Message boards :
Number crunching :
Let the Optimisations begin.... :).
Message board moderation
Author | Message |
---|---|
Gary Roberts Send message Joined: 31 Oct 99 Posts: 95 Credit: 2,301,228 RAC: 0 |
I just saw a note over on the boinc_dev list from Eric Korpela, which said:- I'll modify the tarball maker to also build a setiathome_enhanced tarball. Presumably Tetsuji can do his maniacal thing and we can get back to some slightly quicker crunch times for the beta Test Project... Anyone seen the Maverick?? :). |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
First off, I've not looked at this that closely, but I think it's basically Berkeley playing catchup with what we've already been doing (unofficially). My understanding is that berkeley are introducing the FFTW3 code into the official client (or at least into a beta "enhanced" client). This is basically what Maverick, Metod , myself and others have been doing from Eric Heien's original FFTW3 source patches. Maverick has laterly gone on to use Intel's IPP libs in place of FFTW3, but it is very unlikely that Berkeley will adopt these as they're closed source. But they appear slightly faster on some Intel setups. Obviously the important thing here is that now we've shown the benefits of using FFTW3, this will filter through the development cycle into the official client so everyone will benefit, not just those testing experimental clients :) I guess phase 1 is drawing to an end (us testing our own experimental clients), and now phase 2 is about to start in the form of an official beta "enhanced" client from Berkeley. Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
The Beta client (setiathome_enhanced) is not nearly as fast as Tesuji's clients. In fact it's even slower on my machines than the original 4.09 client ever was. Aloha, Uli |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
The Beta client (setiathome_enhanced) is not nearly as fast as Tesuji's clients. In fact it's even slower on my machines than the original 4.09 client ever was. The "Enhanced" Beta version is slower than the original on my puter as well. |
TeraGram Send message Joined: 20 Jul 99 Posts: 187 Credit: 7,784,928 RAC: 0 |
But they appear slightly faster on some Intel setups. Are you just being understated? "Slightly"? Maverick's optimizations have cut the time for my PII MMZ laptop phenomenally! Work ID Sent Results 18184042 19 Jun 2005 18:52:09 UTC 20 Jun 2005 11:05:05 UTC opt 18071793 18 Jun 2005 12:10:56 UTC 19 Jun 2005 17:37:27 UTC no opt And my big machine, a 3.00 GHz P4 MT SSE2: 18191838 19 Jun 2005 20:59:26 UTC 19 Jun 2005 23:59:47 UTC opt 18132392 19 Jun 2005 4:35:47 UTC 19 Jun 2005 11:12:44 UTC no opt |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
But they appear slightly faster on some Intel setups. Sorry - I meant relative to the optimized FFTW3 clients, not the original Berkeley clients!!! ie, IPP is slightly faster than FFTW3 for single precision transforms on (some) Intel hardware :) Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
The Beta client (setiathome_enhanced) is not nearly as fast as Tesuji's clients. In fact it's even slower on my machines than the original 4.09 client ever was. Is this compiled from source, or a downloaded binary? Do you have the FFTW3 libs installed? They may not have staticly linked them?? Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Is this compiled from source, or a downloaded binary? I refered to the downloaded official 4.00 setiathome_enhanced for SETI Beta in comparison to Tetsuji's setiathome_4.11 for the normal SETI. [edit] All on Windoze. [/edit] Aloha, Uli |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
I refered to the downloaded official 4.00 setiathome_enhanced for SETI Beta in comparison to Tetsuji's setiathome_4.11 for the normal SETI. [edit] All on Windoze. [/edit] Don't forget that the 'Enhanced Client' includes extra processing to gain another 1.5dB sensitivity for detecting signals. Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Crunch3r Send message Joined: 15 Apr 99 Posts: 1546 Credit: 3,438,823 RAC: 0 |
First off, I've not looked at this that closely, but I think it's basically Berkeley playing catchup with what we've already been doing (unofficially). I think thats not a good idea, because my new client for alpha cpus based fftw3 patched source is doing worse with fftw3. Maybe on i386 cpus it´s working but not on alpha cpus. It increased processing time amlost twice !!! So i think there should be a standart port and a fftw port of seti core client.
Join BOINC United now! |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
[quote] I think thats not a good idea, because my new client for alpha cpus based fftw3 patched source is doing worse with fftw3. Maybe on i386 cpus it´s working but not on alpha cpus. It increased processing time amlost twice !!! So i think there should be a standart port and a fftw port of seti core client. [quote] I believe the config will check for the presence of FFTW3 and use it if it's present. If not, it should use the standard maths libs as before. *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.