Message boards :
Number crunching :
New AMD Driver slower?
Message board moderation
Author | Message |
---|---|
Send message Joined: 5 Mar 14 Posts: 24 Credit: 500,964,006 RAC: 0 |
So i don't send myself or anyone else down the wrong path I want to be explain everything that I changed. I noticed I was getting a high-ish amount of error WUs, mostly modified fit after I noticed my daily average slipping. I thought maybe 7 concurrent was too much and maybe it needed a bit more CPU. So i turned down to 6 concurrent but I also updated driver to the newest amd crimson hotfix. I noticed it greatly reduced cpu usage on Collatz on another card. Anyways after reducing concurrency and running the new driver, I noticed fewer errors but went further and turned off modified fit. Currently at 6 concurrent i'm getting around 222sec per non modified-fit wu. I used to get around 100-120 secs on non-modified fit wus. Here are some outputs: Old driver Platform 1 information: Name: AMD Accelerated Parallel Processing Version: OpenCL 2.0 AMD-APP (1912.5) Vendor: Advanced Micro Devices, Inc. Extensions: cl_khr_icd cl_khr_d3d10_sharing cl_khr_d3d11_sharing cl_khr_dx9_media_sharing cl_amd_event_callback cl_amd_offline_devices Profile: FULL_PROFILE Using device 0 on platform 1 Found 1 CL device Device 'Tahiti' (Advanced Micro Devices, Inc.:0x1002) (CL_DEVICE_TYPE_GPU) Driver version: 1912.5 (VM) Version: OpenCL 1.2 AMD-APP (1912.5) Compute capability: 0.0 Max compute units: 32 Clock frequency: 1050 Mhz Global mem size: 3221225472 Local mem size: 32768 Max const buf size: 65536 Double extension: cl_khr_fp64 Build log: Estimated AMD GPU GFLOP/s: 4301 SP GFLOP/s, 1075 DP FLOP/s Using a target frequency of 10.0 Using a block size of 8192 with 273 blocks/chunk Using clWaitForEvents() for polling (mode -1) Range: { nu_steps = 320, mu_steps = 1600, r_steps = 1400 } Iteration area: 2240000 Chunk estimate: 1 Num chunks: 2 Chunk size: 2236416 Added area: 2232832 Effective area: 4472832 Initial wait: 0 ms Integration time: 100.745453 s. Average time per iteration = 314.829540 ms Integral 0 time = 102.021333 s Running likelihood with 37047 stars Likelihood time = 0.411575 s <background_integral> 0.000254840452682 </background_integral> <stream_integral> 61.112214151510841 79.895548307261919 0.982550827806658 </stream_integral> <background_likelihood> -3.141897295188446 </background_likelihood> <stream_only_likelihood> -16.172241066585244 -5.891336555601848 -41.880991865214206 </stream_only_likelihood> <search_likelihood> -2.898416444780255 </search_likelihood> 18:20:10 (7836): called boinc_finish Newer driver: Platform 1 information: Name: AMD Accelerated Parallel Processing Version: OpenCL 2.0 AMD-APP (2004.6) Vendor: Advanced Micro Devices, Inc. Extensions: cl_khr_icd cl_khr_d3d10_sharing cl_khr_d3d11_sharing cl_khr_dx9_media_sharing cl_amd_event_callback cl_amd_offline_devices Profile: FULL_PROFILE Using device 0 on platform 1 Found 1 CL device Device 'Tahiti' (Advanced Micro Devices, Inc.:0x1002) (CL_DEVICE_TYPE_GPU) Driver version: 2004.6 (VM) Version: OpenCL 1.2 AMD-APP (2004.6) Compute capability: 0.0 Max compute units: 32 Clock frequency: 1050 Mhz Global mem size: 3221225472 Local mem size: 32768 Max const buf size: 65536 Double extension: cl_khr_fp64 Build log: Estimated AMD GPU GFLOP/s: 4301 SP GFLOP/s, 1075 DP FLOP/s Using a target frequency of 10.0 Using a block size of 8192 with 273 blocks/chunk Using clWaitForEvents() for polling (mode -1) Range: { nu_steps = 320, mu_steps = 1600, r_steps = 1400 } Iteration area: 2240000 Chunk estimate: 1 Num chunks: 2 Chunk size: 2236416 Added area: 2232832 Effective area: 4472832 Initial wait: 0 ms Integration time: 213.290785 s. Average time per iteration = 666.533702 ms Integral 0 time = 218.105308 s Running likelihood with 37047 stars Likelihood time = 0.431816 s <background_integral> 0.000254884792084 </background_integral> <stream_integral> 61.108815281212131 80.106593895321026 0.982497705285343 </stream_integral> <background_likelihood> -3.141935983154649 </background_likelihood> <stream_only_likelihood> -16.186953952170722 -5.889633635476240 -41.880813709490866 </stream_only_likelihood> <search_likelihood> -2.898416486402882 </search_likelihood> 12:49:19 (1936): called boinc_finish Can anyone else confirm slower run times on the newer driver? |
Send message Joined: 27 Jul 11 Posts: 21 Credit: 235,255,105 RAC: 0 |
Yes, i noticed that too(40s with 1 WU MWhome 1.02 vs 32s previously). They done something with power management in DP calc mode i presume. So i changed power consumption to 15% in Crimson OverDrive panel and i got previous results. |
©2024 Astroinformatics Group