21) Message boards : Application Code Discussion : CUDA for Milkyway@Home (Message 26754) Posted 30 Jun 2009 by [P3D] Crashtest Post: Are there any news about the CUDA App for MW (running on Windows) ? I would like to compare the "Power" of the CUDA-Cards running MW unit with Double Prec. like the current ATI-Cards.

22) Message boards : Number crunching : 64 CPUs on a Q6600 ???? (Message 25767) Posted 17 Jun 2009 by [P3D] Crashtest Post: The problem is - MW is only sending work for up to 8 Cores/CPUs whatever. ... so the current maximum WU-cache/bunker is 48 WUs per System. If you need more WU/s you have to run some VMs on your system If you got a 12 Core System like a Dual-Istanbul-System you wont get enough work. But if you got a Singlecore-CPU like the (my) old Athlon(64) with a MW-working GPU - 6 WUs are a bad joke... so 48 WUs are nice !

23) Message boards : Number crunching : 64 CPUs on a Q6600 ???? (Message 25752) Posted 17 Jun 2009 by [P3D] Crashtest Post: "....and with a full cache of 64 x 6 workunits in progress!" thats wrong - you can only get up to 48 WUs even if you got more than 8 CPUs/Cores etc

24) Message boards : Number crunching : 64 CPUs on a Q6600 ???? (Message 25707) Posted 16 Jun 2009 by [P3D] Crashtest Post: its a mod. boincmgr to get more MW work made by a user of the glorious Team Planet3Dnow! : it reports 64 CPUs to boinc project even if you got only 1 core it requests every 60s work for MW if you belong to the correct team

25) Message boards : Number crunching : Random screen freeze (Message 25543) Posted 15 Jun 2009 by [P3D] Crashtest Post: got this too: PC1 - 4830 1gb, win7x64 - catalyst 9.5 and PC2 - 4830 ,5gb, srv03x64 - catalyst 8.12 with app "e" and the current "f" but only sometimes ...

26) Message boards : Number crunching : this site has been slow (Message 25122) Posted 12 Jun 2009 by [P3D] Crashtest Post: @Travis: This project should devide the project to 2 hosts (or more): - Seti@Home is running on 10 hosts / computers -> http://setiathome.berkeley.edu/sah_status.html - Rosetta@home is running on 5 hosts / computers -> http://boinc.bakerlab.org/rosetta/rah_status.php Running 1 massive boincproject on a small computer like MW is doing cant work - and if the CUDA-DP-App gets ready ..... DoS. Dont even try to run MW_GPU-project on the current system too ... Small question: Whats the systemconfig of the current project host / computer / server ? - CPUs - RAM - HDDs (Raid ?) - UPS

27) Message boards : Number crunching : How can I crunch on both GPUs in a ATI4870x2? (Message 25062) Posted 11 Jun 2009 by [P3D] Crashtest Post: yes you can / should be able to do this

28) Message boards : Number crunching : ps_sgr_*F_2s_1 (Message 25060) Posted 11 Jun 2009 by [P3D] Crashtest Post: ps_sgr_210F_2s_1_4369_1244753324_0 - 45s on 4830 - 27,7 Credits ps_sgr_226F_2s_1_5_1244752853_0 - 45s on 4830 - 27,7 Credits ok so far, but dont call them big ....

29) Message boards : Number crunching : new joking WU-size/type ps_new_11 and ps_new_13 (Message 25044) Posted 11 Jun 2009 by [P3D] Crashtest Post: ps_newtest_2... same problem - done in >5s on a 4830

30) Message boards : Number crunching : new joking WU-size/type ps_new_11 and ps_new_13 (Message 25022) Posted 11 Jun 2009 by [P3D] Crashtest Post: the new ps_new_15 got the same problem - done in >4s on a slow 4830

31) Message boards : Number crunching : new joking WU-size/type ps_new_11 and ps_new_13 (Message 25015) Posted 11 Jun 2009 by [P3D] Crashtest Post: MW is joking again ! The new WU-types ps_new_11 and ps_new_13 are done is less than 4s with ATI GPUs ....

32) Message boards : Number crunching : How can I crunch on both GPUs in a ATI4870x2? (Message 24988) Posted 11 Jun 2009 by [P3D] Crashtest Post: Its possible to crunsh n-WUs on y-GPUs, Gipsel's app 0.19f: - n-WUs (setting n2 = 2WUs per GPU) on y-single-gpu-cards or (y/2) 38?0x2 / 48?0x2 cards or a mix of it: Example: you can crunsh 6 MW-WUs at the same time if you got: a 4870x2 and a 4870 = 3x 4870-GPU and the setting n2 or 3x ATI4870-Cards = 3x 4870-GPU ! current maximum is 4 4870x2-Cards = 8 GPUs, with n2 = 16WUs; with n3 = 24WUs ... BUT: all cards must be active (Monitor or dummy pluged in or Xfire active) REAL EXAMPLE = http://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=48886 core_client_version>6.5.0</core_client_version> <![CDATA[ <stderr_txt> Running Milkyway@home ATI GPU application version 0.19f by Gipsel allowing 2 concurrent WUs per GPU CPU: Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz (8 cores/threads) 2.67772 GHz (282ms) CAL Runtime: 1.3.158 Found 3 CAL devices Device 0: ATI Radeon HD 4800 (RV770) 1024 MB local RAM (remote 2047 MB cached + 2047 MB uncached) GPU core clock: 780 MHz, memory clock: 1000 MHz 800 shader units organized in 10 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads supporting double precision Device 1: ATI Radeon HD 4800 (RV770) 1024 MB local RAM (remote 2047 MB cached + 2047 MB uncached) GPU core clock: 780 MHz, memory clock: 1000 MHz 800 shader units organized in 10 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads supporting double precision Device 2: ATI Radeon HD 4800 (RV770) 512 MB local RAM (remote 2047 MB cached + 2047 MB uncached) GPU core clock: 780 MHz, memory clock: 1000 MHz 800 shader units organized in 10 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads supporting double precision 2 WUs already running on GPU 0 2 WUs already running on GPU 1 2 WUs already running on GPU 2 No free GPU! Waiting ... 27.8616 seconds. Starting WU on GPU 0 ....

33) Message boards : Number crunching : Power Usage (Message 24960) Posted 11 Jun 2009 by [P3D] Crashtest Post: 1,6kWh is too much for 2 CPUs and 2x2 GPUs: the 4870x2 use up to 300W - 2x = 600W the CPUs dont use more than 150W - 2x = 300W + HDDs etc = 100W 600W+300W+100W= 1000W -> 1,0kWh (1,3kWh if its a "bad" PSU !)

34) Message boards : Number crunching : Power Usage (Message 24944) Posted 11 Jun 2009 by [P3D] Crashtest Post: Well my powermeter shows 510W if I crunsh MW with 2 ATI GPUs and crunsh abc@home with the 2 Quadcore Opterons ...

35) Message boards : Number crunching : 3rd.in - optimized apps (Message 24788) Posted 10 Jun 2009 by [P3D] Crashtest Post: My fault - S@Hcal wasnt a S@H / Boinc "project" it was an ATI and its closed, because ATI is working hard on OpenCL for the mass and BOINC got the plan to support OpenCL soon Andrew Dodd wrote: Right now we’re focusing all of our efforts into supporting OpenCL, so you will not see CAL based acceleration for Seti@Home. I know it’s not the answer you wanted to hear, but we really think getting a solid OpenCL drivers is our number 1 goal So DA / Berkeley should change their plan: 3) CUDA 4) ATI 5) OpenCL 6 to n) Larabee, etc. could/should be shrinked to 3) OpenCL because nVIDIA: OpenCL on top of CUDA ATI: OpenCL on top of CAL/Brook+ Fruit: OpenCL native Linux: OpenCL native Intel Larabee : OpenCL native / software emul .... so all we need is OpenCL and a "recent" driver and not 3 to n fu**ing systems, apps ect

36) Message boards : Number crunching : 3rd.in - optimized apps (Message 24574) Posted 8 Jun 2009 by [P3D] Crashtest Post: I heard of only one single build - thats nothing comp. to the hundrets of CUDA-builds before public. But the work stopped before it realy started. Its only a simple alpha test of running S@H on CAL - no optimizations (run the CPU-WUs, not the larger CUDA-WUs) as a kind of tech. demo. If some here is porting S@H to support CAL that will be great - I would test it on all my machines.

37) Message boards : Number crunching : 3rd.in - optimized apps (Message 24567) Posted 8 Jun 2009 by [P3D] Crashtest Post: Why should ATI use a "enemy" API if they got there own and CAL was first ! "No. That has nothing to do with cuda... the "optimized" part is the fall back mode if the cuda app crashes..." S@H only use about 30-70% of the nVIDIA GPUs for calc, rest is calced on CPU (with SSE, SSE2 .....) and the org. APPs arent the best ! Even with the GPUs, S@H could not use then "good" enough - a PC with 8 nVIDIA GPUs got only a rac of 58,126.16 - I could get a higher rac if I only use 2 normal ATI 4850 in a small PC on MW ( MW-RAC x 0,75 > S@H-RAC ) Using 4 ATI 4870x2 you could get a RAC ( MW-RAC x 0,75 ) of 250000 or more if you get enough WUs. inside the boinc src (http://boinc.berkeley.edu/trac/browser/trunk/boinc/lib/coproc.cpp) as only one example, there's everything CUDA-coded, to use CUDA: A good code would be COPROC_GPU incl. an export "supporting double prec." but theres only COPROC_CUDA and CUDA-fx only - and the project apps only use COPROC_CUDA and not COPROC_GPU! Inside COPROC_GPU could be a CUDA and a CAL-Part ! To add ATI-CAL and OpenCL support, they had to rewrite most parts of the entire sourcecode (Scheduler, Memory-Management, FS) and the project-apps too. If they replace CUDA with OpenCL, they only had to do this once ! (and not 3 or more times) I'm sure they will kick CUDA in Q4/09 - Q1/10 if OpenCL-drivers are stable (SDKs are out from both nVIDIA and ATI, Intel will follow) OpenCL pro: could/should/will work on ATI, nVIDIA, Intel, S3, etc GPUs; x86, arm, Sparc CPUs Windows, Linux, ect OS OpenCL contra: not jet ready - will change with CAL SDK 1.5 and new CUDA-SDK because both will use OpenCL as top layer api and the "old" CAL/CUDA below it The lack of info's reported by boinc will/could be a problem for MW_GPU too, because BOINC didnt send informations about SP or DP to the projects - MW_GPU will use a lot of DP ! THE S@HCAL App isnt public and with DA it wont be until boinc 6.10.x or later; they got some ATI cards !

38) Message boards : Number crunching : 3rd.in - optimized apps (Message 24545) Posted 8 Jun 2009 by [P3D] Crashtest Post: For nvidia you need a "recent" driver and hardware - ATI too Gipsel could build the app using the current CAL SDK, so it would work on Catalyst 9.2 or newer without any file copy/paste, but when it wouldnt work on Catalyst 9.1 or older or it could work with file copy/paste The "normal" nvidia-apps arent CPU-optimized - ATI apps are CPU-optimized but theres a basic version only using SSE to work with old Athlons / K7 /x86 There are CPU-optimized apps for CUDA on S@H too - with the "same" problem : SSE, SSE2, SSE3, SSSE3, SSSE41 .... ?!? This basic "problem" ,if it is one, wont change on MW_GPU because of BOINC and the detection of GPUs - only hardcoded CUDA ! So we ATI-USER will have to copy the new Gipsel-App to the project folder like we have to do on MW. ATI GPU support was planed for BOINC 6.4.X, later for 6.6.X, later for 6.8.X and is now planed for 6.10.X ! But there are 3 main problems: 1. nVIDIA doesnt like the performance of the 48xx ATI-GPUs on S@H, MW (yes theres a "working" S@H ATI app (alpha) since December 2008) 2. ATI/AMD didnt provide enough support (Cards, Documentation, Money) like nVIDIA did 3. CUDA is hardcoded inside BOINC, so they have to change many codelines to add a 3rd working path (1st CPU, 2nd CUDA) Maybe they (BERKELEY) will stop the CAL and CUDA support and move to OpenCL as it was planed for 6.10.x - this would be easier than supporting 3 different APIs (CAL, CUDA & OpenCL) and more hardware could BOINC (ATI, nVIDIA, Intel, S3 ....) and it would be easier for the projects - only to build one single app because OpenCL could run on all "new" GPUs (DX10 or newer) or on CPU.

39) Message boards : Number crunching : Compute Errors (Message 24278) Posted 5 Jun 2009 by [P3D] Crashtest Post: Gipsel got it fixed with version 0.19f: http://www.file-upload.net/download-1684038/Milkyway_0.19f_ATI.zip.html Vielen Dank Gipsel !

40) Message boards : Number crunching : Compute Errors (Message 24238) Posted 5 Jun 2009 by [P3D] Crashtest Post: Yea - stop sending "3s" WU for about 2 days - Gipsel is working on it ! I got 48new WUs at once - all "3s" - 60s later - all crashed ... The Milkyway-Project-Team is waisting WUs ... !

Previous 20 · Next 20

©2024 Astroinformatics Group