BOINC: "Top GPU models" pages are junk

[zombie67]

This is hilarious!

It turns out that the BOINC code that creates the stats pages for the "Top GPU models" is "junk".

http://www.gpugrid.net/forum_thread.php?id=1150&nowrap=true#33084


The problem is the way the performance is calculated, the results are junk. If you used the results they would mislead you into believing some older cards were much better than they are.


So the Seti list is off by a factor of 9.8, an entire order of magnitude. Pure junk.

[DrPop]

Haha! Oh man...with history like Credit New under their belt, WHY am I not surprised? :P

[Al]

This doesn't fix the junk they put out, but this link (https://spreadsheets0.google.com/spreadsheet/pub?hl=en_US&key=0AmE73gddEiprdFBHMml3dmxqTjB4T21Mc3Mya3p1Snc&hl=en_US&gid=77) seems to be a little more real world.

[Slicker]

This doesn't fix the junk they put out, but this link (https://spreadsheets0.google.com/spreadsheet/pub?hl=en_US&key=0AmE73gddEiprdFBHMml3dmxqTjB4T21Mc3Mya3p1Snc&hl=en_US&gid=77) seems to be a little more real world.

Without knowing the overclock, the # of CPU cores allocated vs suggested by the project, and memory speed, the results can vary greatly. Using an app_config on collatz can increase credit by 50% for some GPUs but has no effect on others. Using an entire core per GPU task helps a lot on some projects and not so much on others. Since the drivers and/or BOINC doesn't track the specific results and which GPU completed them (as it can and will start a task on one gpu and finish on another in a machine with multiple GPUs) it is hard to tell from the server side who did what. But, all in all, I agree that the spreadsheet is closer to the real world.