Xeon E5-2630 v1 low benchmark

[dbphillips]

So, I finally went and bought upgrades for my dual capable Dell T5600 and was doing some benchmarking to make sure I got what I expected and found the single CPU was running about 4900 for CPU Mark. I had HT off, as my apps don't use it, so turned it on and got about 6500. But the average is 10947, so I want to make sure I deal with any issues before adding more variables. Separate RAM (40GB - 16/16/4/4) and hard disk (striped 7200 drives) benchmarks look great. I'm running a Quadro 5000 and Dell H310 RAID.

I've checked BIOS and msconfig settings and can't seem to find anything to explain the discrepancy. Even the charts show the 2630 should be around 8900 (not sure why the chart numbers are lower than the app results).

Any ideas?

[Tim (PassMark)]

After you turned Hyper-threading back on was the "Number of processes" in the preferences set back to 12?
This should have displayed a warning message when PerformanceTest was started if the number of processes set didn't match the number available for the CPU.

Are you using the 64bit version of PerformanceTest?

[dbphillips]

Thanks for the quick response, Tim.

I'm pretty sure I installed the 64 bit version, but can't seem to verify. It's version 9, which stated 32 and 64 bit, and I try to use 64 bit whenever possible, so if the option was in the install, should be 64.

I let the app adjust the processes (but Task Manager already showed 12 logical processors) and got up to 8100, so about 10% off of the chart, but more than 25% below the app.

[Tim (PassMark)]

If you are on a 64bit version of Windows it should be running the 64bit version of PerformanceTest (though this can be confirmed by checking the About window form the Help menu).

There is an extensive list of issues that can cause lower than expected benchmark results in this post.

I believe the chart data in PerformanceTest and the charts displayed on the website use slightly different scripts, so it's possible the website script is more aggressive in removing some of the outlying results (or it is more likely only including results form V9 of PerformanceTest) from the calculation so we'll look into that.

[dbphillips]

Yeah, I'm not seeing anything there that is obvious. My build is about 6 months old, and obviously with 40gb, I'm running 64 bit. CPU temperature is fine.

Will the test run in safe mode? I understand tangential hijacking of the system by applications, but my primary goal is to verify I can run full speed and see how that compares to my new configuration. Starting down 25% is a little worrisome.

[dbphillips]

Okay, upgraded to E5-2670 and get about 11250, right about 25% slower than the average.

[dbphillips]

For some reason, the dual scores are much closer to average. Actually, a little above.

*Well, for the app values. This is low 13% from the charts.

[Tim (PassMark)]

I've checked how the average is being calculated in PT and the results displayed for the selected/same model in PerformanceTest is an average of all results in the database for that model, including systems with more than one CPU. So on Xeon systems with only one CPU the result will look a little slower and on multi CPU system will look closer to average.

[dbphillips]

Oh. What's the thinking behind that? It seems if there is a separate multi CPU chart the data would be separated.

[Tim (PassMark)]

Having more than one available result per CPU model complicates the chart generation and lookup considerably I believe and would then need a separate result for each possible combination. For the time being we are leaving it as it is but will make sure it's noted in the FAQ and help file.