Benchmark Horror

Author	Message
stephan_t Send message Joined: 20 Oct 05 Posts: 129 Credit: 35,464 RAC: 0	Message 1779 - Posted: 26 Oct 2005, 8:54:34 UTC Checking my stats I realized one of my machines had it's RAC slashed by quite a bit. Going through the logs I spotted the benchmarks running right before the decrease in RAC. And indeed, the benchmark were reporting 633/1386 when before they were reporting 1377/2596. So I ran the benchmarks again and still received the dreadful 633/1386 results. Checked the task manager, only Rosetta was taking CPU time. Everything else was idle. Unconvinced, I restarted the box and ran the benchmark first thing. This time I got my 1377/2596 results back. So my RAC is back to normal. Does anyone has anymore information on that kind of behaviour? It might explain why some people have an incredible RAC even when they only run one mediocre machine while other have a low RAC running 4 boxes rosetta full-time. Team CFVault.com http://www.cfvault.com ID: 1779 · Rating: 0 · rate: / Reply Quote

[BAT] tutta55 Send message Joined: 16 Sep 05 Posts: 59 Credit: 99,832 RAC: 0	Message 1783 - Posted: 26 Oct 2005, 9:38:22 UTC This machine of your apparently is not a hyperthreading machine? If it is, it should mention 2 cpu's instead of one. Might this be the problem? In case the machine is a HT machine your benchmarks are reasonable for a 2.66GHz box. If it isn't, then they are way too low. One possible reason for your benchmarks decreasing is that your cpu, an Intel, throttles down due to overheating. BOINC.BE: For Belgians who love the smell of glowing red cpu's in the morning Tutta55's Lair ID: 1783 · Rating: 0 · rate: / Reply Quote

stephan_t Send message Joined: 20 Oct 05 Posts: 129 Credit: 35,464 RAC: 0	Message 1786 - Posted: 26 Oct 2005, 10:56:16 UTC No it's not HT. My 2 other P4s are HT. You are right, I suspect overheating - that's the only logical explanation. Team CFVault.com http://www.cfvault.com ID: 1786 · Rating: 0 · rate: / Reply Quote

stephan_t Send message Joined: 20 Oct 05 Posts: 129 Credit: 35,464 RAC: 0	Message 1792 - Posted: 26 Oct 2005, 13:13:48 UTC Actually, can anyone explain why this guy, with a single Athlon (AuthenticAMD AMD Athlon(tm) 64 Processor 3000+), gets 355.43 rac when I've been crunching 24/7 with 4 machines and only get 349.67 after 4 days? This is really confusing :( Team CFVault.com http://www.cfvault.com ID: 1792 · Rating: 0 · rate: / Reply Quote

Andrew Send message Joined: 19 Sep 05 Posts: 162 Credit: 105,512 RAC: 0	Message 1793 - Posted: 26 Oct 2005, 13:19:50 UTC Last modified: 26 Oct 2005, 13:21:37 UTC RAC doesn't really mean that much... only because RAC can be artifically inflated... eg. if a person merges a few computers, then their RAC sky rockets or if a project alters the stats manually like CPDN did a while back. ID: 1793 · Rating: 0 · rate: / Reply Quote

stephan_t Send message Joined: 20 Oct 05 Posts: 129 Credit: 35,464 RAC: 0	Message 1796 - Posted: 26 Oct 2005, 13:55:51 UTC I can also see a flaw with using the benchmark as part of the credit calculation - one could bench at full speed or even overclock, then clock down or disable Hyperthreading to max out the processing time. Team CFVault.com http://www.cfvault.com ID: 1796 · Rating: 0 · rate: / Reply Quote

Andrew Send message Joined: 19 Sep 05 Posts: 162 Credit: 105,512 RAC: 0	Message 1797 - Posted: 26 Oct 2005, 14:06:55 UTC Last modified: 26 Oct 2005, 14:26:25 UTC Well that be hard, because the boinc client runs the benchmark periodically, so you'd have to catch it everytime. However, someone could modify the code of the boinc client, to just generate a static number for the benchmarks. This is taken care of by most projects by sending out multiples of a WU and taking the average or the middle claimed credit value etc. With projects that just send out one WU and use the benchmark values then people could play silly buggers. This happened to Pirates, if I remember correctly. PS: Check out this for an alternative to the current benchmark system proposed by Paul "The Wiki-man" D Buck. PPS: Sorry Paul, just made that up. :P ID: 1797 · Rating: 0 · rate: / Reply Quote

Paul D. Buck Send message Joined: 17 Sep 05 Posts: 815 Credit: 1,812,737 RAC: 0	Message 1817 - Posted: 27 Oct 2005, 4:31:39 UTC - in response to Message 1797. Last modified: 27 Oct 2005, 4:38:51 UTC PS: Check out this for an alternative to the current benchmark system proposed by Paul "The Wiki-man" D Buck. PPS: Sorry Paul, just made that up. :P Andrew, It is good ... maybe I ought to add that to my signature? === edit You forgot the "." after the "D", the only mistake ... :) ID: 1817 · Rating: 0 · rate: / Reply Quote