What's the best hardware - more cores on CPU or more CPUs?

[drkirkby]

I'm soon to take delivery of a Dell 7920 workstation

https://www.dell.com/en-uk/work/shop/wo ... pt7920emea

I purchased used from eBay. The Dell 7920 is a very configurable workstation, with a base price of around £2600 (GBP) from Dell, but rising to £100,000 (GB) if you stuff it full of RAM (3 TB of the fastest RAM is £43,000) and CPU (a pair of them is around £15,000), disks etc. What I have bought is much more modest, with a single 8-core Xeon Silver 4110 CPU

Intel website
https://ark.intel.com/content/www/us/en ... 0-ghz.html

# of Cores 8
# of Threads 16
Processor Base Frequency 2.10 GHz
Max Turbo Frequency 3.00 GHz
Cache 11 MB L3 Cache
# of UPI Links 2
TDP 85 W

Passmark website.
https://www.cpubenchmark.net/cpu.php?cp ... Hz&id=3106

and 16 GB RAM. But I intend upgrading both the RAM and CPUs.

I'm wondering if a pair of 10-core CPUs would be better/worst for openEMS than a single 20-core CPU. I realise it is one of those "it depends" type questions. I'm tempted to get as many cores on a single CPU as possible, as it can be upgraded, whereas one can't upgrade the CPUs without swapping them out if one uses two CPUs with less cores.

I will look for used CPUs and RAM. The RAM for this machine is pretty expensive, as its DDS4 ECC RAM. Error correcting RAM always costs more than the non-error-correcting variety.

Would a NVINIA Tesla K80 be supported?

https://www.nvidia.com/en-gb/data-center/tesla-k80/


4992 NVIDIA CUDA cores with a dual-GPU design
Up to 2.91 teraflops double-precision performance with NVIDIA GPU Boost
Up to 8.73 teraflops single-precision performance with NVIDIA GPU Boost
24 GB of GDDR5 memory
480 GB/s aggregate memory bandwidth
ECC protection for increased reliability
Server-optimised to deliver the best throughput in the data center

Adding one of those is a possible option, as its the only Linux supported GPU on a bit of commercial software I want to run. But I suspect many of the other CUDA GPUs would work. (Loads work with the Windows version - I rather suspect that less have been tested on Linux.)

Dave

[thorsten]

Hi,

openEMS does not use the GPU and it does not scale very well with the amount of CPU cores. Usually you get the best speed using only 2 or 4 threads.
The CPU cache size and CPU to main memory bandwidth/speed is much more important I think.

I hope this helps.
br
Thorsten

[drkirkby]

Hi,

openEMS does not use the GPU and it does not scale very well with the amount of CPU cores. Usually you get the best speed using only 2 or 4 threads.
The CPU cache size and CPU to main memory bandwidth/speed is much more important I think.

I hope this helps.
br
Thorsten

Thank you. It does help. I need to sort the RAM out. Currently there’s one 32 GB RDIMM in there, but I have a 64 GB LRDIMM on order. But I recently realised that I will get best performance with 4 modules. Unfortunately they are quite expensive. 4x64 GB = 256 RAM modules will cost more than I paid for the complete machine.