I'm sorry, I don't know the answer, but I can tell you that I've been crunching with BOINC before LLVM4.0 on my Fedora system, so I'm not sure you need the latest libclc to get what you want. If libclc source says it needs 4.0, then, I assume, that's correct. If you can't find 4.0, I agree, that's odd, but it must be somewhere.
I have libclc-0.2.0-3. I know there was a problem this a while back: after a mesa upgrade, libclc couldn't complie things and needed a patch. I don't know if that was added by Fedora maintainers or not, but I'm sure it made it upstream by now. I've been crunching for about a month or so since the last flurry of problems and I have only that libclc and LLVM3.9 It could be that more of that sort of thing is happening, though, and things are out of sync, again.
I also did the following, but not sure if its necessary
$ sudo usermod -a -G video YourUserName
This add your username to the video group
If you reboot and try to run an Einstein ATI GPU application, it will still probably fail. In my case this fail origined in that the application tried the mesa-opencl firstly. The key is the to get rid of this failing driver.
$ sudo apt purge mesa-opencl-icd
does the trick. After rebooting, my Boinc Einstein ATI GPU applications start running smootly.
I got my AMD APU A10-7850K to work with Einstein ATI GPU applications on Ubunt 16.04 64 bit.
This is what I did:
First, download the AMDGPU-Pro Driver and install.
Good news Bent, it appears what you have done is uninstalled (i guess after installing?) the Open source mesa drivers (thread topic) and then installed the AMDGPU-Pro driver.
I got my AMD APU A10-7850K to work with Einstein ATI GPU applications on Ubunt 16.04 64 bit.
This is what I did:
First, download the AMDGPU-Pro Driver and install. You find the driver at
interesting that the AMDGPU-Pro Driver works on ur APU, as it shouldnt according to AMD...i've got a r7-250 and a r9-270..maybe i should give it a shot installing the amdgpu-Pro driver then. How's the performance compared to Windows or old fglrx ?
As this thread started about mesa driver (open source) and this is more about the amdgpupro driver (closed source) i'm going to put a reply to this in the amdgpu-pro thread
Yes, that is the machine. No invalids yet. I am running two cpu applications concurently to the gpu app. Tournaround is about 10000 seconds for the gpu app. I cannot compare it to my previous solution. Both app and data has changed. However, my felling is that the speed is deasant taking into account what kind of processor it is. The combined score i comparable to my 4 core/8 thread Xenon with pure cpu applications.
Does your GPU turnaround time decrease if you reduce the CPU apps to 1?
10000s per WU sounds a bit long for me. My Core i5 4xxx hosts on Windows do 2 concurrent GPU tasks in circa 17000s. That is roughly 8500s for one WU and thus quite a bit faster than your GPU. On paper, your GPU is twice as fast as the HD 4600 in the Core i5! In addition, my hosts run some other CPU-heavy non-BOINC tasks, which may reduce GPU performance...
Yes, that is the machine. No invalids yet. I am running two cpu applications concurently to the gpu app. Tournaround is about 10000 seconds for the gpu app.
i agree with matman, its slower than i expected. 4 ur comparison- i have an a8-7600 running under Windows that takes about 2 hours/WU (BRP4G)
i'll start testing this evening
p.s.: tested on a8-7600.. and its working with amdgpu-pro ! +Its quite fast..Bent maybe u should deactivate cpu-WUs..here is the first 1..took about 6000s- not validated yet
I'm sorry, I don't know the
)
I'm sorry, I don't know the answer, but I can tell you that I've been crunching with BOINC before LLVM4.0 on my Fedora system, so I'm not sure you need the latest libclc to get what you want. If libclc source says it needs 4.0, then, I assume, that's correct. If you can't find 4.0, I agree, that's odd, but it must be somewhere.
I have libclc-0.2.0-3. I know there was a problem this a while back: after a mesa upgrade, libclc couldn't complie things and needed a patch. I don't know if that was added by Fedora maintainers or not, but I'm sure it made it upstream by now. I've been crunching for about a month or so since the last flurry of problems and I have only that libclc and LLVM3.9 It could be that more of that sort of thing is happening, though, and things are out of sync, again.
I got my AMD APU A10-7850K to
)
I got my AMD APU A10-7850K to work with Einstein ATI GPU applications on Ubunt 16.04 64 bit.
This is what I did:
First, download the AMDGPU-Pro Driver and install. You find the driver at
http://support.amd.com/en-us/kb-articles/Pages/AMD-Radeon-GPU-PRO-Linux-Beta-Driver%E2%80%93Release-Notes.aspx
I also did the following, but not sure if its necessary
$ sudo usermod -a -G video YourUserName
This add your username to the video group
If you reboot and try to run an Einstein ATI GPU application, it will still probably fail. In my case this fail origined in that the application tried the mesa-opencl firstly. The key is the to get rid of this failing driver.
$ sudo apt purge mesa-opencl-icd
does the trick. After rebooting, my Boinc Einstein ATI GPU applications start running smootly.
I hope this may help some of you.
Best regards
Bent, Oslo, Norway
Bent Vangli wrote:I got my
)
Good news Bent, it appears what you have done is uninstalled (i guess after installing?) the Open source mesa drivers (thread topic) and then installed the AMDGPU-Pro driver.
Very interesting that the Spectre APU is working.
This host https://einsteinathome.org/host/11290193 ?
Any invalids? I'm finding running Beta tasks give more invalids than non-Beta.
Bent Vangli wrote:I got my
)
interesting that the AMDGPU-Pro Driver works on ur APU, as it shouldnt according to AMD...i've got a r7-250 and a r9-270..maybe i should give it a shot installing the amdgpu-Pro driver then. How's the performance compared to Windows or old fglrx ?
thx in advance
As this thread started about
)
As this thread started about mesa driver (open source) and this is more about the amdgpupro driver (closed source) i'm going to put a reply to this in the amdgpu-pro thread
Yes, that is the machine. No
)
Yes, that is the machine. No invalids yet. I am running two cpu applications concurently to the gpu app. Tournaround is about 10000 seconds for the gpu app. I cannot compare it to my previous solution. Both app and data has changed. However, my felling is that the speed is deasant taking into account what kind of processor it is. The combined score i comparable to my 4 core/8 thread Xenon with pure cpu applications.
Regards, Bent
Does your GPU turnaround time
)
Does your GPU turnaround time decrease if you reduce the CPU apps to 1?
10000s per WU sounds a bit long for me. My Core i5 4xxx hosts on Windows do 2 concurrent GPU tasks in circa 17000s. That is roughly 8500s for one WU and thus quite a bit faster than your GPU. On paper, your GPU is twice as fast as the HD 4600 in the Core i5! In addition, my hosts run some other CPU-heavy non-BOINC tasks, which may reduce GPU performance...
Bent Vangli wrote:Yes, that
)
i agree with matman, its slower than i expected. 4 ur comparison- i have an a8-7600 running under Windows that takes about 2 hours/WU (BRP4G)
i'll start testing this evening
p.s.: tested on a8-7600.. and its working with amdgpu-pro ! +Its quite fast..Bent maybe u should deactivate cpu-WUs..here is the first 1..took about 6000s- not validated yet
https://einsteinathome.org/workunit/264198330
Happyl Good ide. I will try
)
Happyl
Good ide. I will try reducing CPU apps, first to one and detect speed and then to zero. I will then report back, probably after 3-4 days.
Bent :-)
Ok, as promissed above: With
)
Ok, as promissed above:
With one CPU app, each GPU work unit about 8350 seconds.
With none CPU app, each GPU work unit about 7200 seconds.
Best regards
Bent, Oslo, Norway