HPC Hardware

indgepr01 · January 24, 2022, 10:47

We currently have a small (128 core) cluster that has been running quietly in the background for the lat 4 years and time has come to upgrade it, we are going to double the core count.

Due to the age of the old hardware we cannot duplicate it, so we will be using a different processor etc. will it be possible to run a single job on all All 256 cores or will we need to keep the jobs to the 2 sections of identical hardware ?

We will be running pretty much all the Ansys products

Thanks in advance

Phil

flotus1 · January 24, 2022, 11:01

You could theoretically do that.
Without doing anything about the heterogeneous hardware, the older (slower) cores will limit performance gains. You effectively get twice the performance of your old hardware, while paying for new hardware.
Next step is to account for this with load balancing. Basically, the goal is to assign smaller chunks of work to the slower hardware. To a point where both the old and new cores get to a synchronization point at the same time. Pretty sure you can do that with most Ansys products, it's just an additional and potentially cumbersome step in your workflow.

I guess it depends what exactly the existing hardware is whether this is worth it. You could potentially just double performance by getting 2 cluster nodes with 128 of the latest and greatest CPU cores. Not having to buy additional Ansys licenses for doubling core count could easily justify the hardware cost.

January 24, 2022, 10:47	HPC Hardware	#1
indgepr01 New Member Join Date: Jan 2022 Posts: 1 Rep Power: 0	We currently have a small (128 core) cluster that has been running quietly in the background for the lat 4 years and time has come to upgrade it, we are going to double the core count. Due to the age of the old hardware we cannot duplicate it, so we will be using a different processor etc. will it be possible to run a single job on all All 256 cores or will we need to keep the jobs to the 2 sections of identical hardware ? We will be running pretty much all the Ansys products Thanks in advance Phil

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
General recommendations for CFD hardware [WIP]	flotus1	Hardware	19	February 29, 2024 12:48
AMD Epyc hardware for ANSYS HPC	chefbouza74	Hardware	4	October 26, 2021 11:51
CFD Hardware setup by small FEA benchmark. Would it make sense?	Twenty	Hardware	2	May 12, 2021 10:03
Can we merge HPC Pack licenses?	Phillamon	FLUENT	0	January 24, 2014 02:59
Need help with hardware specs for a Fluent HPC	John	FLUENT	0	September 25, 2008 10:24

January 24, 2022, 11:01		#2
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	You could theoretically do that. Without doing anything about the heterogeneous hardware, the older (slower) cores will limit performance gains. You effectively get twice the performance of your old hardware, while paying for new hardware. Next step is to account for this with load balancing. Basically, the goal is to assign smaller chunks of work to the slower hardware. To a point where both the old and new cores get to a synchronization point at the same time. Pretty sure you can do that with most Ansys products, it's just an additional and potentially cumbersome step in your workflow. I guess it depends what exactly the existing hardware is whether this is worth it. You could potentially just double performance by getting 2 cluster nodes with 128 of the latest and greatest CPU cores. Not having to buy additional Ansys licenses for doubling core count could easily justify the hardware cost.