I just tried to get MeerKAT for a machine which had come out of daily room-heating shutdown without success. On looking at the applications page, I noticed that the BRP7 section is empty at the moment.
I imagine enough has been learned to initiate one or more adjustments before further deployment.
I apologise on behalf of the idiots that program the Boinc scheduler. I've gone through this severe bug with them before on github and they never fixed it, Richard I think you were involved. But randomly Boinc will download way too many tasks. Out of the 1100 in progress BRP7 tasks on the status page, 630 of them are on one of my computers. I told it to get a 1 day queue, instead it's got what it believes is an 18 day queue, pretty daft since it breaks both my request and the deadline by a factor of 18! I shall abort a fair number and let the rest of you take some.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I'm awaiting, with bated breath, the outcome of this workunit as it compares both ATI/NVIDIA ( OpenCL/CUDA ) and Linux/Windows.
Cheers, Mike.
( edit ) I could start a book on it : say 2 to 1 ( ie. bet 2 to win 1 ) that it goes to a tie with a third machine being called in to decide the quorum ... :-)
( edit ) It's also an Intel/AMD CPU compare for that matter.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I wanted to test out an actual legacy GPU, to see how a CUDA 5.5 era GPU would actually fare with these tasks since the project claimed to want to keep such an old CUDA version for compatibility with older GPUs. to test this, I used the v0.03 CUDA55 an v0.01 OpenCL Linux applications that both showed to process successfully (that I saved offline before they were pulled). I formed working app_info files so that I was able to download fresh tasks without an app version from the project. both app_info files proved to successfully download and process tasks for a modern 3080Ti/Ubuntu22.04/Kernel5.15 system with modern drivers.
the Test System is Ubuntu 20.04.4, 5.15 kernel, X79 platform with an E5-1680v2 Xeon CPU, a GTX 550Ti and nvidia 390.154 drivers installed from the nvidia proprietary driver package. the 340/384 drivers would not install (guessing a kernel incompatibility).
attempting to run either the v0.03 cuda55 app or the v0.01 opencl app on the 550Ti, resulted in immediate failures. however, swapping in a GTX 1060, with the exact same drivers, processing worked. so it's not the fault of the drivers, and I have no reason (at the moment) to believe the hardware itself is faulty.
even though the 390 drivers support this GPU for use as a display, it seems they don't support CUDA on this GPU anymore. a little digging showed that CC 2.1 (Fermi generation) lost CUDA support in CUDA SDK 9.0 and after, meaning an older version of the driver (and kernel for compatibility) will be necessary for any CUDA processing with these old GPUs. I will try Ubuntu 18.04 and the latest 340.108 drivers.
another interesting observations comes from the successful runs of these new tasks with v0.03 CUDA on a 3080Ti. these appear to be full size tasks. and processing took 6 minutes at 1x, and 11 minutes at 2x. only a slight speed advantage to running 2x it seems. about 5x longer running than the FGRPB1G tasks on the same hardware.
even though the 390 drivers support this GPU for use as a display, it seems they don't support CUDA on this GPU anymore. a little digging showed that CC 2.1 (Fermi generation) lost CUDA support in CUDA SDK 9.0 and after, meaning an older version of the driver (and kernel for compatibility) will be necessary for any CUDA processing with these old GPUs. I will try Ubuntu 18.04 and the latest 340.108 drivers.
Ouch. Nvidia never were good with software. When I download software for an old AMD, I'm given the latest version that supported that card, not a newer version which half supports it.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Of eight MeerKAT v0.05 tasks run on my Windows host with an AMD GPU, only one had an error while computing.
Only one has validated. That one had a quorum partner that was also a Windows host with an AMD GPU.
There was a pretty material productivity improvement from running 2X over 1X. I suspect that 3X would give enough more to make me choose that. GPU utilization at 2X was in the low 90s.
I also got some "new" BRP7-cuda55-WUs. On my machine with the RTX 3090, it seems it doesn't use the GPU, but only runs on the CPU. GPU-Utilisation is between 1% and 10%, runtime until now about 23.5 minutes and about 15% progress.
[/quote]
The WUs seem to be stuck. They're at 97/98 % after nine hours, further progress very very slow. Still no GPU-usage.
Since Einstein task list doesn't allow filtering by app, I'm finding it hard to see mine. Most of them seemed to work, but quite a few (maybe 5-10%) seemed to be taking way too long so I aborted them. The first valid MeerKAT I can see in the list (so presumably the latest in time) validated using my AMD 3GB and someone else's AMD 2GB. The first pending I can see in the list errored on 4 different Nvidias, and my AMD is waiting for it to be resent again.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I just tried to get MeerKAT
)
I just tried to get MeerKAT for a machine which had come out of daily room-heating shutdown without success. On looking at the applications page, I noticed that the BRP7 section is empty at the moment.
I imagine enough has been learned to initiate one or more adjustments before further deployment.
I apologise on behalf of the
)
I apologise on behalf of the idiots that program the Boinc scheduler. I've gone through this severe bug with them before on github and they never fixed it, Richard I think you were involved. But randomly Boinc will download way too many tasks. Out of the 1100 in progress BRP7 tasks on the status page, 630 of them are on one of my computers. I told it to get a 1 day queue, instead it's got what it believes is an 18 day queue, pretty daft since it breaks both my request and the deadline by a factor of 18! I shall abort a fair number and let the rest of you take some.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I'm awaiting, with bated
)
I'm awaiting, with bated breath, the outcome of this workunit as it compares both ATI/NVIDIA ( OpenCL/CUDA ) and Linux/Windows.
Cheers, Mike.
( edit ) I could start a book on it : say 2 to 1 ( ie. bet 2 to win 1 ) that it goes to a tie with a third machine being called in to decide the quorum ... :-)
( edit ) It's also an Intel/AMD CPU compare for that matter.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I have some interesting
)
I have some interesting observations.
I wanted to test out an actual legacy GPU, to see how a CUDA 5.5 era GPU would actually fare with these tasks since the project claimed to want to keep such an old CUDA version for compatibility with older GPUs. to test this, I used the v0.03 CUDA55 an v0.01 OpenCL Linux applications that both showed to process successfully (that I saved offline before they were pulled). I formed working app_info files so that I was able to download fresh tasks without an app version from the project. both app_info files proved to successfully download and process tasks for a modern 3080Ti/Ubuntu22.04/Kernel5.15 system with modern drivers.
the Test System is Ubuntu 20.04.4, 5.15 kernel, X79 platform with an E5-1680v2 Xeon CPU, a GTX 550Ti and nvidia 390.154 drivers installed from the nvidia proprietary driver package. the 340/384 drivers would not install (guessing a kernel incompatibility).
attempting to run either the v0.03 cuda55 app or the v0.01 opencl app on the 550Ti, resulted in immediate failures. however, swapping in a GTX 1060, with the exact same drivers, processing worked. so it's not the fault of the drivers, and I have no reason (at the moment) to believe the hardware itself is faulty.
even though the 390 drivers support this GPU for use as a display, it seems they don't support CUDA on this GPU anymore. a little digging showed that CC 2.1 (Fermi generation) lost CUDA support in CUDA SDK 9.0 and after, meaning an older version of the driver (and kernel for compatibility) will be necessary for any CUDA processing with these old GPUs. I will try Ubuntu 18.04 and the latest 340.108 drivers.
another interesting observations comes from the successful runs of these new tasks with v0.03 CUDA on a 3080Ti. these appear to be full size tasks. and processing took 6 minutes at 1x, and 11 minutes at 2x. only a slight speed advantage to running 2x it seems. about 5x longer running than the FGRPB1G tasks on the same hardware.
_________________________________________________________________________
Mike Hewson wrote: I could
)
Actually if I placed a 2 to 1 bet with you, I'd expect to bet 1 to win 2.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Ian&Steve C. wrote:even
)
Ouch. Nvidia never were good with software. When I download software for an old AMD, I'm given the latest version that supported that card, not a newer version which half supports it.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Dont know if this helps. But
)
Dont know if this helps. But 7 up, and 7 errors. 14 in queue.
Of eight MeerKAT v0.05 tasks
)
Of eight MeerKAT v0.05 tasks run on my Windows host with an AMD GPU, only one had an error while computing.
Only one has validated. That one had a quorum partner that was also a Windows host with an AMD GPU.
There was a pretty material productivity improvement from running 2X over 1X. I suspect that 3X would give enough more to make me choose that. GPU utilization at 2X was in the low 90s.
[quote=Ereignishorizont]I
)
[quote=Ereignishorizont]
I also got some "new" BRP7-cuda55-WUs. On my machine with the RTX 3090, it seems it doesn't use the GPU, but only runs on the CPU. GPU-Utilisation is between 1% and 10%, runtime until now about 23.5 minutes and about 15% progress.
[/quote]
The WUs seem to be stuck. They're at 97/98 % after nine hours, further progress very very slow. Still no GPU-usage.
Since Einstein task list
)
Since Einstein task list doesn't allow filtering by app, I'm finding it hard to see mine. Most of them seemed to work, but quite a few (maybe 5-10%) seemed to be taking way too long so I aborted them. The first valid MeerKAT I can see in the list (so presumably the latest in time) validated using my AMD 3GB and someone else's AMD 2GB. The first pending I can see in the list errored on 4 different Nvidias, and my AMD is waiting for it to be resent again.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.