Posts by RichardLux

1) Message boards : Number crunching : 7970s Dont Run (Message 5236) Posted 24 Jul 2013 by RichardLux Post: I believe MilkyWay is the only gpu project that uses double precision, I do not believe Moo does. I have half a dozen AMD 5770 gpu's and they all worked here just fine, but since they are not double precision they will not work at MilkyWay. If it is like that, then now again I think my GPU performs bad :D
2) Message boards : Number crunching : 7970s Dont Run (Message 5229) Posted 23 Jul 2013 by RichardLux Post: @mikey 1 of 8 CPU threads is free always, if I process GPU tasks. 7 CPU threads are used for other projects. The GPU is on maximum 94% load always (not sure why not 100%, maybe a limiter in the Moo/dnetc software?). I think everything is fine, but I did not know that GPU tasks are much bigger than CPU tasks. Thats why I thought there is some problem. Now I think my GPU performs very well ;) Or do you think it should/must perform better? (1'344'000 GFLOPs task within about 18 minutes) @All Will this fix be available in an official release soon? Update: One of my CPU tasks had 15750 GFLOPs and it took 83 minutes. If I compare everything then I get following... -> GPU -> 1168.70 GFLOPS -> CPU (1 thread) -> 3.13 GFLOPS -> CPU (8 threads) -> 25 GFLOPS -> The GPU is 46 times faster 1168.70 GFLOPS is more than described in the AMD specifications for double precision floating operations. If single precision would be enough then it could compute 4 times faster :D But I guess there is no performance improvement possible, because the project needs double precision (or whatever it is doing^^). I am not familiar with that, so I am shutting up now ;)
3) Message boards : Number crunching : 7970s Dont Run (Message 5219) Posted 22 Jul 2013 by RichardLux Post: Thats not right. I think the GPU will be feed with some more packages to crunch than the CPU can do it. Yes, I thought that some minutes after I posted it. I saw that the GPU tasks are much bigger (1'344'000 GFLOPs) compared with the CPU tasks, but at the moment I dont have a CPU task in my queue and I cannot compare it, but I think it was around 4000 GFLOPs for CPU tasks (?). Thanks.
4) Message boards : Number crunching : 7970s Dont Run (Message 5216) Posted 22 Jul 2013 by RichardLux Post: Hi all! Yah, the solution provided by rebirther works also for me with 7970 card (Windows 7 64 bit and CCC 13.4) Thanks :) Actually rebirther already mentioned the performance topic, but I also want to add my thoughts and experience here... With my i7 3770K with 8 cores I can finish 8 tasks within 96 minutes. With GPU I can finish 1 task within about 18 minutes. This means I can finish 2.67 more tasks with my CPU in same time. I thought that a GPU can compute much faster. I guess it is a question of the input data/type of calculation, which can be still optimized in the software, that it performs faster. Is it like that?