Hello, is this app still considered to be a beta test, comparing results from cpu version? I wonder when we get a gamma ray pulsar discovery through GPU app. Regards.
Hello, is this app still considered to be a beta test, comparing results from cpu version?
No, not since the CPU app versions vanished from the FGRPB1G application months ago.
Basically both applications are doing the same search on different parts of the parameter space. The workunit's "bandwidth" (basically FFT size) is tuned for the different processing units, though.
Quote:
I wonder when we get a gamma ray pulsar discovery through GPU app.
So do we. But in the current "binary" search it's really completely random whether we'll get a discovery from the CPU or the GPU part of the search (first).
We here do have a lot of GPUs that we can't use because of lack of x86-windows-version-GPU-app. Is it possible to feed use something with at least one type of app?
Apologies if this has been answered recently. I've been away for a while.
Will there be a CUDA version of this app in the future? I am currently not running Einstein because the OpenCL app uses a full CPU thread per task on my Nvidia cards, which takes away from my ability to run projects on my CPU as well.
Go back about 10 messages in this thread and find this message.
Considering that that report is several months old and didn't address any planned attempts to resolve the issue I think asking again now is reasonable. The initial CUDA port was a performance failure; but we have no way of knowing if that resulted in shelving it temporarily or indefinitely; or if work has been ongoing to try and fix it. And if so, has any progress been made.
Anyone else seeing a high
)
Anyone else seeing a high number of "Computation error" results on 1.20?
Jim check your
)
Jim check your antivirus..
First couple of things in stderr is "network access denied"
Next one "can't find" which usually means anti virus quarantined it or remeoved that file
Zalster wrote:Jim check your
)
Yeah, I saw those. The network denied message would make sense only if it was running between 0500-0700 mdt, as I disable network access at that time.
The not found is strange, as there was no AV action logged (only thing on the crunchers is Defender).
Subsequent jobs are succeeding, so perhaps it was a hiccup.Almost looks as the the PC got dumb for a bit (hope it's not another SSD preparing to die)
Thanks for the note!
Zalster wrote:Next one "can't
)
After looking a bit closer, it appears to me that the files it could not open were output files, not the DAT input file.
Hello, is this app still
)
Hello, is this app still considered to be a beta test, comparing results from cpu version? I wonder when we get a gamma ray pulsar discovery through GPU app. Regards.
Arif Mert Kapicioglu
)
No, not since the CPU app versions vanished from the FGRPB1G application months ago.
Basically both applications are doing the same search on different parts of the parameter space. The workunit's "bandwidth" (basically FFT size) is tuned for the different processing units, though.
So do we. But in the current "binary" search it's really completely random whether we'll get a discovery from the CPU or the GPU part of the search (first).
BM
We here do have a lot of GPUs
)
We here do have a lot of GPUs that we can't use because of lack of x86-windows-version-GPU-app. Is it possible to feed use something with at least one type of app?
Apologies if this has been
)
Apologies if this has been answered recently. I've been away for a while.
Will there be a CUDA version of this app in the future? I am currently not running Einstein because the OpenCL app uses a full CPU thread per task on my Nvidia cards, which takes away from my ability to run projects on my CPU as well.
Go back about 10 messages in
)
Go back about 10 messages in this thread and find this message.
Cheers,
Gary.
Gary Roberts wrote:Go back
)
Considering that that report is several months old and didn't address any planned attempts to resolve the issue I think asking again now is reasonable. The initial CUDA port was a performance failure; but we have no way of knowing if that resulted in shelving it temporarily or indefinitely; or if work has been ongoing to try and fix it. And if so, has any progress been made.