We do now have an application version (FGRPB1G) that uses the Apple M GPU (via OpenCL). The initial part of the computation requires double precision, so this is done on the CPU.
Also too, currently, the application is built for the x86_64-apple-darwin platform. We may optimize that in the future. However, this app was mostly made to get our server code to deliver Apple M GPU apps.
My 8 core m1 gpu takes about 30min for a job while burning about 6 watts of power, My nvidia T600 gpu's (in my old work windows pc's) takes around 35-40 min and uses about 35 watts of power (according to my killawatt plug meter)
Will be interesting to see the results from other m1/m2 gpu core counts.
I've been running the beta for macOS Sonoma, and apparently it is not compatible. I keep getting a computation error immediately after the task starts. See https://einsteinathome.org/task/1520606259 as an example. Disappointing! Any suggestions or thoughts? Hopefully once Sonoma comes out, this can be resolved.
understand, thanks for the
)
understand, thanks for the update nonetheless :)
We do now have an application
)
We do now have an application version (FGRPB1G) that uses the Apple M GPU (via OpenCL). The initial part of the computation requires double precision, so this is done on the CPU.
Also too, currently, the application is built for the x86_64-apple-darwin platform. We may optimize that in the future. However, this app was mostly made to get our server code to deliver Apple M GPU apps.
BM
I've seen a few tasks pop up
)
I've seen a few tasks pop up and they do compute noticeably faster - thanks for the update !
+1
)
+1
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Thanks Bernd, Really pleased
)
Thanks Bernd, Really pleased with the app update.
My 8 core m1 gpu takes about 30min for a job while burning about 6 watts of power, My nvidia T600 gpu's (in my old work windows pc's) takes around 35-40 min and uses about 35 watts of power (according to my killawatt plug meter)
Will be interesting to see the results from other m1/m2 gpu core counts.
Great news, seems to be
)
Great news, seems to be running well!
100 completed with 8 invalid
M1 Max with 32-core GPU - around 30 minutes to complete
Thanks for building this - I've been running GRPS5 for M1 on 5 cores, and now the throughput has doubled with the addition of the GPU
I've been running the beta
)
I've been running the beta for macOS Sonoma, and apparently it is not compatible. I keep getting a computation error immediately after the task starts. See https://einsteinathome.org/task/1520606259 as an example. Disappointing! Any suggestions or thoughts? Hopefully once Sonoma comes out, this can be resolved.
The task in not getting the
)
The task is not getting the necesssary permissions to access the OpenCL resources in the host and fails to compile the application.
Likely the resource locations got moved in the beta OS from where the stock application expects them for the previous OS version.
Is Sonoma the release that
)
Is Sonoma the release that finally removes OpenCL?
As FGRPB1G will run out of
)
As FGRPB1G will run out of avaiable work soon, we built an arm64 App for BRP4G that uses the Apple_M GPU via OpenCL.
Indeed it seems that there are problems with our OpenCL code on Sonoma. I'll ask around what to do about this.
BM