Raistmer wrote:

Also, any task result I looked into has many restarts. Try to be patient a little and not fiddle with settings. Allow few tasks to complete on their own, w/o restarts and re-sheduling between GPUs. Then provide links to their results on web page.

How much MB WUs/VGA card simultaneously? 3?

What in cmdline.txt file?

If I set -no_cpu_lock the above mentioned errors happens:
Task postponed: Suspicious spike results, host needs reboot or maintenance
...or...
Task postponed: Triplet data corruption, retry from checkpoint.

---

How abot Boinc processor usage settings? Use nn% of processors. Use at most n processors?

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

Task postponed: Triplet data corruption, retry from checkpoint.

So you need to check if AMD own OpenCL samples work well.

Leave app running in default regime w/o any additional settings.
Will it produce valid results in number being launched on all your GPUs?
Only when valid execution for all 4 GPUs with default settings will be proven firmly worth to try to speedup/improve things.

BTW, should this monster GPU support OpenCL 2.0 ?
I see only 1.2 support in task results...
Maybe, not optimal driver version for this GPU?

EDIT: not sure will you find OpenCL samples inside latest SDK. Maybe 2.0 only...
Then try older one:
http://developer.amd.com/tools-and-sdks/opencl-zone/amd-accelerated-parallel-processing-app-sdk/

Task postponed: Triplet data corruption, retry from checkpoint.

So you need to check if AMD own OpenCL samples work well.

Leave app running in default regime w/o any additional settings.
Will it produce valid results in number being launched on all your GPUs?
Only when valid execution for all 4 GPUs with default settings will be proven firmly worth to try to speedup/improve things.

Hm, it depend from which side I look - 'default'... ;-)

I set '0.33' in app_info.xml and '-no_cpu_lock -hp' in cmdline.txt. This would be 'default'?


BOINC say the VGA cards support OpenCL v2.0.

---

BOINC say the VGA cards support OpenCL v2.0.

-no_cpu_lock as default override
How it can be default if you provide option to override default settings???
Default means no options supplied.

From stderr:

Name: Fiji
Vendor: Advanced Micro Devices, Inc.
Driver version: 1800.8 (VM)
Version: OpenCL 1.2 AMD-APP (1800.8)

That can be issue...

Just sounds like more of the same to me;

...apparently there is a bug on the AMD driver and the R9 390X series of GPU start to produce a lot of error when you run 2 or more wu at a time.

I've gotten errors like that in the past when i've tried running multiple WUs per GPU

Seems Mike got a lot of Errors with his new ATI card as well, http://setiathome.berkeley.edu/results.php?hostid=5735690&state=6&appid=. Looks like the Errors have recently stopped, maybe he can suggest a fix...

BOINC say the VGA cards support OpenCL v2.0.

-no_cpu_lock as default override
How it can be default if you provide option to override default settings???
Default means no options supplied.

From stderr:

Name: Fiji
Vendor: Advanced Micro Devices, Inc.
Driver version: 1800.8 (VM)
Version: OpenCL 1.2 AMD-APP (1800.8)

That can be issue...

OK, but then all 12 GPU apps will be fixed at CPU-thread#0.
Or I should let run just 1 MB WU/GPU?

Maybe I should test the above mentioned AMD beta driver?

---

Yep. First to do is to separate completely mixed factors now.
CPU affinity behavior, multiple tasks per GPU behavior, multiple GPU per se behavior.

So, affinity locked as default, 1 task per GPU as default, await at least 10 tasks completions ON EACH of 4 GPUs before touch anything again.

Or I should let run just 1 MB WU/GPU?

yes.

Maybe I should test the above mentioned AMD beta driver?

maybe

So nocpulock fixes the affinity problem, then maybe you could try turning hyper threading back on

I remember getting triplet errors on catalyst 14.9 and running 2 wu per gpu so I could only run 2 wu per gpu with 14.4 with no errors. With newer drivers I could only run 1 wu per gpu without errors.

One choice is to use any of the newer drivers but run only one wu per gpu. It's good to make note of runtimes

I know it's an older driver, but if you want to run more than one wu per gpu I would try 14.4. Make note of runtimes because someone at einstein reported newer driver with one wu was more efficient than older driver with two wu. He didn't provide runtimes for 3 wu per gpu so that would require more testing.

You won`t get it working with this driver running multiple tasks.

Try just one task per GPU until a better driver is available.
Also use -no_CPU_lock in comandline to avoid CPU affinity mask 1.
One instance requires between 30% and 70% of a CPU core.

---

With each crime and every kindness we birth our future.

Raistmer wrote:

Yep. First to do is to separate completely mixed factors now.
CPU affinity behavior, multiple tasks per GPU behavior, multiple GPU per se behavior.

So, affinity locked as default, 1 task per GPU as default, await at least 10 tasks completions ON EACH of 4 GPUs before touch anything again.

Well results of...

VGA card 0
http://setiathome.berkeley.edu/result.php?resultid=4434235420
http://setiathome.berkeley.edu/result.php?resultid=4434235462
http://setiathome.berkeley.edu/result.php?resultid=4434235216
http://setiathome.berkeley.edu/result.php?resultid=4434235217
http://setiathome.berkeley.edu/result.php?resultid=4434235228
http://setiathome.berkeley.edu/result.php?resultid=4434235263
http://setiathome.berkeley.edu/result.php?resultid=4434235331
http://setiathome.berkeley.edu/result.php?resultid=4434235344
http://setiathome.berkeley.edu/result.php?resultid=4434235374
http://setiathome.berkeley.edu/result.php?resultid=4434235390

VGA card 1
http://setiathome.berkeley.edu/result.php?resultid=4434235415
http://setiathome.berkeley.edu/result.php?resultid=4434234398
http://setiathome.berkeley.edu/result.php?resultid=4434235235
http://setiathome.berkeley.edu/result.php?resultid=4434234742
http://setiathome.berkeley.edu/result.php?resultid=4434235284
http://setiathome.berkeley.edu/result.php?resultid=4434235302
http://setiathome.berkeley.edu/result.php?resultid=4434235352
http://setiathome.berkeley.edu/result.php?resultid=4434234844
http://setiathome.berkeley.edu/result.php?resultid=4434235370
http://setiathome.berkeley.edu/result.php?resultid=4434235402

VGA card 2
http://setiathome.berkeley.edu/result.php?resultid=4434235396
http://setiathome.berkeley.edu/result.php?resultid=4434234918
http://setiathome.berkeley.edu/result.php?resultid=4434235183
http://setiathome.berkeley.edu/result.php?resultid=4434235441
http://setiathome.berkeley.edu/result.php?resultid=4434235443
http://setiathome.berkeley.edu/result.php?resultid=4434235460
http://setiathome.berkeley.edu/result.php?resultid=4434235297
http://setiathome.berkeley.edu/result.php?resultid=4434235070
http://setiathome.berkeley.edu/result.php?resultid=4434235332
http://setiathome.berkeley.edu/result.php?resultid=4434235334

VGA card 3
http://setiathome.berkeley.edu/result.php?resultid=4434235392
http://setiathome.berkeley.edu/result.php?resultid=4434235404
http://setiathome.berkeley.edu/result.php?resultid=4434235412
http://setiathome.berkeley.edu/result.php?resultid=4434235167
http://setiathome.berkeley.edu/result.php?resultid=4434235182
http://setiathome.berkeley.edu/result.php?resultid=4434235457
http://setiathome.berkeley.edu/result.php?resultid=4434235311
http://setiathome.berkeley.edu/result.php?resultid=4434235373
http://setiathome.berkeley.edu/result.php?resultid=4433115667
http://setiathome.berkeley.edu/result.php?resultid=4433115414

What should I do now?

Thanks.

---

Look at my result

Maximum single buffer size set to:256MB
Number of period iterations for PulseFind set to:40
SpikeFind FFT size threshold override set to:2048
TUNE: kernel 1 now has workgroup size of (64,1,4)
oclFFT global radix override set to:256
oclFFT local radix override set to:16
oclFFT max WG size override set to:256
oclFFT max local FFT size override set to:512
oclFFT number of local memory banks set to:64
oclFFT minimal memory coalesce width set to:64
CPU affinity adjustment disabled
Priority of worker thread raised successfully
Priority of process adjusted successfully, high priority class used

---

With each crime and every kindness we birth our future.

If I set -no_cpu_lock the above mentioned errors happens:

Messages in BOINC:
Task postponed: Suspicious spike results, host needs reboot or maintenance
...or...
Task postponed: Triplet data corruption, retry from checkpoint.

Maybe I get it a try again?
Can't remember if I tested it already with just 1 WU/GPU.

---

Put this in comandline.txt file.

-sbs 256 -period_iterations_num 40 -spike_fft_thresh 2048 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -hp -no_cpu_lock

Just one taks per GPU.
Everything else will not work.

Also Cat 15.7 is better than 15.7.1.
I also got problems using it.

---

With each crime and every kindness we birth our future.

What do you have in your cmdline.txt file for AP?

Thanks.

---

What do you have in your cmdline.txt file for AP?

Thanks.

-oclFFT_plan 256 16 256 -tune 1 64 4 1 -tune 2 64 4 1 -ffa_block 2830 -ffa_block_fetch 2830

---

With each crime and every kindness we birth our future.

i think you'e only getting the triplet errors because you're on the new drivers and you're running multiple wu per gpu

if you decide to stay with that driver, then going back to one wu per gpu should stop the triplet errors

Mike, which problems you had with AMD Catalyst v15.7.1?
It looks like my PC made a self reboot with it.


How to uninstall the AMD Catalyst?
Windows 8.1: Programs and Features
Uninstall 'AMD Catalyst Install Manager'?

If I install the AMD Catalyst,
I should use the 'Express' or the 'Custom' installation?
Last time I made the 'Custom' installation and let all checked (IIRC, 6 entries, incl. Raptr).
(next time I'll uncheck Raptr, because it's just a gamer online thing)
Which is really needed and must be installed (for SETI crunching)?

If I would like to test AMD Catalyst v15.9.1 Beta, (someone tested this version already, with which OS/hardware?)
there is all in it or I need additional software?
Because the file names are different, with (normal) and without (beta version) 'with .NET 4.5'.


woohoo, IIRC, the R9 Fury X is available since mid 2015.
The v15.7.1 was released 2015/07/29.
The v15.7 was released 2015/07/08.
The v14.12 was released 2014/12/09, I guess it wouldn't work with my VGA cards.

---