Configuration recommendation for Opteron 6376

[nadams]

Hi all,

I'm looking to do some stress testing on my new HP BL465 servers, and was trying to figure out what the best configuration would be. I'm only going to have the servers available for about a week, but was hoping to contribute some to the project while they're burning in. I have three blades, each consisting of two Opteron 6376 processors (32 total cores per blade).

I loaded up the regular client, and have one set at CPU:29 and another at CPU:30. The CPU:29 client is only seeing around 29,000ppd... the CPU:30 is at around 32k PPD. Is this just because of the project that they're running, or an issue with the configuration? I expected them to be... faster.

[Nathan_P]

That's low, you do have a passkey don't you? Are the cpu's staying in turbo mode or are they dropping into power saving mode?

For reference a dual L5640 xeon setup 12c/24t @ 2.4 gets ~50k PPD on any core A3 project and 46-49k on a core A4 project

[nadams]

That's low, you do have a passkey don't you? Are the cpu's staying in turbo mode or are they dropping into power saving mode?

For reference a dual L5640 xeon setup 12c/24t @ 2.4 gets ~50k PPD on any core A3 project and 46-49k on a core A4 project

I do have a passkey, and it is configured. These are also the first two WUs for this setup, so I'll continue to watch it. It looks like I'm getting better performance at CPU:29 than CPU:32... would that be expected? Even though there's no GPU folding going on, does it need to have some free cores available? Oh, and I'm on an A3 project on both.

[Joe_H]

CPU:29 will be changed to a lower setting automatically since 29 is a "large" prime number. What it is changed to should show up in the log file at the beginning of processing a WU. As for which number will be used, 28 is a possibility, but 24 is also a high probability. Why you are seeing better PPD at 30 versus 29 is due to the fewer number of cores and also depends on what else is taking CPU time on the system and possibly which project the WU is coming from.

[nadams]

So what's the general recommendation for SMP? Should I stick to cpu:30 and above just based on PPD alone? I'm not going to dig too far into the optimizations, because I'm typically a 30k ppd contributor anyway... A single one of these blades matches my normal daily output... and I can only run them for a week.

[PantherX]

Welcome to the F@H Forum nadams,

In your case, a single CPU Slot containing the maximum number of CPUs you want to allocate to folding. However, please note that some numbers might be rounded down by the FahCore since that number could cause issues, e.g. 11 will automatically be rounded down to 10.

[nadams]

Thanks... for the purposes of my testing, I set all 3 clients to CPU:32.