Interesting benchmark difference between OS

Author	Message
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0	Message 63734 - Posted: 18 Oct 2009, 6:26:51 UTC Last modified: 18 Oct 2009, 6:27:22 UTC Win7: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid= Linux: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1167365 Same hardware. ID: 63734 · Rating: 0 · rate: / Reply Quote

dcdc Send message Joined: 3 Nov 05 Posts: 1833 Credit: 120,109,643 RAC: 14,483	Message 63737 - Posted: 18 Oct 2009, 11:11:17 UTC Last modified: 18 Oct 2009, 11:11:33 UTC the host ID is missing from the windows machine's link ;) ID: 63737 · Rating: 0 · rate: / Reply Quote

Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0	Message 63747 - Posted: 18 Oct 2009, 18:49:56 UTC - in response to Message 63737. the host ID is missing from the windows machine's link ;) https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1080649 well... i couldn't edit. so there's the WIn7 ID: 63747 · Rating: 0 · rate: / Reply Quote

DJStarfox Send message Joined: 19 Jul 07 Posts: 145 Credit: 1,250,162 RAC: 0	Message 63756 - Posted: 19 Oct 2009, 16:02:51 UTC - in response to Message 63734. It's tough to compare Linux vs Windows, even with the same hardware and same version of BOINC. This is because BOINC is compiled differently for each of those platforms. There will be minor variations in benchmarks. ID: 63756 · Rating: 0 · rate: / Reply Quote

Jaykay Send message Joined: 13 Nov 08 Posts: 29 Credit: 1,743,205 RAC: 0	Message 63757 - Posted: 19 Oct 2009, 16:51:40 UTC and benchmark results may vary even because of mouse movement... if you read some threads about benchmark result here you would know that. ID: 63757 · Rating: 0 · rate: / Reply Quote

mikey Send message Joined: 5 Jan 06 Posts: 1896 Credit: 10,138,586 RAC: 20,966	Message 63808 - Posted: 24 Oct 2009, 11:17:13 UTC - in response to Message 63747. the host ID is missing from the windows machine's link ;) https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1080649 well... i couldn't edit. so there's the WIn7 Making it a clickable link and saying that there are differences but not HUGE differences, a couple hundred here or there: Operating System Linux 2.6.28-15-generic Memory 3024.27 MB Cache 3072 KB Measured floating point speed 2269.92 million ops/sec Measured integer speed 6419.12 million ops/sec Operating System Microsoft Windows 7 x86 Edition, (06.01.7100.00) Memory 3071.24 MB Cache 488.28 KB Measured floating point speed 2470.59 million ops/sec Measured integer speed 5414.3 million ops/sec Yes the Win7 machine has a 20% faster integer speed but the floating point speed is very close, we are talking only about a 10% difference in floating point speeds. So I guess it comes down to which projects use which one to a better advantage. Meaning put Linux on those using integer speed as their basis for crunching and Win7 on those using floating point speed. I am hoping here that all you did was put one OS on, do the test and then put the other OS on the same machine and then run the test again? Or at least build 2 identical machines? Ram speed and motherboards may make up for some of the differences seen. ID: 63808 · Rating: 0 · rate: / Reply Quote

Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0	Message 63825 - Posted: 25 Oct 2009, 21:32:20 UTC - in response to Message 63808. the host ID is missing from the windows machine's link ;) https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1080649 well... i couldn't edit. so there's the WIn7 Making it a clickable link and saying that there are differences but not HUGE differences, a couple hundred here or there: Operating System Linux 2.6.28-15-generic Memory 3024.27 MB Cache 3072 KB Measured floating point speed 2269.92 million ops/sec Measured integer speed 6419.12 million ops/sec Operating System Microsoft Windows 7 x86 Edition, (06.01.7100.00) Memory 3071.24 MB Cache 488.28 KB Measured floating point speed 2470.59 million ops/sec Measured integer speed 5414.3 million ops/sec Yes the Win7 machine has a 20% faster integer speed but the floating point speed is very close, we are talking only about a 10% difference in floating point speeds. So I guess it comes down to which projects use which one to a better advantage. Meaning put Linux on those using integer speed as their basis for crunching and Win7 on those using floating point speed. I am hoping here that all you did was put one OS on, do the test and then put the other OS on the same machine and then run the test again? Or at least build 2 identical machines? Ram speed and motherboards may make up for some of the differences seen. same machine. dual boot. ID: 63825 · Rating: 0 · rate: / Reply Quote

P . P . L . Send message Joined: 20 Aug 06 Posts: 581 Credit: 4,865,274 RAC: 0	Message 63826 - Posted: 25 Oct 2009, 21:46:51 UTC Hi Chilean. Mine are different every time they run up or down, not to worry. ID: 63826 · Rating: 0 · rate: / Reply Quote

mikey Send message Joined: 5 Jan 06 Posts: 1896 Credit: 10,138,586 RAC: 20,966	Message 63839 - Posted: 26 Oct 2009, 9:04:35 UTC - in response to Message 63825. I am hoping here that all you did was put one OS on, do the test and then put the other OS on the same machine and then run the test again? Or at least build 2 identical machines? Ram speed and motherboards may make up for some of the differences seen. same machine. dual boot. Even better! ID: 63839 · Rating: 0 · rate: / Reply Quote

dumas777 Send message Joined: 19 Nov 05 Posts: 39 Credit: 2,762,081 RAC: 0	Message 63852 - Posted: 27 Oct 2009, 7:29:37 UTC - in response to Message 63734. Win7: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid= Linux: https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1167365 Same hardware. As already mentioned the BOINC benchmark test is not a great indicator of performance. A much better measure is how many credits per hour over a long period Rosetta awards for each OS on identical hardware. In my experience on my Mac Pro, Rosetta seems to give very close to the same as on windows through boot camp (didn't benchmark long enough to statistically measure but has to be within %10). Therefore Rosetta seems to be one of the better projects to run on a non Windows machine (along with the fact they are doing excellent science). On my Linux box it also runs at a credit rate that I would expect. The only drawback is of course the memory usage on the project rates towards the upper end so make sure you allocate at least 512meg a core. Btw wtf is it called minirosetta when there is nothing mini about its memory usage. ID: 63852 · Rating: 0 · rate: / Reply Quote