Performance benchmark for custom workstations?

[simondkloss]

Hello,

I was wondering if there is something like a performance benchmark for custom build workstations. The issue is that I am currently trying to decide weather to run the workstation with 64Gb (2x32) or 128Gb (4x32) of DDR5 Ram. Using two ram slots would enable running the Ram at a higher pace of 5600 Mhz, while at 4x32 the Ram is limited to 4000 Mhz. Does VASP benefit more from higher clockspeeds or from more capacity?

My Setup:
Intel I9 13900K
4x32 Gb Kingston Fury 5600 Mhz
Gigabyte Z790 Aorus Elite Ax
Samsung Evo SSD

Run on Ubuntu 22.04. The VASP Build (v6.4.2) is exactly as described on the VASP Wiki page for Ubuntu 22.04.

BW
Simon

[manuel_engel1]

Hi Simon,

Unfortunately, we don't have any such benchmark data that we can share. Each setup is different and the maximum achievable performance always depends on all the components in your computer.

I think most importantly, the amount of RAM you require depends a lot on what kinds of calculations you are going to perform. Some VASP calculations require a lot of memory to run successfully. In general, I would therefore recommend to go with more RAM.

However, here, your choice appears to be between 2x32 and 4x32 of the same type of RAM. Usually, it's quite easy to upgrade or even switch RAM, especially in a desktop machine. In this case, I would suggest to start with 2x32. This will probably net you the highest performance. And when your applications require more RAM you can always upgrade later.

[simondkloss]

Hi Manuel,

Thanks for your reply. Would it be possible to clock the VASP test and use the time required to perform the "make test" as benchmark? Or do you think these tests are not representative in terms of a "real life" calculation?

BW
Simon

[manuel_engel1]

That is an interesting question. Honestly, I don't know if there is a correlation between the performance in the tests and the performance in actual calculations. The tests are somewhat designed to not put too much strain on your system because they should be simple and fast. As part of our Continuous Integration, we run the tests with 4 cores per test. For benchmarking, you may want to increase that number to try and saturate your CPU.

All in all, I would say that running the tests for benchmarking is probably a decent idea but I wouldn't read too much into the results.