Gee I actually only had 3 Intel iMacs in history : first generation from 2007 to 2010, 2nd gen from 2010 to 2020 and x gen (my current) since 2020 until now (and don't plan to change in any near future). AFAIR it has always been AMD GPU (not sure for the 1st iMac though, too old)
@CpgAcw I can see your Intel machine actually has the Intel integrated GPU "declared" (and nothing alse) but the tasks this host is crunching are only Gamma-ray pulsar search #5 (FGRP5) = CPU and not GPU. Your M4 host is also crunching Binary Radio Pulsar Search (Arecibo,GBT,A) v2.08 using apple_gpu-opencl.
But as I said in April I already switched to "GPU friendly" apps for my AMD, not much choice now : amicable or primegrid AFAIK.
Not sure why the i3 mini did not crunch gpu tasks. I never looked into it because I only had it online to get my RAC going when I came back to Boinc after many years away.
Probably because it only has 1.5 GB of video memory.
The M4 machine uses "unified" memory so it has more options for the OS to run things.
The i3 machine is currently offline. I will probably repurpose it to Asteroids@Home cpu tasks when I finish rebuilding my "fleet" onto proper shelving and reworking my lan.
This is @CpgAcw. I had to temporarily change my ID for about 24 hours for another BOINC activity to find me. My actual ID is Phil with the Marvin the Martian avatar.
Oh you mean this "Mac OS X on Intel / 1.34 (opencl-intel_gpu-lion-Beta)" from 2021... I did not even know that there were some mac with Intel GPU chipset that could be used by boinc...
Some interesting info on O3 on my mac: I decided to test O3 again (the fact that the AF is going to pick it up as a back-up project for our forthcoming RAID has absolutely nothing to do with it, any connection would be purely coincidental and completely beyond my control :-D)
And things have changed ! Firstly I noticed that the ‘class plan’ has changed because in order to configure my app_config I had to change the GW-opencl-ati-2 to GW-opencl-ati-3 for it to work, but more importantly, to my great surprise, it started to use a huge amount of GPU resources: around 11-12%. That's fantastic, before it was 0. And to do this it needs about 1/2 thread of CPU all the time (without adding any app_config changes at first). So I've let it do a number of tasks like that since yesterday and it seems stable. They take about 4h40 on my venerable GPU.
So this morning I say to myself ‘let's get crazy’ and I turn my app_config back on (hence the find above) so that it does 3 of them at the same time: bad idea, GPU usage for each falls to less than 5% and 30% or less of CPU for one thread.
So now I've set both the CPU and GPU parameters to 0.5 in the app_config, I'm doing 2 tasks at the same time, GPU usage is back up to around 7-8% (and sometimes a bit more) each and CPU to just under 1/2 thread each, so overall I'm using a bit more GPU. It's not crazy, but it's a lot better than 0.
And there's a good reason behind it: I realise a new version of the app was released recently
Mac OS 10.5 or later running on Intel 64 Bit
Version 1.16 (GW-opencl-ati-3)
20 Jan 2025, 14:39:38 UTC
So it's still not fantastic, but maybe on a more recent machine + GPU it performs much better ?
But in any case I'm really pleased that the developer has shaken things up and got things moving!
Gee I actually only had
)
Gee I actually only had 3 Intel iMacs in history : first generation from 2007 to 2010, 2nd gen from 2010 to 2020 and x gen (my current) since 2020 until now (and don't plan to change in any near future). AFAIR it has always been AMD GPU (not sure for the 1st iMac though, too old)
@CpgAcw I can see your Intel machine actually has the Intel integrated GPU "declared" (and nothing alse) but the tasks this host is crunching are only Gamma-ray pulsar search #5 (FGRP5) = CPU and not GPU. Your M4 host is also crunching Binary Radio Pulsar Search (Arecibo,GBT,A) v2.08 using apple_gpu-opencl.
But as I said in April I already switched to "GPU friendly" apps for my AMD, not much choice now : amicable or primegrid AFAIK.
Not sure why the i3 mini did
)
Not sure why the i3 mini did not crunch gpu tasks. I never looked into it because I only had it online to get my RAC going when I came back to Boinc after many years away.
Probably because it only has 1.5 GB of video memory.
The M4 machine uses "unified" memory so it has more options for the OS to run things.
The i3 machine is currently offline. I will probably repurpose it to Asteroids@Home cpu tasks when I finish rebuilding my "fleet" onto proper shelving and reworking my lan.
This is @CpgAcw. I had to temporarily change my ID for about 24 hours for another BOINC activity to find me. My actual ID is Phil with the Marvin the Martian avatar.
Phil
I thought I was wrong once, but I was mistaken.
Phil wrote:Not sure why the
)
Probably because there is no Einstein app able to process on the Intel integrated GPU stuff, is there ?
[AF>Le_Pommier wrote:
)
there is a MacOS Intel GPU app for the BRP4 project, but not for O3AS.
_________________________________________________________________________
Oh you mean this "Mac OS X on
)
Oh you mean this "Mac OS X on Intel / 1.34 (opencl-intel_gpu-lion-Beta)" from 2021... I did not even know that there were some mac with Intel GPU chipset that could be used by boinc...
Some interesting info on O3
)
Some interesting info on O3 on my mac: I decided to test O3 again (the fact that the AF is going to pick it up as a back-up project for our forthcoming RAID has absolutely nothing to do with it, any connection would be purely coincidental and completely beyond my control :-D)
And things have changed ! Firstly I noticed that the ‘class plan’ has changed because in order to configure my app_config I had to change the GW-opencl-ati-2 to GW-opencl-ati-3 for it to work, but more importantly, to my great surprise, it started to use a huge amount of GPU resources: around 11-12%. That's fantastic, before it was 0. And to do this it needs about 1/2 thread of CPU all the time (without adding any app_config changes at first). So I've let it do a number of tasks like that since yesterday and it seems stable. They take about 4h40 on my venerable GPU.
So this morning I say to myself ‘let's get crazy’ and I turn my app_config back on (hence the find above) so that it does 3 of them at the same time: bad idea, GPU usage for each falls to less than 5% and 30% or less of CPU for one thread.
So now I've set both the CPU and GPU parameters to 0.5 in the app_config, I'm doing 2 tasks at the same time, GPU usage is back up to around 7-8% (and sometimes a bit more) each and CPU to just under 1/2 thread each, so overall I'm using a bit more GPU. It's not crazy, but it's a lot better than 0.
And there's a good reason behind it: I realise a new version of the app was released recently
Mac OS 10.5 or later running on Intel 64 Bit
Version 1.16 (GW-opencl-ati-3)
20 Jan 2025, 14:39:38 UTC
So it's still not fantastic, but maybe on a more recent machine + GPU it performs much better ?
But in any case I'm really pleased that the developer has shaken things up and got things moving!
[AF>Le_Pommier wrote:
)
It's not a new version. it's the same old version, just re-released with new config files.
the difference in behavior is due to the app responding differently to the new input data and associated command line arguments.
the application itself is unchanged since 1.5 years ago.
_________________________________________________________________________