Thanks for noticing that. Yesterday I was looking at some tasks that were running on a machine running O2MDF (AMD GPU) and just happened to notice the 7 day deadline. I thought that it must be because AMD tasks were test status but then wondered (since quorum partners shouldn't have different deadlines, I imagine) if perhaps both types were still test status.
I hadn't seen anybody commenting about it and it wasn't causing any issues so I didn't feel like hassling you about it. I'm sure you have enough on your plate as it is and I'm grateful that you do keep things going as well as you do. It's a tremendously large workload that you have to cope with. Congratulations for the great service you give.
The "Beta Test" restriction for AMD GPUs has been relieved.
The App version 2.03 only fixes a minor technical issue, the app now correctly reports timing statistics (which part of the computation took how much time) back to us. In case you're curious - it's the result file ending in "_4" in the projects directory - there's no personal information in it; previous App versions wrote that in a file "GCT.timing" in the respective slot directory, which doesn't get reported.
This new version doesn't seem to be deployed yet for AMD. There's one listed for nvidia. I've actually set NNT some time ago so as to run the work cache down and provide results with the timing information as early as possible. I'm not about to run out just yet but it would be nice to know if the new app will arrive before the weekend or not. I certainly can't hold out until Monday :-).
I write here since it seems that for AMD on macOS nothing has changed since the beta and since I reported it various times I don't know if it just got lost in the forums or is fixed in 2.03 but my machine doesn't get them yet?
I get the 2.02 version and they take 2 hours to finish giving 1000 credits and the GPU has about 50% load
A FGRP task takes 15 minutes to finish and gives 3400 credits = 27 times as much credit as the GW app. So looking only at the credits it's way more efficient to run GW tasks on the CPU only if the WU size is the same as the CPU version I'd turn the GPU tasks for AMD/macOS off as CPU crunching would be more efficient then.
I just started crunching these new task and I certainly agree that the credits are by far to low.
What really bothers me however is the fact that they initially were announced to run in about 5-6 minutes, but ended up running for 70-80 minutes. Additionally, running these tasks made the boinc manager to re-announce the FGRP tasks to take a whopping 9 hours, compared to the ~34min they usually take (on a GTX960).
I cannot but conclude that the new app is at least an order of magnitude less efficient that it was thought to be (running on ordinary GTX nvidia cards). B.I.G reported a mere 50% GPU load, and I can see on my GPU temperature that the GPU seems to run a lot cooler that it does crunching the FGRPs: A bit under 70deg now compared to slightly over 80deg normally.
I will disable these tasks for now and continue with the old FGRPs until I see a new app version.
I get the 2.02 version and they take 2 hours to finish giving 1000 credits and the GPU has about 50% load
The credits are probably set for the Windows version. The 2.02 version takes 17 minutes on my RX 570 under Win7 64-bit, so you would be getting your usual credits if your RX 580 were on Windows.
If you are running concurrent tasks (3x or more) try reducing to 2x. Yesterday I saw my first validate error and, very quickly after that, three TIME_LIMIT_EXCEEDEDs. I reduced from 4x to 3x and the problem has gone away.
The 'work content' (and perhaps the memory requirement) of tasks directed at the Vela pulsar seemingly has changed so if that doesn't get modified, the volunteer will probably need to ease off to compensate, if such problems continue to show up.
Gary I'm running 2X on a GTX1060 3GB. GPU-Z shows about 2/3 of the memory being used. I seem to be getting only a couple a day. When I look at my wingmen they are often showing validate errors also. Methinks this app is pushing the envelope.
Thanks for noticing that.
)
Thanks for noticing that. Yesterday I was looking at some tasks that were running on a machine running O2MDF (AMD GPU) and just happened to notice the 7 day deadline. I thought that it must be because AMD tasks were test status but then wondered (since quorum partners shouldn't have different deadlines, I imagine) if perhaps both types were still test status.
I hadn't seen anybody commenting about it and it wasn't causing any issues so I didn't feel like hassling you about it. I'm sure you have enough on your plate as it is and I'm grateful that you do keep things going as well as you do. It's a tremendously large workload that you have to cope with. Congratulations for the great service you give.
Cheers,
Gary.
Gary Roberts
)
I just want to second that! Keep up the good work!
The "Beta Test" restriction
)
The "Beta Test" restriction for AMD GPUs has been relieved.
The App version 2.03 only fixes a minor technical issue, the app now correctly reports timing statistics (which part of the computation took how much time) back to us. In case you're curious - it's the result file ending in "_4" in the projects directory - there's no personal information in it; previous App versions wrote that in a file "GCT.timing" in the respective slot directory, which doesn't get reported.
BM
Bernd Machenschalk wrote:The
)
This new version doesn't seem to be deployed yet for AMD. There's one listed for nvidia. I've actually set NNT some time ago so as to run the work cache down and provide results with the timing information as early as possible. I'm not about to run out just yet but it would be nice to know if the new app will arrive before the weekend or not. I certainly can't hold out until Monday :-).
Cheers,
Gary.
I write here since it seems
)
I write here since it seems that for AMD on macOS nothing has changed since the beta and since I reported it various times I don't know if it just got lost in the forums or is fixed in 2.03 but my machine doesn't get them yet?
I get the 2.02 version and they take 2 hours to finish giving 1000 credits and the GPU has about 50% load
A FGRP task takes 15 minutes to finish and gives 3400 credits = 27 times as much credit as the GW app. So looking only at the credits it's way more efficient to run GW tasks on the CPU only if the WU size is the same as the CPU version I'd turn the GPU tasks for AMD/macOS off as CPU crunching would be more efficient then.
I just started crunching
)
I just started crunching these new task and I certainly agree that the credits are by far to low.
What really bothers me however is the fact that they initially were announced to run in about 5-6 minutes, but ended up running for 70-80 minutes. Additionally, running these tasks made the boinc manager to re-announce the FGRP tasks to take a whopping 9 hours, compared to the ~34min they usually take (on a GTX960).
I cannot but conclude that the new app is at least an order of magnitude less efficient that it was thought to be (running on ordinary GTX nvidia cards). B.I.G reported a mere 50% GPU load, and I can see on my GPU temperature that the GPU seems to run a lot cooler that it does crunching the FGRPs: A bit under 70deg now compared to slightly over 80deg normally.
I will disable these tasks for now and continue with the old FGRPs until I see a new app version.
Good luck with the tuning of the new app!
B.I.G wrote:I get the 2.02
)
The credits are probably set for the Windows version. The 2.02 version takes 17 minutes on my RX 570 under Win7 64-bit, so you would be getting your usual credits if your RX 580 were on Windows.
Validate errors have
)
Validate errors have reappeared starting 11/30
https://einsteinathome.org/task/902153723 https://einsteinathome.org/task/902215413 https://einsteinathome.org/task/902244878 https://einsteinathome.org/task/902244880Betreger wrote:Validate
)
If you are running concurrent tasks (3x or more) try reducing to 2x. Yesterday I saw my first validate error and, very quickly after that, three TIME_LIMIT_EXCEEDEDs. I reduced from 4x to 3x and the problem has gone away.
The 'work content' (and perhaps the memory requirement) of tasks directed at the Vela pulsar seemingly has changed so if that doesn't get modified, the volunteer will probably need to ease off to compensate, if such problems continue to show up.
Cheers,
Gary.
Gary I'm running 2X on a
)
Gary I'm running 2X on a GTX1060 3GB. GPU-Z shows about 2/3 of the memory being used. I seem to be getting only a couple a day. When I look at my wingmen they are often showing validate errors also. Methinks this app is pushing the envelope.