[cfeng]

I built a new computer specifically for OpenFoam.

I am using the new eight-core AMD FX 8350 processor, 16 gb GSkill ram, running at 90% of the L3 cache speed (fast!), and two hard drives in RAID1.

I was expecting this computer to be extremely quick, but it is running identical simulations about 10x slower than the local i7 workstations at school.

Could I have setup something incorrectly during installation of Ubuntu 12.04 or OpenFoam 2.20?

Any help would greatly appreciated, Thanks.

[wyldckat]

Greetings Charlie and welcome to the forum!

You should have done some more research on the Hardware forum: http://www.cfd-online.com/Forums/hardware/
There you should have found at least this thread: http://www.cfd-online.com/Forums/har...re-6-core.html

There are a few details to keep in mind:

1. How fast is your RAM? 1600MHz or 2166MHz?
2. How did you install OpenFOAM 2.2.0? Was it from source code or the from the Deb pack?
3. If I remember correctly, i7 processors have quad-channel DDR3 support, while the 8350 processors only support dual-channel DDR3. If you check the Hardware thread, you'll somewhere read about how much this can hurt performance...
   If you had purchased an AMD Opteron, then it could have quad-channel...

In conclusion:

• For your CPU, you should have the fastest RAM you can purchase... without going overboard, because from what I've read, more than 2166MHz is probably not worth the money... but I could be wrong.
• You should build OpenFOAM from source. Search for the GCC tuning options for the AMD processors.
• Also search for memory tuning options in the BIOS/UEFI, because of the memory searching patterns that can improve the behaviour with unstructured or structured meshes. Again search the Hardware forum.

Best regards,
Bruno

[cfeng]

wyldckat, thanks for your reply. I am using 1600 MHz ram overclocked to 1866 MHz running in dual channel. Also, OpenFoam was installed from the Deb Pack. I think the problem comes from running my case using all 8 cores in parallel, so the L2 cache is split between the two cores, causing slow memory reads.

A previous thread talked about optimizing mpirun for core affinity. Can't say I understand cache very well, but I'll compare with running cases split between 4 sets of 2 cores.