Message boards :
Number crunching :
Underused 970 ?
Message board moderation
Author | Message |
---|---|
Send message Joined: 30 Dec 09 Posts: 21 Credit: 75,540,465 RAC: 0 |
I used to crunch with an old GTX 460. That toaster did its job honestly. Crunching for example a MW 1.36 in roughly 90 seconds. Then i got a nice second hand 970 with five times the number of CUDA cores and an advertised quadruple FLOPS output. And i see the 970 crunches a MW 1.36 in 50 seconds. A 80% increase. Nice but below what i expected. Should i toy with the app_info.xml to make it crunch several WU's at the time ? I don't know at all how to do that. I could toy a bit with the "frame per seconds" parameter too. I talk about several WU's at the same time because it worked well when i was GPU crunching for einstein. But what do i know so i wait for you expertise before doing anything stupid. |
Send message Joined: 30 Dec 09 Posts: 21 Credit: 75,540,465 RAC: 0 |
Found this thread. Pinned. http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=3118#56955 Added an app_config.xml and got my 970 to crunch two milkyway WU's. I got a meagre 17.65% of output improvement. |
Send message Joined: 31 Dec 11 Posts: 17 Credit: 3,171,557,895 RAC: 0 |
MilkyWay uses double precision calculations, and the 460 was capped at I believe 1/12 fp64 and the 970 is capped at 1/32 fp64. Even though the 970 is a much faster GPU (at single precision aka fp32) the cap on fp64 has made all consumer NVIDIA cards after Fermi, with the exception of the OG Titan and Titan Black, very inefficient at double precision work loads. |
©2024 Astroinformatics Group