So, somewhere around June 2016, BOINC GPU crunching on my system broke. It had been working between then and my previous post on this list. It coincided with MESA/LLVM changes and Fedora 24.
It's almost a year later and BOINC/E@H + ATi/Radeon + MESA/OpenCL is still broken. Has anyone been able to restore GPU crunching? I even have a Radeon X, now. I remember reading something about kernel 4.10 having important AMDGPU updates, but that might not be all that is necessary. I'm getting Computation Error (SEGV) immediately.
Yeah, okay, thanks. But that's not this thread. I know people have talked about that here, but that's not what I'm asking.
Since I switched to radeon-si 1.5 years ago, my system has been a lot more stable. Plus, the proprietary drivers are sort of a pain. I used them for probably 14 or 15 years though, so, I realize I could go back to doing that. But, after having everything working correctly on the OSS driver, it's even less enticing, now.
The backtrace is very similar to what it was 5 months ago, but now I'm using a completely different card and driver. I think this means that the problem must be Mesa. Agreed? Does anyone know if this is being tracked somewhere or when Mesa will get a patch for this?
It looks like Mesa 17 just hit update repos this week, so I'm going to update.
Okay, so now I have Mesa 17 and still get jobs failing with computation failure. I'm thinking the problem lies in LLVM? Does anyone agree?
LLVM 3.9.1
My AMDGPU version is xorg-x11-drv-AMDGPU 1.3, but I don't think I can do better than that right now.
Also, this error is for FGRP, not BRP6. Should I open a new thread about that, or is this good enough. I'm assuming this a similar enough problem to that for the old BRP6 that it's okay to keep it here.
I have a similar problem, maybe its related to this discussion, so let me explain here,
I´m using Ubuntu 16.04 64-bit, with Mesa 17.0.7 and BOINC 7.6.33. thankfully, i have been able to set up BOINC so that it detects my GPU using open source mesa drivers (my GPU is an AMD A6-5200 w/ HD 8400) as OpenCL capable.
The thing is *every* task ends with "compute error", and im very pissed off.....
Thanks for posting. That's my problem too! I've been fighting it for over a year. :-(
Is thread is for the BRP6 app, which is old and you probably are not running that app, so your errors are not a result of problems with that app. As you see, they told me to open a new thread for my problem. You can find out exactly what app caused the failure by checking your work unit reports on-line. There are several apps in the E@H set. You should do this anyway since you'll surely be asked more about your error and the error is most easily found in the work unit report on-line. Go to your E@H account and look for the "tasks" link.
Then, post to my thread for FGRPopencl1K-ati as I suspect that one of your apps that is failing, just like for me.
Hey all, So, somewhere
)
Hey all,
So, somewhere around June 2016, BOINC GPU crunching on my system broke. It had been working between then and my previous post on this list. It coincided with MESA/LLVM changes and Fedora 24.
It's almost a year later and BOINC/E@H + ATi/Radeon + MESA/OpenCL is still broken. Has anyone been able to restore GPU crunching? I even have a Radeon X, now. I remember reading something about kernel 4.10 having important AMDGPU updates, but that might not be all that is necessary. I'm getting Computation Error (SEGV) immediately.
26-Apr-2017 22:23:48 [---] OpenCL: AMD/ATI GPU 0: AMD POLARIS10 (DRM 3.9.0 / 4.10.10-200.fc25.x86_64, LLVM 3.9.1) (driver version 13.0.4, device version OpenCL 1.1 Mesa 13.0.4, 8190MB, 8190MB available, 3709 GFLOPS peak)I don't use Mint but their
)
I don't use Mint but their forums seem to suggest the AMD gpupro driver is workable with some small effort.
Yeah, okay, thanks. But
)
Yeah, okay, thanks. But that's not this thread. I know people have talked about that here, but that's not what I'm asking.
Since I switched to radeon-si 1.5 years ago, my system has been a lot more stable. Plus, the proprietary drivers are sort of a pain. I used them for probably 14 or 15 years though, so, I realize I could go back to doing that. But, after having everything working correctly on the OSS driver, it's even less enticing, now.
The backtrace is very similar to what it was 5 months ago, but now I'm using a completely different card and driver. I think this means that the problem must be Mesa. Agreed? Does anyone know if this is being tracked somewhere or when Mesa will get a patch for this?
It looks like Mesa 17 just hit update repos this week, so I'm going to update.
Okay, so now I have Mesa 17
)
Okay, so now I have Mesa 17 and still get jobs failing with computation failure. I'm thinking the problem lies in LLVM? Does anyone agree?
LLVM 3.9.1
My AMDGPU version is xorg-x11-drv-AMDGPU 1.3, but I don't think I can do better than that right now.
Also, this error is for FGRP, not BRP6. Should I open a new thread about that, or is this good enough. I'm assuming this a similar enough problem to that for the old BRP6 that it's okay to keep it here.
Probably would be a good idea
)
Probably would be a good idea to have your own thread. I saw a new post here and immediately thought that BRP6G has been dead for awhile.
Anyway, I setup a RX580 in Ubuntu 16.04 with AMDGPU-Pro drivers for the Pentathlon:
https://einsteinathome.org/host/12530808/tasks
Hello all, I have a
)
Hello all,
I have a similar problem, maybe its related to this discussion, so let me explain here,
I´m using Ubuntu 16.04 64-bit, with Mesa 17.0.7 and BOINC 7.6.33. thankfully, i have been able to set up BOINC so that it detects my GPU using open source mesa drivers (my GPU is an AMD A6-5200 w/ HD 8400) as OpenCL capable.
The thing is *every* task ends with "compute error", and im very pissed off.....
What can i do?
Thanks in advance!
Hi ELHIOR, Thanks for
)
Hi ELHIOR,
Thanks for posting. That's my problem too! I've been fighting it for over a year. :-(
Is thread is for the BRP6 app, which is old and you probably are not running that app, so your errors are not a result of problems with that app. As you see, they told me to open a new thread for my problem. You can find out exactly what app caused the failure by checking your work unit reports on-line. There are several apps in the E@H set. You should do this anyway since you'll surely be asked more about your error and the error is most easily found in the work unit report on-line. Go to your E@H account and look for the "tasks" link.
Then, post to my thread for FGRPopencl1K-ati as I suspect that one of your apps that is failing, just like for me.
https://einsteinathome.org/content/fgrpopencl1k-ati-polaris10-amdgpu-llvm391-mesa-17-crash
Hello PAUL, i did it,
)
Hello PAUL,
i did it, thank you so much!