Posts by Jeroen

1) Message boards : Number crunching : Performance discrepancy (Message 63358) Posted 12 Apr 2015 by Jeroen Post: The applications via this project depend on FP64. In general, consumer Kepler excluding Titan lags behind Fermi architecture due to Kepler having FP64 locked down to 1/24 of the FP32 compute rating. The GT 430 is set at 1/12 of the FP32 rating. By my calculations, the GT 640 still comes ahead due to having a much higher FP32 GFLOPS rating than the GT 430 and should run at approximately 25.5-33.5 GFLOPS FP64 compared to 22.4 GFLOPS FP64 of the GT 430. There are several variants of the GT 640 with different compute ratings. Based on this, my guess would be that the GT 640 is not boosting to full clock speed due to the GPU not being fully loaded. You can setup an app_config.xml file in the milkyway project folder and run two tasks per GPU to see if this increases the GPU load and resulting frequency. <app_config> <app> <name>milkyway</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.05</cpu_usage> </gpu_versions> </app> </app_config>
2) Message boards : Number crunching : AMD Radeon R9 290X (Message 60332) Posted 8 Nov 2013 by Jeroen Post: Hi MW don't run (App version needs CAL but GPU doesn't support it) PrimeGrid don't run (App version needs CAL but GPU doesn't support it)But runing using app_info. Run a little slower than HD 7970 Poem and Einstein. working run a little slower than HD 7970 Collatz and DistRTgen. working run about 50% better than HD 7970 Hello, I was wondering if you could test out your R9-290x with running three Einstein BRP5 tasks (0.33 utilization factor) and post what the run time is per task? If you could run a batch of GPU tasks without any other CPU tasks running that would be great. Thanks, Jeroen
3) Message boards : Number crunching : GPU Requirements (Message 57391) Posted 1 Mar 2013 by Jeroen Post: I have tested Titan last day on milkyway and it works ok. When doubleprecision is enabled it can run 7 tasks in parallel (~97% GPU, 80% of TDP=200W) and can produce cca 300k credits per day. Thanks for sharing your results. Could you try running one task at a time with the 1/3 DP option enabled and post the runtime per task? I wanted to see how the Titan runs compared to the 7970 with a single task running. Also, do the overclocking and power target options work with DP mode enabled?
4) Message boards : Number crunching : Variation in GPU time (Message 56275) Posted 23 Nov 2012 by Jeroen Post: Thanks. I will try to lower the memory frequency. Any extra power savings will be good to have.
5) Message boards : Number crunching : Variation in GPU time (Message 56273) Posted 22 Nov 2012 by Jeroen Post: Hello, I just recently started running a 7970 card for Milkway. I have been running NVIDIA cards for the last decade and am new to recent AMD cards. I am noticing in both Linux (Slackware64 14/Ubuntu 12.10) and Windows 7 64 that there is a fairly significant difference in run time from one task to another. Generally when I first start BOINC, the first task will run 41-57 seconds. Then most every other task after that runs 61-93 seconds. Is this normal? I have no other load on the CPU and have disabled EIST and C-states power savings features. The GPU has a factory frequency of 1100 MHz but runs at 1050 MHz with this particular project and the memory frequency is 1500 MHz. GPU temperature is 55-60C under load. I see no indication in OverDrive that the GPU frequency is dropping under load. Thanks.
6) Message boards : Number crunching : HD7970 on the horizon .. (Message 52440) Posted 13 Jan 2012 by Jeroen Post: I think I might pickup a pair of 7970s for running this project. The best I can get out of my 580 GTX is 128 seconds for a single task which seems to be decent but hopefully I can get even better performance out of the 7970.
7) Message boards : Number crunching : Linux CUDA (Message 52112) Posted 29 Dec 2011 by Jeroen Post: This happens if the ICD is missing from Nvidia's OpenCL driver. How did you install the drivers? I'm running a small custom Linux setup and because of that I had to build and install the drivers and libraries manually. I have Milkyway@home working now. Thank you for the help and info regarding the ICD. I was missing /etc/OpenCL/vendors/nvidia.icd.
8) Message boards : Number crunching : Linux CUDA (Message 52047) Posted 24 Dec 2011 by Jeroen Post: I have been trying to run Milkyway@Home OpenCL app in Linux. However the tasks are failing. Here is one of them: http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=53096855 I made sure libOpenCL.so.1 is in place. This particular system is running other CUDA projects okay (Seti, Einstein, Primegrid, GPUGrid, DNETC) however this is the first one I have tried based on OpenCL. Driver version is 290.10 and GPU is GTX 580. Has anyone seen this issue? Thanks.