i7's over estimating claimed credit

Join PrimeGrid

Read our rules and policies.
Download, install and run the BOINC software used by PrimeGrid.
When prompted, enter the URL:
http://www.primegrid.com/
Select which subprojects to run via the PrimeGrid Preferences.

If you are not using graphics enabled BOINC client, you may use manual registration form.

Author

Message

skildude
Send message
Joined: 18 Jul 08
Posts: 56
ID: 25792
Credit: 159,324,128
RAC: 360,726

I've been following my own progress on the 321 llr challenge and noticed that the i7 computers are all claiming about 2X what every other intel and AMD chip is claiming. I'm beginning to wonder if their is something wrong with the chips. My computer and others routinely claims 230-280 credit. the i7's typically claim 400+ and one actually claimed 800+. I'm not one to toss away free extra credit but it seems something is seriously wrong with those processors

ID: 22783 |

Michael Goetz

Volunteer moderator
Project administrator
Project scientist

Send message
Joined: 21 Jan 10
Posts: 13118
ID: 53948
Credit: 209,043,744
RAC: 119,147

I've been following my own progress on the 321 llr challenge and noticed that the i7 computers are all claiming about 2X what every other intel and AMD chip is claiming. I'm beginning to wonder if their is something wrong with the chips. My computer and others routinely claims 230-280 credit. the i7's typically claim 400+ and one actually claimed 800+. I'm not one to toss away free extra credit but it seems something is seriously wrong with those processors

I'm guessing you have hyperthreading enabled?

This is all pure speculation, but I'm guessing I know what's happening. The only thing that doesn't make any sense is that the older Pentium CPUs should also have this problem if they have HT.

When the BOINC benchmarks run, it runs the floating point benchmark on all the cores, and the integer benchmark on one core. No clue why it does that.

I'm guessing that all the work done at Primegrid uses integer math, rather than floating point. Assuming that's true, the only benchmark of interest is the integer benchmark, which is run on only one thread -- so there's no competing thread (running as a hyperthread on the same core) slowing it down.

When the work units are running, however, all the threads are in use, so they run somewhat slower (about 1/3 slower) due to the effects of hyperthreading. That makes the WU consume more CPU time, as if more work was being done. That, combined with the higher integer benchmark score that was not affected by HT, combines to cause a higher reported claimed credit

It's just a guess, and there's more than one assumption that may be incorrect in this theory.

____________
Please do not PM me with support questions. Ask on the forums instead. Thank you!

My lucky number is 75898⁵²⁴²⁸⁸+1

ID: 22785 |