Posts by Mumak

1) Message boards : Number crunching : Out Of Work? (Message 65394) Posted 6 Oct 2016 by Mumak Post: It was working great, but now I'm sometimes getting: Milkyway@Home 06-Oct-16 9:10:45 Server can't open database That causes to client to back-off and run out of tasks again. Can you please check this?
2) Message boards : Number crunching : Out Of Work? (Message 65382) Posted 4 Oct 2016 by Mumak Post: Today it works great. WUs are properly supplied, so all my GPUs are fully utilized.
3) Message boards : Number crunching : New AMD 480 GPU (Message 65371) Posted 3 Oct 2016 by Mumak Post: RX 480 works for me. Some times for comparison: HD7950: ~13-14 s Fury X: ~15-17 s RX 480: ~20-21 s Not bad..
4) Message boards : Number crunching : WhoÂ´s the MAN or WOMAN whoÂ´s running a Radeon R9 Nano ? (Message 64276) Posted 28 Jan 2016 by Mumak Post: I assume you're talking about the Fury or Nano (Fiji). Ah indeed - I just noticed the topic name ;-) This project uses double precision (DPFP) and those cards have significantly reduced performance in DPFP. Other projects like Einstein@Home are much more benefiting from Fiji-class GPUs (also due to their high memory bandwidth and E@H tasks require a lot). The old Radeon HD 79xx/280x (Tahiti) are still the best performers here, because they have a high DPFP rate (not crippled). Those cards can crunch a MW task in 10-25 secs.
5) Message boards : Number crunching : Insufficient amount WU on the server (Message 64153) Posted 5 Dec 2015 by Mumak Post: I'm too often running out of work and recently getting too many validation errors. Guess I'm switching back to Einstein...
6) Message boards : Number crunching : Problem with running two milkyway work units (Message 64134) Posted 29 Nov 2015 by Mumak Post: You might enable the Milkyway@Home Separation (Modified Fit) units too and since these take a different amount of time, running both types mixed should get them 'out of sync'.
7) Message boards : Number crunching : The increase in the calculation time at Win 10 (x64) (Message 64122) Posted 24 Nov 2015 by Mumak Post: Now after installing the Crimson drivers it's back to normal. Wondering how long it will stay...
8) Message boards : Number crunching : The increase in the calculation time at Win 10 (x64) (Message 64113) Posted 17 Nov 2015 by Mumak Post: Spoke too early - was running well for a few days, but now it's again consuming a lot of CPU time at the beginning. No idea why is that...
9) Message boards : Number crunching : The increase in the calculation time at Win 10 (x64) (Message 64098) Posted 14 Nov 2015 by Mumak Post: FWIW - I checked logs and it seems that it was the upgrade to Win10 TH2 (1511) that fixed things.
10) Message boards : Number crunching : The increase in the calculation time at Win 10 (x64) (Message 64096) Posted 13 Nov 2015 by Mumak Post: Not sure whether it's because of upgrading Win10 to TH2 (1511) or Catalyst to 15.11 Beta, but it works now as before, so no extensive CPU usage anymore.
11) Message boards : Number crunching : The increase in the calculation time at Win 10 (x64) (Message 64039) Posted 25 Oct 2015 by Mumak Post: I tried, but it didn't help. I have been running MW@H on the same machines before summer and there were no such issues. After summer I returned back and see this behavior.
12) Message boards : Number crunching : Run Multiple WU's on Your GPU (Message 63995) Posted 13 Oct 2015 by Mumak Post: Oh, don't dl anything from CNet, it has become one of the worst dl sites ever. SIV has tons of features, but I find it sometimes too difficult to use.. BTW, I'm the author of HWiNFO64 ;-)
13) Message boards : Number crunching : CPU usage increase in modified fit 1.36 (Message 63990) Posted 12 Oct 2015 by Mumak Post: I don't think the issue is on server (code) side. The standard MilkyWay@Home v1.02 application is affected too and that's 3 years old code, the ModFit application is 1 year old. It seems the issue is on some systems only - if I check the results of other machines, some of them spend only ~2 secs on CPU, while others 30-50 secs. Same GPUs, AMD drivers... Could be a combination of new Catalyst + hardware ?
14) Message boards : Number crunching : CPU usage increase in modified fit 1.36 (Message 63969) Posted 6 Oct 2015 by Mumak Post: I (and some others) noticed the same after upgrading to Windows 10: http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=3817 But your computers seem to be on Windows 7, so I'm not sure what might be causing this. Maybe Catalyst upgrade ?
15) Message boards : Number crunching : The increase in the calculation time at Win 10 (x64) (Message 63958) Posted 1 Oct 2015 by Mumak Post: I see a similar issue after upgrading to Win10. One of my machine (R9 280X) spends ~30 secs initial time on CPU, another one (2xHD 7950) ~50 secs. This is a huge overhead against previous results on Win7 (1-3 seconds CPU time per task). Checking results of different users running Win10 and similar cards shows that many others are affected, but I found some that are not and running Catalyst 15.9. Not sure what's the difference and how to fix this. I have tried the latest driver 15.9.1 Beta released yesterday, but it's still the same. However I haven't been able to downgrade the AMD GPU drivers to some older version, it still retains the latest driver installed. Automatic driver update is disabled in Win10.
16) Message boards : Number crunching : New Benchmark Thread - times wanted for any hardware, CPU or GPU, old or new! (Message 63449) Posted 22 Apr 2015 by Mumak Post: OK, so Tesla K20c (ECC off, 758 MHz): 37 s The early times on my machine were with default 705 MHz clock (~40 s)
17) Message boards : Number crunching : New Benchmark Thread - times wanted for any hardware, CPU or GPU, old or new! (Message 63445) Posted 22 Apr 2015 by Mumak Post: That's most probably for single-slot GPUs only. For crunching you need GPUs with better cooling (dual-slot at least), when the airflow in case is not good, you'll need to keep an additional spare slot between them too. So 3 slots per GPU. From my experience the Tahiti tends to run about 8-10 C hotter when doing FP64. Power consumption can go even twice as high. So in the winter I'm doing MW@H and in summer I switch to lighter tasks (like Einstein) ;-) I haven't observed such a big difference between SP and FP64 tasks on the Tesla, but I'll check again.
18) Message boards : Number crunching : New Benchmark Thread - times wanted for any hardware, CPU or GPU, old or new! (Message 63442) Posted 22 Apr 2015 by Mumak Post: I have a Tesla K20c, Quadro K4000 and K4200. Quadro's are not good for crunching, since they are mostly low-profile, thus with not adequate cooling. The Tesla runs much better. I gave it a run on MW@H some time ago and the times were somewhat comparable to Radeon 280/280X. Besides having a much better FP64 performance, they offer ECC, which is a feature that's better to turn off for crunching. One also needs to point out that there are only OpenCL apps on MW available, and the Tesla would perform much better if there would be CUDA FP64 apps. I'm currently running Einstein on the Tesla, but can try MW again if you need more precise numbers. Anyway, it's a too expensive card for such purposes.
19) Message boards : Number crunching : strongest performing milkyway@home AMD card (Message 63132) Posted 10 Feb 2015 by Mumak Post: I don't think it makes sense to switch from 6970 to 7970. Currently the HD 7970 (280X) is the best performer here due to high DP performance. The 290 series (Hawaii) have heavily reduced DP performance, so it's not a good choice for this project. Theoretically the FirePro S9100 might be even better if it doesn't have crippled DP performance like the other consumer Hawaii SKUs. But I don't think anybody tested it here yet.
20) Message boards : Number crunching : Machines runs out of units every few minutes (Message 63041) Posted 18 Jan 2015 by Mumak Post: Hello Admins, anybody here? I'd appreciate a feedback on this issue. I haven't seen such problems yes, it started a few days ago and persists. It causes my machines to run dry of GPU units, thus as a side effect the CPU cores allocated to GPUs start CPU tasks, then after a while when a few GPU units are received, they switch back to GPUs and so on. This results in constant suspending/restarting/reshuffling of CPU tasks.

Next 20