Dear All,
I'm not pretty sure whether it's a good place to ask question but I don't find more appropriate one.
So, I'm going to build a small cluster based on Intel i9. According to funding I have I have 2 options: 2 x i9 (10 cores) + GPU Nvidia RTX 3080 or 3 x i9 and no GPU. What is your suggestion? Which cluster will perform better?
cheers
wojtek
To GPU or not to GPU
Moderators: Global Moderator, Moderator
-
- Global Moderator
- Posts: 474
- Joined: Mon Nov 04, 2019 12:41 pm
- Contact:
Re: To GPU or not to GPU
This question depends on many factors such as the size and type of the calculations that you want to do.
Not all parts of the VASP code are ready to run on the GPU.
Have a look at the information on this page and let us know if there is some information missing:
wiki/index.php/OpenACC_GPU_port_of_VASP
We, the VASP team, don't give this type of hardware recommendation.
Perhaps some users can share their personal experiences helping you make a decision.
I will move this question to the "From users to users forum".
Not all parts of the VASP code are ready to run on the GPU.
Have a look at the information on this page and let us know if there is some information missing:
wiki/index.php/OpenACC_GPU_port_of_VASP
We, the VASP team, don't give this type of hardware recommendation.
Perhaps some users can share their personal experiences helping you make a decision.
I will move this question to the "From users to users forum".
-
- Global Moderator
- Posts: 474
- Joined: Mon Nov 04, 2019 12:41 pm
- Contact:
Re: To GPU or not to GPU
I would like to bring your attention to a note on this wiki page referring to the recommended type of hardware:
Since the graphics card, you mention in your post seems to fall in the category of "gaming" hardware it is perhaps not the best choice if you intend to use it for double-precision floating-point computations (e.g. with VASP).Running VASP on other NVIDIA GPUs (e.g. "gaming" hardware) is technically possible but not advisable: these GPUs are not well suited since they do not offer fast double-precision floating-point arithmetic (FP64) performance and in general have smaller memories without error correction code (ECC) capabilities.