No,

I'm not too bothered about those - the incidence rate is low and it always happens very soon after the task's started so little effort is wasted.

Matt

After many WUs errored I reset my Radeon 7950 (100% stable on folding@home @1005MHz, watercooled, no overvolt) back to its factory clock of 880MHz and now it magically seems to work without error on GPUGRID 9.17(mt).
(AthlonX4 610e 2.4GHz, four CPUs used, 54% GPU load).

Edit:"without error" could be too much, but it could, at least once, work...

---

What is your avg PPD in f@h with this card and clock?

2 (old) errors with 9.15, 111 valids with 9.17.

HD7950 boost core clock@1000Mhz, undervolted, running very smooth, 7 cpu threads (8 available), gpu usage 58-59 percent, gpu 50-52 degrees Celsius, idle-BOINC@90 percent cpu usage (57-174) +118 watts.

Congratulation to this well-oiled Beta AMD support ! Wake up from hibernate with no lags from checkpointing.

I would say about 83'000 ppd on folding@home.
http://folding.extremeoverclocking.com/user_summary.php?s=&u=654291

---

Strange behaviour:
1) My client request only work for cpu, but i download gpu amd app
2) My client says it's using 6 cpu core and not an amd gpu (like, for example poem): (1cpu+1gpu) in boinc manager
3) Gpu-z gives me 90% gpu load, so app is using gpu[/img]

Strange behaviour:
1) My client request only work for cpu, but i download gpu amd app
2) My client says it's using 6 cpu core and not an amd gpu (like, for example poem): (1cpu+1gpu) in boinc manager
3) Gpu-z gives me 90% gpu load, so app is using gpu[/img]

But all expected and correct.
For technical reasons we have to deploy this as appearing to BOINC as CPU-only.

The old "CPU" app is gone now.
Matt

Matt,

Could you expand a little on those "technical reasons", please - primarily for the benefit of us BOINC-watchers in the community?

I'm familiar with Specifying plan classes in C++ and Specifying plan classes in XML: I also remember you saying that the BOINC server here doesn't have all the latest features, which may restrict the degree to which you can use the XML route.

But I think you're perhaps the first project developer who has attempted to combine the features of MT and ATI plan_class specifications. That could be useful to other projects as well, if the new gromacs version is a success.

Are you perhaps finding that this requirement is inhibiting you from combining the two features?

GPU plan classes
Note: plan classes containing the substrings 'nvidia', 'cuda', and 'ati' are reserved for GPU applications. Plan classes for AMD GPUs must contain the substring 'ati'; plan classes for NVIDIA GPUs much contain either 'nvidia' or 'cuda' as a substring; plan classes for OpenCL applications must include 'opencl' as a substring.

(from Plan classes)

If so, we ought to draw it to the BOINC developers' attention, so that the limitation can be lifted or worked round in the next release of BOINC server software. I'd be happy to try and help liaise with David Anderson, if that's the case.

Matt,

Could you expand a little on those "technical reasons", please - primarily for the benefit of us BOINC-watchers in the community?

Sure. Our BOINC scheduler is too old to understand the OpenCL plan class, and if we use the "ati" plan class the BOINC client rejects the application if the CAL runtime isn't available (which is the case with any new AMD driver set).

Eventually we'll upgrade the server, but that's not happening any time soon.

Matt

Known issues are:
* Checkpoint restart not working

Hope in the next version....

Known issues are:
* Checkpoint restart not working

Hope in the next version....

Actually it might be. give it a go and tell me.

Known issues are:
* Checkpoint restart not working

Hope in the next version....

Actually it might be. give it a go and tell me.

No.

Actually it might be. give it a go and tell me.

No.

+1. If i restart my pc, wus restart from 0%

Actually it might be. give it a go and tell me.

No.

+1. If i restart my pc, wus restart from 0%

Can you direct me to the output of a job where a restart has happened please?

Can you direct me to the output of a job where a restart has happened please?

This http://www.gpugrid.net/workunit.php?wuid=10639347 wu was interrupted after ~ 10% progress by stopping BM. After restart of BM it began again with 0%.
Wu will be finished in ~ 1.5 hrs from now.

Can you direct me to the output of a job where a restart has happened please?

This http://www.gpugrid.net/workunit.php?wuid=10639347 wu was interrupted after ~ 10% progress by stopping BM. After restart of BM it began again with 0%.
Wu will be finished in ~ 1.5 hrs from now.

wu finished and validated.
http://www.gpugrid.net/result.php?resultid=13808408

13808430

Initial situation: 2 cpu wus (wcg) on my 6 cores (4 free core) and no gpu wu
Start this gpu wu and it uses all cpu cores (stopping wcg wus).
I pause the wu (at 11%) and immediately the 2 cpu wus restart. After two minutes i restart the wu and it use only 4 cpu core and other two continue to run.
At 12% i close Boinc Manager and, after restart, the wu uses all 6 cpu core, but it's 0%

Do you need other test?

or about 260'000 ppd on Poem@home
with GPU load 85%

---

Any news about checkpoint??

Hello,
Working on a i7 3770 with a 6670 card.
It's using 8 CPU cores + GPU.
The most significent issue i have : I'm also running WorldCommunityGrid WU on this box and GPUGrid tasks pauses every other tasks to compute 1 GPRUGRID WU.
Then uploads / download new one.
Looks like all other tasks are just suspended forever and can't be turned on without suspending GPUGrid. I "solved" this by network limiting the computer so GPUGrip WU finishes then WCG WU can compute and then results are send (once every 24h)
Maybe i ve not well configured BOINC and preferences ... Not sure
(Not sure if it s really clear this explaination ....)