NVIDIA RTX PRO 6000 Blackwell Workstation Edition PCB Layout Leaked By Insider

[T0@st]

Over the past weekend, members of the Chiphell discussion board started posting truly NDA-busting photo material—one example made headlines a few days ago. A fairly convincing list of next-gen NVIDIA RTX PRO Blackwell series graphics cards appeared online just over a month ago; only a smattering of physical specimens have emerged since then. As pointed out by interested Chiphellers, Leadtek Chinese language websites have started listing a small selection of upcoming "Blackwell" generation professional SKUs.

The previously leaked PCB design was linked to Leadtek/NVIDIA's "blower-style" RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB model. A brave Chiphell forumite has shared shots of another alleged internal component; a shorter PCB design has come to light—in VideoCardz's expert opinion, this stubby unit is destined to be contained within the Leadtek-made (non-Max-Q) RTX PRO 6000 Blackwell Workstation Edition enclosure. Official Team Green promotional renders have already presented this darker alternative to existing Founders Edition gaming-oriented siblings. According to VideoCardz, Team Green's Professional Blackwell series review embargo is still in effect and official launch window information is still not a publicly-known quantity. The freshly leaked bare PCB seems to borrow design elements—namely a dual-sided GDDR7 memory module mounting setup—from NVIDIA's familiar GeForce RTX 5090 Founders Edition model.



View at TechPowerUp Main Site | Source

 

[bonehead123]

Looks cool & all, hopefully there will be at least 10-15 more of them available at launch time, hehehehe

 

[ncrs]

Looks cool & all, hopefully there will be at least 10-15 more of them available at launch time, hehehehe

RTX Pro 6000 is the reason for the lack of 5090 availability since they use the same chip.
I'm getting "around $9000" preliminary price for RTX Pro 6000 Workstation from my company's vendor, and the cheapest 5090 on newegg is a $1559 combo with a PSU ~$2900. There will be a lot of demand for RTX Pro due to its VRAM, so I expect they will sell very well.

Edit: fixed price since I mistook 5080 for a 5090.

 

[Jawz]

cheapest 5090 on newegg is a $1559 combo with a PSU

That's a 5080. Cheapest 5090 I see is $2,849.98. Still though, I see the point you're making.

 

[ncrs]

That's a 5080. Cheapest 5090 I see is $2,849.98. Still though, I see the point you're making.

You're right, sorry for the confusion.

 

[HBSound]

I like this unit, but I wish Nvidia would supply the GPUs of the 5090 so that everyone is not paying up to $2000 over retail for the 5090. I would almost make the sacrifice for this kind of GPU over the 5090, because the resale value down the road will be significantly more substantial for the professional GPU compared to the gaming GPU. I sold off a A6 not to long ago, and I used for almost three to four years, and what I paid for, I sold it for.

I would like to see in a real-world scenario what will push the limits of a GPU that can handle 96GB of VRAM. I deal in the world of CAD, 3D, BIM and rendering, and something like this has be thinking, do I actually have models that would consume a GPU like this???

 

[kondamin]

I like this unit, but I wish Nvidia would supply the GPUs of the 5090 so that everyone is not paying up to $2000 over retail for the 5090. I would almost make the sacrifice for this kind of GPU over the 5090, because the resale value down the road will be significantly more substantial for the professional GPU compared to the gaming GPU. I sold off a A6 not to long ago, and I used for almost three to four years, and what I paid for, I sold it for.

I would like to see in a real-world scenario what will push the limits of a GPU that can handle 96GB of VRAM. I deal in the world of CAD, 3D, BIM and rendering, and something like this has be thinking, do I actually have models that would consume a GPU like this???

you might not now, but it wont take long when you do have the vram as nothing will stop you from going further

 

[bonehead123]

I would like to see in a real-world scenario what will push the limits of a GPU that can handle 96GB of VRAM. I deal in the world of CAD, 3D, BIM and rendering, and something like this has be thinking, do I actually have models that would consume a GPU like this???

Well, my IT/procurement group has already started writing up purchase orders for these cards, as we do all of the above types of work, and they will be a great upgrade for our workstations, because our current RTX 5000/48GB cards are beginning to limit our productivity, especially with the live/real time renders that need to implement all the design/engineering changes/revisions/updates/inputs from the various teams....

This was the same scenario back when we had the 4x series cards, which worked well at that time, but our workloads have increased significantly since then, both in quantity and complexity....

And as for resale value, well I sold those cards for almost as much as they paid for them, and I suspect it will be the same this time around

 

[tps3443]

Looks cool & all, hopefully there will be at least 10-15 more of them available at launch time, hehehehe

I wish that was the problem. The big issue is that scalpers literally making $30,000 dollars in a day on RTX 5090’s. 1 guy was snatching up (12) 5090FE cards. These people run businesses. And literally OWN the inventory on and after launch day. So it becomes impossible to obtain. RTX 6090 will be worse.

 

[HBSound]

T0@st—When this GPU is released, will you test the "NVIDIA RTX PRO 6000" to check its performance for PCI-E Performance Scaling? I am curious: Does this GPU need all 16 PCI lanes, or can it achieve 95% of the performance on 8 PCI lanes?

Thank you in advance.

 

[tps3443]

T0@st—When this GPU is released, will you test it as you did for the 5090 to check the performance for PCI-E Performance Scaling? I am curious: Does this GPU need all 16 PCI lanes, or can it do 95% of the performance on 8 PCI lanes?

Thank you in advance.

I run a 5090 on PCIe Gen3 x16, this is equivalent to PCIe Gen4 x8, or PCIe Gen5 x4.

You will lose 4% performance at 1440P, and 2% performance loss at 4K. It is very minor! Sometimes there is no performance loss at all. But essentially with 4K you can expect 0-2% performance loss.

 

[HBSound]

I run a 5090 on PCIe Gen3 x16, this is equivalent to PCIe Gen4 x8, or PCIe Gen5 x4.

You will lose 4% performance at 1440P, and 2% performance loss at 4K. It is very minor! Sometimes there is no performance loss at all. But essentially with 4K you can expect 0-2% performance loss.

I'm sorry. I meant to insert the A6000, not the 5090. When I hit the copy and paste button, I did not replace the 5090 with the Workstation Pro A6000.

I updated my original post.

T0@st—When this GPU is released, will you test the "NVIDIA RTX PRO 6000" to check its performance for PCI-E Performance Scaling? I am curious: Does this GPU need all 16 PCI lanes, or can it achieve 95% of the performance on 8 PCI lanes?

 

[igormp]

I'm sorry. I meant to insert the A6000, not the 5090. When I hit the copy and paste button, I did not replace the 5090 with the Workstation Pro A6000.

I updated my original post.

T0@st—When this GPU is released, will you test the "NVIDIA RTX PRO 6000" to check its performance for PCI-E Performance Scaling? I am curious: Does this GPU need all 16 PCI lanes, or can it achieve 95% of the performance on 8 PCI lanes?

It depends on what kind of tasks you're talking about.
If you're thinking about games, it's going to behave a bit better than a 5090, and the PCIe scaling will be almost identical, which was already explained above.

For other tasks which might require more data transfers through PCIe (like model training), then the impact of PCIe bandwidth may become more evident, but that would also apply to a 5090.

 

[HBSound]

Thank you!