1
$\begingroup$

I am a researcher looking to start performing DFT calculations using VASP. I am planning to build a new supercomputer for VASP compilation and am trying to determine whether a CPU or GPU would offer better performance. After conducting various investigations, I am having difficulty finding a detailed analysis comparing the VASP DFT calculation performance of specific CPU and GPU products.

CPU: Intel® Xeon® Gold 6530 (32 cores, 2.1GHz, 160MB, 3UPI, 270W) (Total 64 cores) GPU: NVIDIA L40S GDDR6 48GB PCI-E4.0 x16

I would appreciate it if anyone could share insights on the VASP performance of these two products.

Thank you.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

I don't know for certain, but I expect that the performance of VASP on that GPU would be awful. VASP requires good 64-bit (FP64 or "double-precision") floating-point performance, but the L40S is almost entirely comprised of FP32 and lower-precision cores. The FP64 capabilities of the L40S are there simply to ensure that FP64 code will actually run, the performance is 1/64th of the 32-bit (FP32 or "single-precision") performance; see Figure 1 of:

https://images.nvidia.com/aem-dam/en-zz/Solutions/technologies/NVIDIA-ADA-GPU-PROVIZ-Architecture-Whitepaper_1.1.pdf

The stated FP32 performance is 91.6 TFLOPS, which means that the FP64 performance is about 1.43 TFLOPs. This is nowhere near any modern CPU or GPU; that is not to say that the L40S is a bad card in general, the performance for FP32 is very good, it just isn't designed for 64-bit maths.

Improve this answer
$\endgroup$

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.