Message boards :
Number crunching :
3rd.in - optimized apps
Message board moderation
Previous · 1 . . . 25 · 26 · 27 · 28 · 29 · 30 · 31 . . . 40 · Next
Author | Message |
---|---|
Send message Joined: 8 Nov 08 Posts: 178 Credit: 6,140,854 RAC: 0 |
ATI needs some of my money and I can't give it to them w/o a Linux version so I can buy another card or two. Talk to Cluster Physik. I believe he's said he won't release a Linux version of the current app (it's not even GPL'd like the rest of the source), but when the CUDA app source is release, he'll tweak that and release it as GPL (as he has to), and that's when we should get a Linux app. That's all I'm waiting for to shift my desktop over to Linux here from Win7. |
Send message Joined: 7 Mar 08 Posts: 5 Credit: 231,427 RAC: 0 |
still no hope for a HD3xxx card below 3850 like a 3650? I have some of those which want to contribute! |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
still no hope for a HD3xxx card below 3850 like a 3650? I have some of those which want to contribute! Talk to Travis. The GPU project that is (was?) planned to be set up was announced to accept also GPUs only capable of single precision (that would include the complete HD2000, HD3000 and HD4000 series). The app for ATIs is basically ready, but the project is not up and running. I've posted the results of the new app (not compatible with the current MW project, the GPU project uses much longer WUs) already on June 21st in the code discussion forum. So the project can be sure it calculates the correct stuff with the required precision. |
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0 |
Is there going to be a Linux version of this bad boy? Originally I thought BOINC would support ATI GPU in the meantime. The problem is really that with the lack of support for ATI GPUs there is the need of quite some OS specific code in the app to handle the scheduling (would not be needed otherwise). The plan was to get rid of it with the release of an ATI aware BOINC version which would make the Linux port a 'piece of cake'. In principle the scheduling stuff should be easily portable to Linux. The problem is I've neither experience with Linux nor an appropriate testing environment. So if the GPU project gets off the ground, I will publish the code of the ATI version (or will give it to Travis who can do everything he wants with it) and maybe someone else will do the Linux adaption of the scheduling (won't be hard for someone who knows how mutexes and semaphores work). |
Send message Joined: 29 Jun 09 Posts: 2 Credit: 14,721,568 RAC: 0 |
I would like to thank all you guys that put your time and efforts into zslip and the ati optimized apps. i have been a firm believer that some somebody would get the code right and ati could show its real power. I have a 4890HD @ 1ghz that can usually crank a wu out every 45 - 50 seconds So thanks Alot Ryan |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
I would like to thank all you guys that put your time and efforts into zslip and the ati optimized apps. i have been a firm believer that some somebody would get the code right and ati could show its real power. Thanks for that Ryan. I certainly appreciate the ATI optimized apps and the effort that Cluster Physik and his colleagues have put into them, and I'm glad zslip.com is still useful. |
Send message Joined: 29 Jun 09 Posts: 2 Credit: 14,721,568 RAC: 0 |
I'm curious about some of the code in the app_info_xml <flops>1.0e11</flops> <avg_ncpus>0.1</avg_ncpus> <max_ncpus>1</max_ncpus> i assume flops means its capped @ 1 and the cpu thing is that 1 chip or how does that work with quad cores? the other thing if you have 2 4870 x2's would that be like 4 gpu so you could do like 2 wu per gpu? we have a few guys from places round the world who are curious. Im really curious how all this works or if there are changes i can make to get mine doing more. 100,000 pionts per day is not enough |
Send message Joined: 17 May 09 Posts: 22 Credit: 161,135,083 RAC: 0 |
I'm curious about some of the code in the app_info_xml I have tried 2 4870 x2 and its more trouble than its worth. Using my i7 920 with one 4870 x2 doing 7 climatepredection WUs and using Gispel 19f code I am also able to do 6 MW WUs as well . no problems no hicups. at the moment averging 7000 credits per hour 170000 credits a day. I find its very important to do 2 applications and get a balance between CPU based apps and the GPU MWs . about 18 tasks at once , if more it wastes processing time. Ross |
Send message Joined: 11 Apr 08 Posts: 81 Credit: 57,755,743 RAC: 0 |
|
Send message Joined: 15 Aug 09 Posts: 9 Credit: 318,640 RAC: 0 |
Greetings, I have been a Boinc user for about 2 years, but I just discovered MilkyWay yesterday. I have been watching the development of the CUDA interface for some time, but being a loyal ATI user, I am very pleased and excited that someone has finally taken up the challenge of implementing the ATI GPUs for computations. I now have it working pretty well, so I thought I would share what I have done. I downloaded the MW package and installed it and made the necessary file name changes, etc. and it didn't seem to work at all. It quickly blew thru about 20 WU's giving "computation error" on each one. I made some changes to the CMDLINE variable but they didn't seem to help. It was only later that I discovered that the BoincTray process was still running even after aborting the Boinc app. Obviously, this needed to be aborted for the APP_INFO changes to take hold. The documentation mentioned aborting the "service", and I do know the difference between a "service" and a "process" as they appear in two different places on the TaskManager. So I was looking in the wrong place. My mis-conception from the docs thru me off for a few hours. Even though I killed the BoincMgr application (aborts two processes) the BoincTray still seems to keep running. Then I discovered that my Boinc was an older version so I updated it to v6.6.36. In reading the forum I also discovered a new version of the ATI Catalyst driver posted over at Guru3D.com. It was version 9.8 and I was running 9.7 which is the latest version posted on the ATI support site. Amazing to me that the latest drivers are NOT available thru ATI, but thru an enthusiast site. That is very valuable to know. Anyway, I am running a Vista Ultimate X64 on a dual quad-core Xeon with 16GB ram and a Radeon 3870. I am really amazed at how fast MilkyWay is running thru the work queue. It is taking about 6-8 minutes to complete one WU, whereas the Rosetta WUs I am also crunching are taking about a day to reach 50% completition. And I am still able to use my system for just about everything else I need it for. Not so sure about PhotoShop which I am just about to fire up, since I know it also may used the GPU directly. Now using Boinc v6.6.36 and Catalyst v9.8 The CMDLINE variable I am using is "f20 n2" and as I said it is working smooth as silk. I realize that this is early in what will be a continuous development process. But so far it looks very good to me. This was definately worth the 5+ hours I spent to get it running correctly. n2n |
Send message Joined: 27 Aug 07 Posts: 915 Credit: 1,503,319 RAC: 0 |
It was only later that I discovered that the BoincTray process was still running even after aborting the Boinc app. The boinctray process monitors your computer activity. It's needed for your computer prefs Suspend work while computer is in use? If you have this set to NO you can kill that process. I just keep it disabled. me@rescam.org |
Send message Joined: 28 Aug 07 Posts: 35 Credit: 88,863,882 RAC: 3,506 |
If I have the optimised app and app_info file from zslip will that stop me getting the new CUDA wu that are out for NVIDEA GPU's. I am asking because I was running this new PC on the standard application from MW and I also got the new CUDA wu's as well to process. I want to make sure there isn't anything I need to add to the app_info so that it runs the CUDA application as it is not in the app_info ? The app_info I downloaded is this one. <app_info> <app> <name>milkyway</name> </app> <file_info> <name>astronomy_0.19_SSSE3.exe</name> <executable/> </file_info> <app_version> <app_name>milkyway</app_name> <version_num>16</version_num> <file_ref> <file_name>astronomy_0.19_SSSE3.exe</file_name> <main_program/> </file_ref> </app_version> <app_version> <app_name>milkyway</app_name> <version_num>19</version_num> <file_ref> <file_name>astronomy_0.19_SSSE3.exe</file_name> <main_program/> </file_ref> </app_version> </app_info> Proud Founder and member of Have a look at my WebCam |
Send message Joined: 16 Jun 09 Posts: 85 Credit: 172,476 RAC: 0 |
They should work together okay |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
If I have the optimised app and app_info file from zslip will that stop me getting the new CUDA wu that are out for NVIDEA GPU's. They should work together okay The app_info from zslip only affects work for ATI GPUs. The 6.*.* versions of BOINC that detect CUDA will start or stop work for CUDA, not the presence of any app_info for ATI. |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
The 'zslip' optimized apps web page has been transferred to; http://www.brilliantsite.com/milkyway zslip.com will redirect to the above link for a while, until the meter runs out... |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
Version 0.19g is now available for ATI GPUs. Thanks again to Gipsel and to the testers : HiRN, Emploi, Indiana_74 and other people of Team Planet3DNow!, http://www.planet3dnow.de You can also find downloads here; http://www.brilliantsite.com/milkyway Don't forget to read the readme.txt file that comes with the downloads. |
Send message Joined: 14 Feb 09 Posts: 999 Credit: 74,932,619 RAC: 0 |
|
Send message Joined: 24 Dec 07 Posts: 1947 Credit: 240,884,648 RAC: 0 |
I'm seeing 0.19g taking 3 to 5 seconds longer per wu. |
Send message Joined: 16 Feb 09 Posts: 109 Credit: 11,089,510 RAC: 0 |
I've noticed on my system that on average, 19g is taking an additional 15 seconds of GPU time and 40 seconds of wall time per task for the same batch of work. 3 concurrent tasks running on HD4870, Win7_64, Cat 9.5. Perhaps the new brook64.dll file allows for a more recent Catalyst version to be run? I haven't had the time to experiment further. |
Send message Joined: 30 Dec 07 Posts: 311 Credit: 149,490,184 RAC: 0 |
Yes I noticed processing time increased by just over 50% with 0.19g. Couldn't get GPU load over about 60% no matter what wait factor I tried. GPU processing was jerky with load occasionally falling to zero. Not as jerky as Collatz with -k512 but the same kind of effect. Runs really cool like Collatz too. Mucked around with it for half an hour then changed back to 0.19f and went to bed. |
©2024 Astroinformatics Group