I have two NVIDIA RTX A4500 on my board (bought it as part of Lenovo Thinkstation PX) and I’m trying to boost my GPU performance. I’ve a few questions and I’d be looking for your help with them.
I can add two other RTX A4500 GPU cards since my board allows up to 4 cards. Which card models would be more efficient and would integrate well with my current GPU cards?
How about these ones: RTX A5000 GPU? RTX A6000 GPU? RTX 5000 ADA 32GB?
Would these work with the two GPU cards I already have (and I’m already using NVLink)?
Hello @m_farrag and welcome to the NVIDIA developer forums.
In principle any workstation GPUs will work together nicely as long as they are supported by the same driver. It all depends on how the workloads are distributed.
But in general you should consult your system integrator first regarding compatibility. There are many things to consider, including but not exclusive:
- adequate power supply
- proper cooling
- implications to PCIe speeds (2 GPUs might run with 2x PCIe x16 but four will likely run at x8 or even slower)
- NVLINK bridges only support two GPUs at a time on workstation platforms and is discontinued in Ada generation workstation GPUs (which also means RTX 5000 Ada would not run with NVLink)
- what kind of workloads are you planning to run, do they even support multi-GPU setups?
So I cannot give any recommendation here, I am sorry.
to add what Markus says (since I just answered similar in a different thread… ;-) :
- we DON’T recommend to mix GPUs of different generations - this may cause issues with compatibility, hence we often need to advertise the largest COMMON denominator of features, making some of the exclusive benefits of the lastest generation unused…
- multiple GPUs NEVER act seamlessly as a single bigger/larger GPU! For single threaded rasterized rendering tasks, which are limited to a single CPU core, they can only run on a single GPU! Lots of SW tricks needed to benefit from a 2nd GPU, basically in putting other (side-)jobs on that GPU…
Parallel jobs, like raytracing NICELY benefit from a 2nd+ GPU, almost perfect scaling.
Parallel compute jobs, it depends, often need to be explicitly written to benefit from multiple GPUs. Check for each workload/app, if it supports multi-GPU…
Even with multiple (similar) jobs, WinOS does NOT automagically distribute them across all GPUs equally, so often explict GPUaffinity needs to be programmed in… :-(.
so, overall, DON’T mix, and if you see ZERO benefit, or have a chance for some significant perf improvement depends 100% on your workload - so pls provide details about what you are trying to run on multiple GPUs…
1 Like