Message boards :
Number crunching :
Lunatics Windows Installer v0.41 Release Notes
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 17 · Next
Author | Message |
---|---|
Karsten Vinding Send message Joined: 18 May 99 Posts: 239 Credit: 25,201,931 RAC: 11 |
I was just a bit tweaked when someone simply waved off SSE3 when we (AMD K10 users) had an excellent one once. Seeing how many of the super crunchers who helped test during the V7 beta are overwhelmingly Intel/Cuda, I just wanted to stand up and remind everyone that not all crunchers are Intel/Cuda. It was partially me that waved off the SSE3 app. I'm one of the Alfa testers on Lunatics, and though I havent been very active on their forums this time (many personal things have taken precedence), I have tested the various versions. My rigs are all AMD, and my second cruncher is a Phenom II. On this machine the difference between SSE2 and SSE3 versions of the new app was small, SSE2 sometimes fastest, SSE3 other times. This could probably change. Therefore I see no problem with the team concentrating on a few major versions to get it out fast. There are also no SSE4 versions, because the gains were small compared to SSSE3, and it was prioritized to get the installer out fast. You must understand they where under some time pressure, because of the short windows between release on main, and promised release on Lunatics. I havent tested the final version (of SSE3) yet, so cant tell how big the difference is, but I guess it will be small. But of course every little bit counts :) Regarding the old AKV8B_SSE3_AMD, I was actually one of the testers (as I remember we were only 2 or 3 testers), who in cooperation with Jason Gee tested an app that turned out to be faster on all the CPU's I had at the time. If I remember correctly (Jason can correct me if I have the facts wrong), the app was mainly faster because it left out a (IPP?) library, that handicaped AMD, and used another instead. The actual amount of SSE3 code was not very great it was basically the SSE3 app tweaked a little, with some parts replaced (some of it by SSE2 code). The app did require SSE3 to run, it didn't use it all that much. |
Link Send message Joined: 18 Sep 03 Posts: 834 Credit: 1,807,369 RAC: 0 |
William wrote: It's Jason's build using statically linked libraries - those are traditionally faster than the dynamically linked ones (i.e. Raistmer's r555 and r1797) In other words: use the old one if you have it? |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Raistmers AP apps are faster on my machine. I did use r557 first also and it seemed faster but after about 10 completed tasks 555 was faster. And r_1917 is based on the same code. But on non AVX hosts pre win 7 SP1 r_1797 shoud be faster. With each crime and every kindness we birth our future. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
Do all be aware that the new r1846 applications for MB v7 on CPU go through a process called "wisdom generation" - the first couple of times you run them, they spend quite a bit of time testing what works best on your particular hardware. After that, tasks should run a lot quicker - so wait before making final comparisons with earlier, or different, versions. |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
AKV8B_SSE3_AMD Is there a new app released yet. I installed the installer 41 and the sse2 cashed this machine while running x41zc for multibeam. It was running fine but the nvidia driver 320.18 failed this BSOD my rig i am running now x41zc cuda 42 in which I have identified as cuda_fermi and not cuda42 ver 7 running well. I am still using the old AKV8B_SSE3_AMD app to run 603 and I am still getting them but I do not know how much longer they will send. need an sse3 app for cpu multibeam Amd Phenom 2 1100T @ 4.00 ghz win 7 sp1 x64 Nvidia 460 running @ 850 core Michael Miles |
atlov Send message Joined: 11 Aug 12 Posts: 35 Credit: 32,718,664 RAC: 34 |
Is BOINC able to crunch GPU tasks assinged to different cuda versions at the same time? Let's say I'm using <count>0.5</count> for all apps. PS: Thanks to the Luncatics team! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
Is BOINC able to crunch GPU tasks assinged to different cuda versions at the same time? Let's say I'm using <count>0.5</count> for all apps. Not with the installer. You will just have one CUDA program installed - the version you chose at installation. But - if you had SETI tasks already downloaded from before your installation, they might be marked ('branded', to revive a term we used to use) for a variety of versions. They will all be processed OK by the version you installed, and yes, tasks with different 'brandings' can be crunched at the same time. And if you do work for other projects that use different CUDA levels, they can be crunched at the same time too (with appropriate <count> values). I regularly run Einstein's cuda32 tasks on the same card and at the same time as SETI's cuda50 tasks. |
atlov Send message Joined: 11 Aug 12 Posts: 35 Credit: 32,718,664 RAC: 34 |
IC. I also noted the installer chose the cuda50 executeable for all cuda versions in app_info ... that explains everything. Now back to the sandbag barriers at the river ... |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
William wrote:It's Jason's build using statically linked libraries - those are traditionally faster than the dynamically linked ones (i.e. Raistmer's r555 and r1797) As I said, I've no idea what sort of improvements Raistmer (and Urs) have been doing. You might want to run a speed comparison too - as Mike found on his Phenom the r555 was faster than r557. YMMV... A person who won't read has no advantage over one who can't read. (Mark Twain) |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
AKV8B_SSE3_AMD Ok, once again the AMD saga - this time from the peanut gallery. [you all realise that at going on 3 years at this I'm not really the person to write either 'a short history of Lunatics' or 'a short history of optimised Seti' ?] Let's just roll back to v0.40 then. CPU apps in there were AKv8b2 (note the 2, minor bug update). They were build by Jason with Intel compilers - ICC/IPP - Intel proprietary code that is targetted specifically at Intel CPUs, thereby achieving a nice speed improvement through all SIDs. There was a specific AMD app, since the other apps had a tendency to crash on AMD. Note 'Intel proprietary code' in conjuntion with 'Open Source/GPL'... which led to withdrawal of the installers. So this time round, Joe build with MinGW/GCC, to avoid that problem. The downside is of course that they no longer use Intel's optimisation, so are slower than they used to be - they are still faster than stock though, being based on the original AK, and the further optimisations that happened. Since we don't use Intel, we shouldn't really have a problem with AMD... @ Michael Miles - what crashed the NV driver or the CPU app? A person who won't read has no advantage over one who can't read. (Mark Twain) |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
IC. I also noted the installer chose the cuda50 executeable for all cuda versions in app_info ... that explains everything. I specifically designed the aistubs, so that they would be showing ther plan_class of the app that is actually running. If your new work shows 'cuda50' then you've installed the 5.0 flavour. Now back to the sandbag barriers at the river ... Good luck! A person who won't read has no advantage over one who can't read. (Mark Twain) |
__W__ Send message Joined: 28 Mar 09 Posts: 116 Credit: 5,943,642 RAC: 0 |
I did use r557 first also and it seemed faster but after about 10 completed tasks 555 was faster. That would be nice (running XP Pro SP3), but i can't go through a complet AP-WU with r1917. My frist try some weeks ago ended in a stop at ~50-60%. After a ~2 hour wait i switchted back to r557. Second try yesterday with the 0.41 installer ends with same result. Can there be a significant pause of processing a WU with the r1917 specially during the first 10 WUs ? Maybe i have to wait longer. __W__ _______________________________________________________________________________ |
__W__ Send message Joined: 28 Mar 09 Posts: 116 Credit: 5,943,642 RAC: 0 |
It will fix itself after a short number of tasks. In my case it only happened to the very first task run with the new app.Thank you for this tip. Will give it a try tomorrow when the just runnig AP-WUs have finished. __W__ _______________________________________________________________________________ |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Its possible there is a problem on your host to create correct wisdom file. Make sure no AV proggy is interfering with the Boinc data folder. Check your HD for E/A errors. With each crime and every kindness we birth our future. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
|
__W__ Send message Joined: 28 Mar 09 Posts: 116 Credit: 5,943,642 RAC: 0 |
Its possible there is a problem on your host to create correct wisdom file.HD is rockstable, all Boinc parts are excluded in AV and a wisdom file with some entries was created - correct or not, i don't know. __W__ _______________________________________________________________________________ |
__W__ Send message Joined: 28 Mar 09 Posts: 116 Credit: 5,943,642 RAC: 0 |
Multicore host? Core2Duo T7600 2.33 MHz __W__ _______________________________________________________________________________ |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Multicore host? That is my second guess. I know you have a dual core. Free one core when the next ap unit starts. Maybe other tasks interfering whilst wisdom will be created. With each crime and every kindness we birth our future. |
__W__ Send message Joined: 28 Mar 09 Posts: 116 Credit: 5,943,642 RAC: 0 |
Free one core when the next ap unit starts.EST ~11 hours left ;-) __W__ _______________________________________________________________________________ |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
AKV8B_SSE3_AMD Is there a new app released yet. I installed the installer 41 and the sse2 cashed this machine while running x41zc for multibeam. It was running fine but the nvidia driver 320.18 failed this BSOD my rig i am running now x41zc cuda 42 in which I have identified as cuda_fermi and not cuda42 ver 7 running well. I am still using the old AKV8B_SSE3_AMD app to run 603 and I am still getting them but I do not know how much longer they will send. need an sse3 app for cpu multibeam Amd Phenom 2 1100T @ 4.00 ghz win 7 sp1 x64 Nvidia 460 running @ 850 core @ Michael Miles - what crashed the NV driver or the CPU app? William I am not sure what crashed the machine but just before it did my clock speed downclocked on the 460 to 50 mhz then it just BSOD this was the cpu app that i used AKv8c_Bb_r1846_winx86_SSE2x I noticed they worked but when the got to 99.99% done they just sat there and did not finish. at one point it went from 23% to 900.00% done then back and forth several times then they just error-ed out The old ak app does not work for v7 multibeam but I am still getting 603 files after copied my old files back into project directory. I will try a manual install and see if that helps at all |
©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.