Message boards :
Number crunching :
Using SSE2 path...
Message board moderation
Author | Message |
---|---|
Send message Joined: 19 Jul 10 Posts: 589 Credit: 18,926,725 RAC: 4,683 |
Any ideas on why is this computer using SSE2 instead of SSE3 for the GPU app? The CPU is SSE3 capable, runs the SSE3_AMD optimized SETI app and I'm using the current stable version of BOINC (6.12.34) and as I have seen on other hosts using that version, they are using SSE3, so should be new enough. |
Send message Joined: 8 May 10 Posts: 576 Credit: 15,979,383 RAC: 0 |
It was using some Windows function to check for SSE3 support but it doesn't work in Windows XP. The current GPU versions haven't been updated yet since that was fixed; the current CPU one should still detect SSE3 on XP. |
Send message Joined: 19 Jul 10 Posts: 589 Credit: 18,926,725 RAC: 4,683 |
OK, than it works as good as it can... thought it might be something wrong on my end. Thanks. |
Send message Joined: 17 Feb 08 Posts: 363 Credit: 258,227,990 RAC: 0 |
Any ideas on why is this computer using SSE2 instead of SSE3 for the GPU app? The CPU is SSE3 capable, runs the SSE3_AMD optimized SETI app and I'm using the current stable version of BOINC (6.12.34) and as I have seen on other hosts using that version, they are using SSE3, so should be new enough. I wouldn't bother much about that. There's no use for SSE3 instructions while calculating the "likelihood". (takes only a few sec...) And since you're running the app on a AMD CPU, it's highly likely that SSE3 code will be quite a bit slower after all. FWIW, i'm running a custom 0.91 app here on a AMD X2 5200+ that uses the SSE3 code path (ICC 12.x.x, avg 3.02 sec/likelihood/calc. )and it's slower than your AMD X2 5600+ using the stock app (2.4 sec.). ;) Join Support science! Joinc Team BOINC United now! |
Send message Joined: 19 Jul 10 Posts: 589 Credit: 18,926,725 RAC: 4,683 |
FWIW, i'm running a custom 0.91 app here on a AMD X2 5200+ that uses the SSE3 code path (ICC 12.x.x, avg 3.02 sec/likelihood/calc. )and it's slower than your AMD X2 5600+ using the stock app (2.4 sec.). ;) Hmm... than that's one of those cases, when it turns out to be better, when everything is not working too perfect ;-). Actually, once this is fixed in some future version and will take than the slower path, it could be nice to have a command line parameter to choose the path, at least for people like me, who like playing around with things like that. |
Send message Joined: 8 May 10 Posts: 576 Credit: 15,979,383 RAC: 0 |
Actually, once this is fixed in some future version and will take than the slower path, it could be nice to have a command line parameter to choose the path, at least for people like me, who like playing around with things like that.There is since it's needed for testing. --force-sse2, --force-sse3 etc. |
Send message Joined: 19 Jul 10 Posts: 589 Credit: 18,926,725 RAC: 4,683 |
There is since it's needed for testing. --force-sse2, --force-sse3 etc. Cool, must have missed that one in the list of command line parameters. |
Send message Joined: 14 Feb 09 Posts: 999 Credit: 74,932,619 RAC: 0 |
There is since it's needed for testing. --force-sse2, --force-sse3 etc. I don't think he listed it, I copied the command line parameters from his original post. |
©2024 Astroinformatics Group