Message boards :
Number crunching :
a question about work distribution
Message board moderation
Author | Message |
---|---|
Send message Joined: 2 Sep 04 Posts: 453 Credit: 193,569,815 RAC: 9,173 |
Just checked a Little bit around and now I'm wondering about this: Several of my boxes have prozessor-Features sse2 ssse3 pni This page shows that there are sixtrack-versions for sse2 sse3 pni. So how is decided what my Client gets to crunch ? Is always sent work for the highest or fastest prozessor-Feature or is the sse2 sse3 pni assigned to a workunit first and the Scheduler has to look to find a box that has this Feature ? Supporting BOINC, a great concept ! |
Send message Joined: 29 May 11 Posts: 10 Credit: 2,810,862 RAC: 0 |
I think it is a standard BOINC mechanism that over the course of time will try all variations of an app that can run on a device to try and figure out which is the most efficient code set. Once BOINC has figured out which runs best it will still occasionally send a different version to check and see if anything has changed from a performance perspective. -Steve |
Send message Joined: 12 Jul 11 Posts: 857 Credit: 1,619,050 RAC: 0 |
Actually it doesn't really matter too much There is a factor of two improvement using SSE (soon AVX I hope). SSE3/pni are the same and marginally better than SSE2. Eric. |
Send message Joined: 27 Aug 05 Posts: 2 Credit: 1,792,903 RAC: 0 |
I think it is a standard BOINC mechanism that over the course of time will try all variations of an app that can run on a device to try and figure out which is the most efficient code set. Once BOINC has figured out which runs best it will still occasionally send a different version to check and see if anything has changed from a performance perspective. Which does not seem to work at all as my machines keep being flooded with SSE2 since years. There does not seem to happen any kind of evaluation/conclusion on server side. Actually it doesn't really matter too much There is a factor of two Which may explain why the server schedulers don't show any conclusive preferences in distributing SSE2 vs SSE3/PNI. Is there any hope for some better optimizations? As much as I loved supporting LHC since the very beginning, it's not much fun seeing resources waisted (on top of stating it's not much fun seeing any good project using 'CreditNew'). |
Send message Joined: 12 Jul 11 Posts: 857 Credit: 1,619,050 RAC: 0 |
Sorry for delay; just saw your message. Eric. Just checked a Little bit around and now I'm wondering about this: In fact sse2/sse3/pni don't make much difference. I am wondering if we still have any boxes without sse2??? |
©2024 CERN