Message boards :
Number crunching :
nm_test 7
Message board moderation
Author | Message |
---|---|
Send message Joined: 30 Aug 07 Posts: 2046 Credit: 26,480 RAC: 0 |
how are the new WUs working (hopefully with app v0.2)? |
Send message Joined: 5 Apr 08 Posts: 11 Credit: 106,689,722 RAC: 0 |
On my Linux box (OpenSuse 11.0 64bit) the WU's are running well with application: 1. milkyway_0.2_x86_64-pc-linux-gnu and 2. 18% faster with a self compiled version from the tar (no changes in the source code, only a few compiler flags set.) It seems that the new (test) WU's need about 2 times more cpu-time to complete than the "old" WU's with milksops application. |
Send message Joined: 30 Aug 07 Posts: 2046 Credit: 26,480 RAC: 0 |
On my Linux box (OpenSuse 11.0 64bit) the WU's are running well with application: Well it's good that they're pretty close. After we get the bugs people have been seeing ironed hopefully we can get squeeze some more performance out of it and get it comparable to milksop's app. Also, are you comparing them with or without cuts in the parameter file (because each cut is an additional integral calculation)? |
Send message Joined: 5 Apr 08 Posts: 11 Credit: 106,689,722 RAC: 0 |
<with or without cuts in the parameter file> ?? I have nothing changed (sources nor parameter files). The only changes i have made were in the "make.linux" file (pathes and some flags), that i could compile (statically) on my machine. So, only you can decide if we are "comparing them with or without cuts". I cannot calculate what "type of cut" will influence the validation of a result. |
Send message Joined: 30 Mar 08 Posts: 25 Credit: 75,915,107 RAC: 0 |
Hello, I have tested the new application on windows 32 bits. It seems that the new (test) WU's need about 2 to 3 times more cpu-time to complete than the "old" WU's with milksops application. 13 min 50s instead of 5 min 35 s with a Q9300 intel core 21 min instead of 8 min on a T7250 core. Hope thes einfos will help to improve your code. Regards |
Send message Joined: 30 Aug 07 Posts: 2046 Credit: 26,480 RAC: 0 |
So you're just comparing the runtime of milksops app with WUs being generated by the server to the test app with WUs being generated from the server? |
Send message Joined: 30 Aug 07 Posts: 2046 Credit: 26,480 RAC: 0 |
Hello, This is good news -- the new WUs are doing a bit more crunching than the old ones -- they're calculating a similar sized integral, but they're also calculating an additional integral which is being removed from the first integral. The cut isn't as large as the first integral, so it seems the app is a bit slower than milksop's, but it seems it's not TOO bad. Hopefully we can get some more feedback for optimizations and make it even faster :) |
©2024 Astroinformatics Group