Message boards : Number crunching : 40 points +/- on the same machine?
Previous · 1 · 2
Author | Message |
---|---|
![]() Send message Joined: 1 Apr 06 Posts: 26 Credit: 176,432 RAC: 0 |
It is time to get the garments ready: Optimized clients are NOT available for every OS/CPU/... therefore they are NOT the solution. Of course, you want the inflated numbers they generate and care less about the actual results they produce. Old credit system = grant based upon benchmarks (some of which were falsified) New credit system = grant based upon actual work produced (not hours ran, Stevea ... output) Was the old system fair - no Is the new system fair - no (but it's a lot closer than the old system) |
Mats Petersson Send message Joined: 29 Sep 05 Posts: 225 Credit: 951,788 RAC: 0 |
It may very well be annoying, irritating, frustrating and confusing that the results vary up and down as much as they do. However, I took all of the results available at the moment for stevea's machines [just under 80 of them], and run it through a spreadsheet. The total score according to the old credit system is 8337.37, with the new 8749.01 - around 5% higher than the old system. The variance, however, is about 10x the old system - 6.08/60.88. Average score per workunit is 106.9 vs 112.17. I'm not sure if this was a meaningfull excercise (I probably should also split it per machine to see if any particular machine is much worse than any other machine). But I guess if stevea would rather have stable but 5% lower scores, the old system would be better... I prefer a bigger variation and higher score myself [partially because several of my machines are Linux-based, where the benchmark score is terribly low but the credit for Linux is actually on par with Windows - there's a factor of 2.3 difference between the claimed and granted credit because the Windows systems are "pulling it up"]. -- Mats |
Message boards :
Number crunching :
40 points +/- on the same machine?
©2025 University of Washington
https://www.bakerlab.org