For quantum chemical calculations, we consider purchasing an AMD Ryzen Threadripper PRO 5995WX processor with 64 cores, which supports up to 2 TB of RAM. However, I would like to make optimal use of this processor for calculations using atomistic molecular dynamics with the Gromacs package.
Task: Execution of four computational tasks simultaneously in parallel on a computer with AMD Ryzen Threadripper PRO 5995WX using Gromacs.
(i) Case (1). If a computer has one AMD Ryzen Threadripper PRO 5995 WX processor, it has 64 cores with 4 GPU 4080 rtx installed.
ii) case (2). Alternatively, one can purchase four computers with an AMD Ryzen 7980 processor with 16 cores and install 1 4080 rtx on each computer.
If I run in (1) case, 4 computational tasks on 16 cores on 1 GPU, and in (2) case, the same tasks, one task at each station.
Will the calculation speed in both cases for one task be approximately the same, and therefore, instead of four computers, will one server with an AMD Ryzen Threadripper PRO 5995 WX with 4 GPU 4080 rtx yield the same calculation results?
this would be by preference- i have 4 system, and decent processor with GPU.
You arent going to run on 4 system together, run in one system and wait for sometime :)
1.if there is huge money get 64 core processor with a good GPU. << still it never going to be my choice>> with modern gpu, you dont need that much cores on cpu.
2. 4 system is better than overpowered sungle system with 4 gpu that has bottleneck at chipset.
For Gromacs calculations, I use an RTX 4080 graphics card and an AMD Ryzen 7980 processor with 32 GB of RAM. This setup allows me to perform calculations at a rate of almost 1 microsecond per day on systems consisting of 15.000 atoms. However, this computer is not sufficient for quantum chemistry calculations, mainly due to the limited RAM size.
I believe that having four separate computers with powerful GPUs would be more effective than using a single computer with four GPUs.
If your simulations can run update on the GPU, the performance difference between 4 GPUs in 1 machine or 4 is going to be small. If update runs on the CPU there could be a larger performance difference, depending on how fast the CPU-GPU communication is.