 900w???
The Threadripper’s are basically cut down EPICS, and thanks to the MI350X the SH5 socket is rated to handle a little north of 1000w. And while that may have some extra GPU fidley bits built into it aid wager the power delivery translates over to what ever they are doing on Threadripper.
 
AMD announces Ryzen Threadripper 9000 series: 9955WX with 96 Zen5 cores and 5.4 GHz boost
"The Threadripper series is making a return at Computex. Following the release of Ryzen 9000 last year, the Zen 5 architecture is finally entering high-end workstations and enthusiast systems. The official launch is expected in July, about a year after the desktop models."
That’s exciting news! The new Threadripper 9955WX with 96 Zen 5 cores sounds super powerful. Can’t wait to see how it performs in real-world tasks!
 
Where will you buy the board?
I grabbed the WRX90 asus sage already, current WRX90 boards should support it with bios update. Went back and forth between the asrock and the asus, the asrock seems more stable but I value the multiple gen5 nvme slots in the Asus more. Hope it's not a POS like a lot of asus stuff lately. I currently have a 3990x rig that's been running 5 years flawlessly with a gigabyte board.

Will be populating this machine with 256gb ddr5, 4x samsung 4tb 9100 pro, a silverstone AIO cooler, silverstone RM52 rackmount case, a shitload of Noctua A12s, and a single RTX 6000 Pro 96gb to start (might go two depending on how long AI training takes).

Pretty wild the compute density we can get these days.
 
I grabbed the WRX90 asus sage already, current WRX90 boards should support it with bios update. Went back and forth between the asrock and the asus, the asrock seems more stable but I value the multiple gen5 nvme slots in the Asus more. Hope it's not a POS like a lot of asus stuff lately. I currently have a 3990x rig that's been running 5 years flawlessly with a gigabyte board.

Will be populating this machine with 256gb ddr5, 4x samsung 4tb 9100 pro, a silverstone AIO cooler, silverstone RM52 rackmount case, a shitload of Noctua A12s, and a single RTX 6000 Pro 96gb to start (might go two depending on how long AI training takes).

Pretty wild the compute density we can get these days.
Sweet baby Jesus, those are some extreme specs. I have a 64c Epyc workstation that feels like overkill and this computer you're describing would run circles around it lol.
 
Can you be specific or cite some examples.
I mean, they're in "a bad cycle" I watched Buildzoid's rambles on his experiences with all the motherboard manufacturers and his conclusion was that all of them have good and bad gens, which I notice as well with the Asus X3D incident, Asrock X3D incident, and, hardware wise, again, every manufacturer has its good and bad boards. Buildzoid's example was the asus maximus (something, but I only remember the chipset) X79, and then the Maximus after that which was X99. The X79 was a rock-solid, overbuilt, really nice board, but the X99 one was just mid-tier at best, and then unstable and buggy at worst. Other brands are the same.

I see this was a bit of a ramble, sorry.
 
That sounds like overkill for a lot of users. What is your use case that needs this much power (at a steep price!)
It's not a personal rig, it's for work. My personal rig is a maxed out MBP 13.

This computer and the 3990x rig I have are used for running RF and EM simulations, as well as training our own neural networks.

Right now, when we are doing product development and research each time we have to wait while training or simulating my engineers are sitting on their asses instead of iterating. In business, speed makes everything compound. If neural networks train twice as fast, my guys are experimenting and iterating twice as fast. I'm getting more bang for the buck out of each day, but more critically we'll finish product dev faster and get to market with new products and solutions faster. When those products and solutions hit, the revenue impact is millions of dollars - and if that revenue comes sooner than later, you can leverage that money sooner to build even MORE value.

The limitation is usually the software on the side, at some point infinite cores won't help software perform better that isn't built for it. But right now on the AI side it scales pretty well. If our use case required the power of Nvidia's even more crazy solutions I could justify it from a business standpoint (and probably will eventually), but right now I think 1-2 RTX 6000 Pros will chop our training time to like an hour, which is on the point of diminishing returns.

EDIT: on that topic, I think the 3990x is some of the best value I've ever gotten out of hardware. I can't believe after five years it's still providing very good performance for simulations and I see no reason to discontinue it. I just needed another compute rig because our team is growing and the resources were conflicted on the 3990x. But it will serve proudly alongside the 9985WX rig in the rack for several more years most likely.
 
Back
Top