• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

AMD Readies Radeon PRO W9000 Series Powered by RDNA 4

btarunr

Editor & Senior Moderator
Staff member
Joined
Oct 9, 2007
Messages
47,792 (7.40/day)
Location
Dublin, Ireland
System Name RBMK-1000
Processor AMD Ryzen 7 5700G
Motherboard Gigabyte B550 AORUS Elite V2
Cooling DeepCool Gammax L240 V2
Memory 2x 16GB DDR4-3200
Video Card(s) Galax RTX 4070 Ti EX
Storage Samsung 990 1TB
Display(s) BenQ 1440p 60 Hz 27-inch
Case Corsair Carbide 100R
Audio Device(s) ASUS SupremeFX S1220A
Power Supply Cooler Master MWE Gold 650W
Mouse ASUS ROG Strix Impact
Keyboard Gamdias Hermes E2
Software Windows 11 Pro
AMD is readying a new line of professional graphics cards based on its latest RDNA 4 graphics architecture. The company has assigned the silicon variant "Navi 48 XTW" to power its next flagship pro-vis product, which will likely be branded under the Radeon PRO W9000 series. According to the source of this leak, the card comes with 32 GB of memory, which is probably ECC GDDR6, across the chip's 256-bit wide memory bus. The product should offer the same core-configuration as the Radeon RX 9070 XT gaming GPU, with 64 compute units worth 4,096 stream processors, 128 AI accelerators, 64 RT accelerators, 256 TMUs, and 128 ROPs.

Besides professional visualization, AMD could target the AI acceleration crowd. The company is hosting the "Advancing AI" press event in June, where it is widely expected to announce its next-generation AI GPUs and updates to ROCm. It could also use the occasion to unveil the Radeon PRO W9000 series product, promoting them to the AI acceleration crowd.



View at TechPowerUp Main Site | Source
 
Was just wondering when we'd see these, earlier today.
-Also, here's the 32GB Navi 48 everyone wanted :laugh:

What's expected pricing? $2,999? W7800 was $2,499, according to the TPU GPU...
 
Was just wondering when we'd see these, earlier today.
-Also, here's the 32GB Navi 48 everyone wanted :laugh:

What's expected pricing? $2,999? W7800 was $2,499, according to the TPU GPU...
It's cheap for actually using a GPU for real and not wasting one on gaming.

I don't get what's so hard about this. Programable GPUs kinda kicked off the nvidia 6800 Ultra. That was the end of GPU for gaming. CUDA kicked off the the 8800 GTX and gaming has been a distant second since then. Gaming doesn't matter. PC gaming especially is the pitty fuck at best. The only gaming things out now are moves to the cloud and slapping RGB on everything. Outside of those two things nothing is for gamers.
 
dos the ROCm work well on windows?

No, it does not. To the best of my knowledge, ROCm on Windows is, as of today, on a conceptual stage and not supported by any consumer-grade or pro-viz graphics card. Of course, there's a way to run it under WSL, but that's just the Linux version of ROCm that works on the few models they support (7900 XTX, etc.)

It's cheap for actually using a GPU for real and not wasting one on gaming.

I don't get what's so hard about this. Programable GPUs kinda kicked off the nvidia 6800 Ultra. That was the end of GPU for gaming. CUDA kicked off the the 8800 GTX and gaming has been a distant second since then. Gaming doesn't matter. PC gaming especially is the pitty fuck at best. The only gaming things out now are moves to the cloud and slapping RGB on everything. Outside of those two things nothing is for gamers.

At $2,999 MSRP, relatively tough sell if the only reason you're buying is to get 32 GB VRAM. The RTX 5090 will deliver that with far superior performance. At this price, the only situation I see this card being viable is to honor ISV support contracts.
 
So this is the 32GB version of the AMD Radeon RX 9070 XT.
Too bad it will cost double or more!
 
It's a Radeon Pro is already know it'll be priced at a very noticeable premium. It's a RX 9070 XT with a bit better driver support and 32GB, but it'll probably be priced significantly higher in spite of costing like maybe $100's more to produce same tired situation from GPU maker's. Hopefully Intel might actually be on the ball with something that's good value for dollar with plenty of VRAM when they do finally come around and launch something to replace the Arc A770 more credibly than the B580 that was a bit of more efficient side grade and no where to be found in terms of stock unless you count heavily scalped price jack up.
 
So this is the 32GB version of the AMD Radeon RX 9070 XT.
Too bad it will cost double or more!

At least triple, since it's a professional GPU and the reason these cost more is because of the driver validation process. But if it costs too much, then it's not worth getting in the largest majority of cases, since the RTX 5090 is a better card in every regard. Need a card for encoding? 5090 is better. Need a card for compute? 5090 is better. Need 32 GB? 5090 is better, its memory is far faster. Need a card for AI? 5090 is better. IMHO, AMD has to price this product very carefully for it to be successful, in fact, if they asked $1499, that'd be a solid middle ground between the RTX 5080 and 5090, around triple the gaming 9070 XT's price.

It's a Radeon Pro is already know it'll be priced at a very noticeable premium. It's a RX 9070 XT with a bit better driver support and 32GB, but it'll probably be priced significantly higher in spite of costing like maybe $100's more to produce same tired situation from GPU maker's. Hopefully Intel might actually be on the ball with something that's good value for dollar with plenty of VRAM when they do finally come around and launch something to replace the Arc A770 more credibly than the B580 that was a bit of more efficient side grade and no where to be found in terms of stock unless you count heavily scalped price jack up.

Drivers aren't better, they're just certified by software vendors and have some of the artificial restrictions lifted. This is the whole shtick with pro-viz cards.
 
At least triple

clap.gif
 
At least triple, since it's a professional GPU and the reason these cost more is because of the driver validation process. But if it costs too much, then it's not worth getting in the largest majority of cases, since the RTX 5090 is a better card in every regard. Need a card for encoding? 5090 is better. Need a card for compute? 5090 is better. Need 32 GB? 5090 is better, its memory is far faster. Need a card for AI? 5090 is better. IMHO, AMD has to price this product very carefully for it to be successful, in fact, if they asked $1499, that'd be a solid middle ground between the RTX 5080 and 5090, around triple the gaming 9070 XT's price.



Drivers aren't better, they're just certified by software vendors and have some of the artificial restrictions lifted. This is the whole shtick with pro-viz cards.

Fair enough broader support for isolated incidents AMD might need to help resolve for a few things and less arbitrarily restrictive features aimed at professional workloads as opposed to consumer entertainment.
 
dos the ROCm work well on windows?
Well? No. But you can run some stuff with it.
If all you need is creating some images or running widely-used LLMs, then it works after a bit of fussing around.
But it definitely doesn't work well in the sense of professional usage as the GPU in the post is meant to be used for.

No, it does not. To the best of my knowledge, ROCm on Windows is, as of today, on a conceptual stage and not supported by any consumer-grade or pro-viz graphics card. Of course, there's a way to run it under WSL, but that's just the Linux version of ROCm that works on the few models they support (7900 XTX, etc.)
You can get the runtime on windows itself and run some stuff on top of it.
Anything more intricate it going to be awful and Linux is the preferred environment for that anyway.
Nonetheless, lots of that stuff is slowly fading away from windows as well, even when it comes to CUDA. Tf does not support windows with GPUs anymore, and Nvidia recommends people to use WSL whenever possible.
 
Looking forward to some benches in Solidworks.
 
if u ask me all that shit ai crap should be expansion cards not in damn gpu's
 
if u ask me all that shit ai crap should be expansion cards not in damn gpu's
GPUs are expansion cards. Have always been. Did you ever buy a GPU after DX 9.0c? If you did congrats you voted for this with your money. PC gaming sucks. It's a waste of resources. You've been out voted. If you want to game on the PC in about a decaded you will rent hours on the cloud. That is PC gaming. This has always been PC gaming. The HIV of gaming. Accept it. You did this. You own this. If you don't want an APU based system and renting hours in the cloud run as far and as fast from the PC as you can.
 
Another expensive workstation card, what's new?
Still using my amazing Radeon Pro VII
I also have a Radeon Pro w5000 and w6000, best to wait 2 years for a good price drop haha.
I always wonder why workstation gpus are always price so higher, anyone tell me please, because it doesn't make any sense.
 
At least triple, since it's a professional GPU and the reason these cost more is because of the driver validation process. But if it costs too much, then it's not worth getting in the largest majority of cases, since the RTX 5090 is a better card in every regard. Need a card for encoding? 5090 is better. Need a card for compute? 5090 is better. Need 32 GB? 5090 is better, its memory is far faster. Need a card for AI? 5090 is better. IMHO, AMD has to price this product very carefully for it to be successful, in fact, if they asked $1499, that'd be a solid middle ground between the RTX 5080 and 5090, around triple the gaming 9070 XT's price.

K but is the 5090 actually available?
 
GPUs are expansion cards. Have always been. Did you ever buy a GPU after DX 9.0c? If you did congrats you voted for this with your money. PC gaming sucks. It's a waste of resources. You've been out voted. If you want to game on the PC in about a decaded you will rent hours on the cloud. That is PC gaming. This has always been PC gaming. The HIV of gaming. Accept it. You did this. You own this. If you don't want an APU based system and renting hours in the cloud run as far and as fast from the PC as you can.
There were few suggestions for cloud-based games. Do you know what happened to them? Are they dead? Because of one simple thing - latency, and I don't see that changing anytime soon.
 
but that's just the Linux version of ROCm that works on the few models they support (7900 XTX, etc.)
You can get ROCm to work on older AMD cards. That part actually isn't difficult; the problem is that ROCm itself is difficult, lol. When it works, it works quite well, but getting it to work ... well, depending on what software you want to use with ROCm, that ranges from annoying to impossible. And this is on Linux. CUDA, by contrast, is supported by basically everything, by default.

igor's right, though. I don't think Windows has much of a future for this sort of workload, over the long term. It's just too inflexible, and the use case for these workloads is niche enough that maintaining support for Windows feels like a hard sell. Presumably Microsoft agrees, given their WSL push over recent years.

In Linux, I can do goofy shit like pairing an AMD GPU wth an Nvidia GPU and use both to share a single AI inference task, through Vulkan. It ain't especially fast, but it also isn't painfully slow--and it's hilarious that I can do it at all. Hypothetically I could toss an Arc GPU in there and run inference on three different GPUs, from three different companies, using three different drivers. What's even funnier is that this clownish scheme is actually easier to set up than ROCm, in many cases.

It'll be interesting to see how AMD's software ecosystem evolves after UDNA. Will they jettison older architectures? I hope not, but it wouldn't shock me.
 
Yeah, these are gonna be wildly expensive, and yeah, a 5090 is going to be the better buy in most circumstances barring maybe B2B, but usually Radeon PRO cards are used in a niche where it's uniquely capable. W7800s were used because it supported a wide range of esoteric displays and had an absolute monster DisplayPort coming out the back, nevermind the spacious memory it had for stuff like medical imaging devices. I imagine the Navi 48 Radeon PRO cards will have similar talents to sell.
 
Last edited:
K but is the 5090 actually available?

IMG_2847.jpg

and what makes you think that this will be "easy to get"?

Another expensive workstation card, what's new?
Still using my amazing Radeon Pro VII
I also have a Radeon Pro w5000 and w6000, best to wait 2 years for a good price drop haha.
I always wonder why workstation gpus are always price so higher, anyone tell me please, because it doesn't make any sense.

Like I said, largely due to driver validation. Since they're aimed at businesses, the large margins also help fund R&D for future architectures

You can get ROCm to work on older AMD cards. That part actually isn't difficult; the problem is that ROCm itself is difficult, lol. When it works, it works quite well, but getting it to work ... well, depending on what software you want to use with ROCm, that ranges from annoying to impossible. And this is on Linux. CUDA, by contrast, is supported by basically everything, by default.

igor's right, though. I don't think Windows has much of a future for this sort of workload, over the long term. It's just too inflexible, and the use case for these workloads is niche enough that maintaining support for Windows feels like a hard sell. Presumably Microsoft agrees, given their WSL push over recent years.

In Linux, I can do goofy shit like pairing an AMD GPU wth an Nvidia GPU and use both to share a single AI inference task, through Vulkan. It ain't especially fast, but it also isn't painfully slow--and it's hilarious that I can do it at all. Hypothetically I could toss an Arc GPU in there and run inference on three different GPUs, from three different companies, using three different drivers. What's even funnier is that this clownish scheme is actually easier to set up than ROCm, in many cases.

It'll be interesting to see how AMD's software ecosystem evolves after UDNA. Will they jettison older architectures? I hope not, but it wouldn't shock me.

Haha, wow, that's insane. I kinda want to see it now
 
Another expensive workstation card, what's new?
Still using my amazing Radeon Pro VII
I also have a Radeon Pro w5000 and w6000, best to wait 2 years for a good price drop haha.
I always wonder why workstation gpus are always price so higher, anyone tell me please, because it doesn't make any sense.
It's in the name, they all pitch a higher price.
 
and what makes you think that this will be "easy to get"?

Nothing? that is kinda the point lol.
You make a comparison that card X will be better for a lot of tasks, my point is that that does not matter as card X is unavailable and this will thus sell like hotcakes as a result.... which you support by insinuating this will be hard to get as well....
 
Another expensive workstation card, what's new?
Still using my amazing Radeon Pro VII
I also have a Radeon Pro w5000 and w6000, best to wait 2 years for a good price drop haha.
I always wonder why workstation gpus are always price so higher, anyone tell me please, because it doesn't make any sense.
ISV certification
 
dos the ROCm work well on windows?
Not yet.

There are however apps on windows that support ROCm Like LM studio.

TechPowerUp are you going to start testing Ai reasoning models anytime soon with GPUs?
I asked for this also would like to see them add LM studio and pick a standard set of LLMs

8B to 14B models for lower vram cards so 8GB to 16GB
27B+ for cards with 24GB of vram and above.

Then have a generic query.

I was compiling something like this from info I found on the internet awhile ago but haven't gone back to it yet.

1745248416213.png
 
Last edited:
Back
Top