Drop in WU size?

Author	Message
huangyizhe Send message Joined: 6 Jan 18 Posts: 1 Credit: 16,311,158 RAC: 0	Message 710 - Posted: 20 Jan 2018, 2:56:19 UTC Last modified: 20 Jan 2018, 3:18:52 UTC I noticed a recent drop in WU size to about 1/8. Is there any particular reason for that? ID: 710 · Rating: 0 · rate: / Reply Quote

Sergei Chernykh Project administrator Project developer Send message Joined: 5 Jan 17 Posts: 534 Credit: 72,451,573 RAC: 0	Message 711 - Posted: 20 Jan 2018, 8:51:16 UTC - in response to Message 710. Last modified: 20 Jan 2018, 8:52:06 UTC We've had a bunch of very GPU-friendly (highly parallelizable) workunits recently, but as far as I can tell, we're back to regular workunits again. All workunits were calibrated to have approximately the same amount of computations for CPU version. Run times on GPU are usually stable too, but sometimes they can vary a lot. ID: 711 · Rating: 0 · rate: / Reply Quote

Hotbit Send message Joined: 30 Aug 17 Posts: 4 Credit: 56,166,167 RAC: 0	Message 728 - Posted: 6 Feb 2018, 13:59:28 UTC - in response to Message 711. Interesting information. As WUs are run on GPUs - is it possible to calibrate them for GPUs rather than CPUs? ID: 728 · Rating: 0 · rate: / Reply Quote

Sergei Chernykh Project administrator Project developer Send message Joined: 5 Jan 17 Posts: 534 Credit: 72,451,573 RAC: 0	Message 729 - Posted: 6 Feb 2018, 16:08:10 UTC - in response to Message 728. The speed increase for GPU over CPU depends on how well a WU can be parallelized, but it's hard to predict without actually running each WU. It's even different between various GPU models. CPUs are more stable because they don't run that many parallel threads. That's why WUs were normalized for CPUs. ID: 729 · Rating: 0 · rate: / Reply Quote