Can RTX 2060 Super and Quadro P2000/4000 co-exist on same m-board?

Can RTX 2060 Super and Quadro P2000/4000 co-exist on same m-board?

Assuming that PS can handle it, and both cards sit in their own PCIe X 16 sloths.
I now have three GPU renderers. Arnold GPU, Redshift GPU, Octane GPU, and DX11/12 engines.
Would I need to have both RTX 20XX drivers and Quadro PASCAL drivers installed? I am thinking yes, but concerned about possible damage to either GPU and mainboard components?
OS is Win 10 Pro 64. CPU is XEON E5-1640V4, ram 64GB (16GB X4 DDR4, ECC,registered, 2666Mhz @ 2400). Mainboard is HP Z440.
Thank you

As long as you have PCIe slots available and a sufficiently sized power supply, you can install any number of GPUs you wish. I seem to recall people reporting 8+ GPUs in one system before in these forums.

Anytime you install PCIe cards yourself there is some risk that you could damage something (mechanically or electrically), but in practice the risks tend to be minimal. Make sure the system is turned off and unplugged from the wall outlet while you are working on it, don’t use brute force when handling any mechanical components (such as little plastic tabs designed to snap into place), and ensure all connectors are inserted properly aligned, straight, and completely. Take precautions against electrostatic discharge.

NVIDIA has unified drivers that work across all generations of supported GPUs. There is some cut-off for outdated hardware, but current drivers support everything from the Kepler architecture to the latest Turing architecture.

Thank you.

However when I did driver search on NVIDIA, I got two different packages. Slightly different version numbers as RTX 2060S is for RTX game boards and Quadrop PASCAL driver is a different package. So lets say both are installed. In NVIDIA Control panel, are both drivers unified? In there is only control panel that lists both boards. Would two drivers work with each other or overwrite the other during installation ? Boards are not SLI or anything like that. The RTX would have the panels plugged into it as it is the faster board. THe Quadro Pascal would be a OpenCL , CUDA, and GPU renderer slave , and its display ports and HDMI ports would remain unused. Or the reverse. The Quadro PASCAL is the panel display board and RTX 2060X is the CUDA and OpenCL slave.
The second consideration is using cryptomining boards as GPU render slaves. I do not cryptomine and dislike practice, but awfull lot of these GTX 1660 and 1660Ti port-less boards are available at reasonable prices. They could be used as inexpensive GPU render slaves. As long as one GPU on system has Display Port or HDMI to drive a panel.

You never have 2 NVIDIA drivers installed. Ever. On any OS. Under any circumstance. It is not possible.

NVIDIA drivers are architecturally unified. This doesn’t meany any driver installer package will work with any GPU, but it does architecturally support the idea that you can only have 1 driver, properly installed, at any given point in time. The way to support multiple GPUs is having that one driver do the task.

If you’re looking for a package that demonstrably supports dissimilar GPUs, I usually recommend the drivers that are bundled with the CUDA installer packages. Those drivers are designed to work on a wide range of GPUs (essentially all GPUs extant at that time).

With a few limited exceptions, NVIDIA generally doesn’t recommend the use of dissimilar GPUs in the system, and functionality is not guaranteed. For example GPU P2P will not work between those 2 GPUs. You’re welcome to do as you wish, of course, and I fully expect that you would not “damage” anything, but of course I can’t and don’t offer guarantees over the internet. I regularly run rigs with multiple dissimilar GPUs. In my view it works fine. But that’s not a statement of support, guarantee/warranty, etc. It’s merely an observation. If you’re concerned that you may damage something in your PC, don’t open the case.

At this year’s GTC I saw a server box with sixteen cards in it. It was made by One Stop Systems:

Interesting. If I recall correctly, the maximum ever reported in these forums over the past ten years was around a dozen? I am not sure for what kind of use case this 16 GPU system would be well suited.

It seems it might bottleneck on a lack of PCIe bandwidth (we would really want PCIe gen3 x16 per GPU), and have an insufficient number of CPU cores for general-purpose accelerated computing (a typical balance is 4-5 physical CPU cores per GPU). And since a well-balanced system typically calls for system memory in the amount of 2x-4x (I would strive towards the higher bound) the total GPU memory, the system memory may be undersized.

Thank you all. THe reason that it came up is that I want to run both RTX 2060 Super FE and Quadro P4000 in same system. Why do I want to do that? Well I already have the boards. Each one sits in its own system. I am experiencing some OpenCL compatibility issues in some applications. My tool set is core around three packages. Maya '19, Houdini 17.5, and Realflow 10 & 20. Specifically the use of OPenCL and CUDA for GPU running of FLIP and FLuid-PDB dynamics. OpenGL performance is, of course, fantastic for both my RTX and Quadro boards. My secondary hobby need is high DX11/DX 12 game performance for flight simulation software. DCS, X-Plane, and Prepare3D. RTX 2060S is completely superior to Quadro P4000 in flightsims. My average FPS at 2560X1440 @144Hz is above 120 and gets up to 190’s. Easily doubling the FPS under Quadro P4000.
Where I am having issues is OpenCL usage. Maya 2019 OpenCL and ArnoldGPU(Beta) is very unstable under RTX 2060S. ArnoldGPU crashes every time in GPU only mode, and drops to CPU mode, or crashes Maya. As it is Beta, I am not that concerned about it just yet. Maya’s dynamics OpenCL usage, when it available, works OK in both RTX and Quadro Pascal, but a lot more stable and available under Quadro Pascal. Houdini is same way, but I am not certain exactly how Houdini’s dynamics engine uses OpenCL. I enable it in solver option, and it runs. I notice that GPU and CPU are loaded at same time. But it is inconsistent, as some OpenCL sims load GPU , while others favor CPU. I am using production 17.5.360 build. Realflow OpenCL is a real problem. It works under Quadro accelerating FLIP, and CUDA works great accelerating Realflow’s PDB solvers. But RTX’s OpenCL is unavailable in RF10 and new RF20. Option is greyed out under FLIPs, but CUDA is available, but not as fast (in RF) as Quadro. Both RTX 2060S and Q P4000 have 8gb. So I want to have both of them running at same time on same system, instead of having two systems.