So, as ATi GPU users will know, CAL was removed from the latest stable Catalyst drivers (for Linux). Yesterday, all of a sudden, my ATi GPUs started crunching SETI@home work again. I couldn't believe it.

Now, I see an app called SETI@home v7 7.04 (opencl_ati5_linux_nocal). I'm correct in saying this is a new app, right?

You can see the apps here:
http://setiathome.berkeley.edu/apps.php

Date for opencl_ati5_linux_nocal is 30 May 2013 but I don't know if it is correct.
It may be new plan_class (new for Linux)
(i.e. it may be old app but new plan_class)

In fact the app (executable) is the same for all these plan classes (or maybe 2 variants for HD 4xxx and HD 5xxx+ ?):
opencl_ati5_linux
opencl_ati5_linux_nocal
opencl_ati_linux
opencl_ati_linux_nocal


Read some (5-10) posts here:
http://setiathome.berkeley.edu/forum_thread.php?id=77102&postid=1685088#1685088

The verdict/fix (28 May 2015): "Eric found a typo in the plan class specification file"
http://setiathome.berkeley.edu/forum_thread.php?id=77102&postid=1685215#1685215

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

I've just added an ATI graphics card to my AMD A8 APU based Linux box (running Mint 17.1 64 bit), and finally managed to get the drivers updated to ATI Catalyst ones so BOINC recognises it, and SETI has given me a few ATI tasks. The ones using application SETI@home v7 7.04 (opencl_ati5_linux) all ran fine. However, I got three allocated to SETI@home v7 7.04 (opencl_ati_linux) which are behaving oddly. Each one runs for 21 seconds elapsed time, about 0.016% progress, then goes to status "scheduler wait" for ten seconds or so, then resets itself to zero percent and zero elapsed time and starts again. Estimated time remaining is not going down. I've suspended them for the time being as they don't seem to be getting anywhere, and stopped SETI from giving me new tasks until I work out what's going on. Should I abort them, or is there something I need to do to make them work?

Hi BilBg,

Well, that explains it, for sure. Thanks for the references.

A typo kept me from making it near the top. Ouch.

(As for the new question about resetting workunits, all I can say is that I had something like that happen in the past, but i don't remember what I did to fix it. I think I posted to this form, so search around more or post a new thread; no one is going to see it here in this one.)

Thanks Porcelain Mouse, I think you're right. :-)

The "search forums" didn't find any matches on it for me, so I think I may be the first to have my particular problem. For the time being, I've aborted all the looping tasks and will wait and see if it gets any more of them. So far, nothing further. Cheers!

Developers/programmers of GPU apps (CUDA, OpenCL) read and post in Number Crunching

You are not "the first to have" "Scheduler wait", e.g.:
http://setiathome.berkeley.edu/forum_thread.php?id=75005

I don't know the cause of your "particular problem" (maybe because of only 256MB GPU memory or the driver version?) so better ask in Number Crunching

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Hmm, ok. I'll try there next. Obviously my forum searches aren't quite right either!

Your last comment about me only having 256Mb on board has highlighted a problem though - the card is a HD6670 with 2Gb of RAM, but Catalyst Control Centre (and BOINC) recognise it as a HD6550D with only 256Mb. I'd just assumed they were the same thing, but hadn't spotted the RAM difference. Reading up on it, they're not the same. I probably have the wrong driver installed I guess, which is a pain as it took me ages to install the thing. I'll do some more searching the web on that line of enquiry before posting back on the Number Crunching page, if I even need to - I suspect if I get the drivers & CCC to report it as a HD6670 like it's supposed to, BOINC may well be a bit happier. Thanks for pointing this out though.

Sorted my problem out now. I've removed the GPU and am back to plain old APU processing with the default xorg xserver display driver. Screen is the right size again, case temp has dropped back to normal, I'm happy, BOINC is happy. Moral of the story - if it ain't broke, don't fix it. Will never try adding a GPU again, more hassle than it's worth! Cheers.