BOINC Manager doesn't measure anything at all - it simply displays what the BOINC Client tells it. I'd have expected BoincTasks to do the same.

I've just built myself a client which (if I've done it right) skips the calculation of pseudo-progress for apps which don't report <fraction_done> - that might make the situation clearer if I can catch a VVVHAR.

(That's the tweak to fraction_done introduced with v7.2.38 - current code in client/app.cpp, ll 703-706, if anyone else wants to do the same)

BOINC Manager doesn't measure anything at all - it simply displays what the BOINC Client tells it. I'd have expected BoincTasks to do the same.

I've just built myself a client which (if I've done it right) skips the calculation of pseudo-progress for apps which don't report <fraction_done> - that might make the situation clearer if I can catch a VVVHAR.

(That's the tweak to fraction_done introduced with v7.2.38 - current code in client/app.cpp, ll 703-706, if anyone else wants to do the same)

You know how to download tasks from the server, and run offline. I sure don't. Take this one as a test, it doesn't have that high AR, only 23.133295, but it did behave exactly as the other with even higher AR WU's

http://setiathome.berkeley.edu/workunit.php?wuid=2092148298

or this one:

http://setiathome.berkeley.edu/workunit.php?wuid=2092148304

Hi!

I have an Apple iMac 11,1 (late 2009) with a ATI Radeon HD 4850. It has been working great for Feb & March until just recently. Now the graphics subsystem seems to stop working - I get a "service progress watchdog" event from the WindowServer. This means I can still see the desktop and can move the cursor, but no "clicks" or keyboard taps are recognized (serviced), so my only recourse is to power off. I have also seen recently some graphics artifacts.

Has anything changed? As I remember (so take with a grain of salt), the OpenCL tasks used to take about an hour, and now they take closer to 3 hours.

John

After looking more closely, I have only had the graphics subsystem stopping for 8.07. My computer worked fine for 8.06.

Jimbocous, which version of BOINC are you running? I can look it up myself, but just wanting to be sure whether it was one of you v7.6.9 hosts, or the v7.6.22 one?

The machine where I observed this is still on 7.6.9.
Thanks, Jim ...

---

If I can help out by testing something, please let me know.
Available hardware and software is listed in my profile here.

You know how to download tasks from the server, and run offline. I sure don't.

Yes, I can do that, and I wrote a tool to help other people do it too. But I wanted to run some tasks online if possible, to see BOINC's progress estimation with my own eyes.

Mostly got mid-AR (0.42): took about 15 minutes on GTX 750 Ti. But that was five minutes to reach 10%, another four minutes to reach 20%, maybe six minutes for the whole final 80%. That's certainly more non-linear than I'm used to.

Giving up staring at the screen for tonight, I'll so some more tomorrow. And yes, probably grab one of those 100-plus ARs to see how they run.

It was explained few times already please use search

I recall your discussion clearly. This is a different issue, in my opinion, and I do not find your attitude helpful.
I spent time trying to make this as clear as possible. Is that not what the Beta process is supposed to be for?
If it is offensive to you for me to identify issues, I can easily desist.

---

If I can help out by testing something, please let me know.
Available hardware and software is listed in my profile here.

It was explained few times already please use search

I recall your discussion clearly. This is a different issue, in my opinion, and I do not find your attitude helpful.
I spent time trying to make this as clear as possible. Is that not what the Beta process is supposed to be for?
If it is offensive to you for me to identify issues, I can easily desist.

Not offensive, but reiterate just waste of time.

---

News about SETI opt app releases: https://twitter.com/Raistmer

---

---

If I can help out by testing something, please let me know.
Available hardware and software is listed in my profile here.

I myself wrote:

Application details for host 76831
http://setiweb.ssl.berkeley.edu/beta/host_app_versions.php?hostid=76831

SETI@home v8 8.09 windows_intelx86 (opencl_atiapu_sah)
SETI@home v8 8.09 windows_intelx86 (opencl_atiapu_SoG)
SETI@home v8 8.09 windows_intelx86 (opencl_ati5_nocal)
SETI@home v8 8.09 windows_intelx86 (opencl_ati5_SoG_nocal)
SETI@home v8 8.09 windows_intelx86 (opencl_ati_nocal)

Downloaded to the project folder:
*opencl_ati_sah.exe
*opencl_ati5_sah.exe
*opencl_ati5_SoG.exe
*opencl_atiapu_sah.exe
*opencl_atiapu_SoG.exe

Just two apps are 'identically' - is this all OK or is there something wrong?

Raistmer wrote:

Did you check their MD5 sums to say they are identical ones?

I myself wrote:

With 'identically' I meant the 'plan_class'.

The host apps details overview show 5 apps, in the project folder are 5 apps.
But just two 'plan_class's are similar (shown in my last/above message).


On the apps overview are shown AMD/ATI apps:
Windows/x86 8.09 (opencl_ati5_cat132)
Windows/x86 8.09 (opencl_ati5_nocal)
Windows/x86 8.09 (opencl_ati5_sah)
Windows/x86 8.09 (opencl_ati5_SoG)
Windows/x86 8.09 (opencl_ati5_SoG_cat132)
Windows/x86 8.09 (opencl_ati5_SoG_nocal)
Windows/x86 8.09 (opencl_atiapu_sah)
Windows/x86 8.09 (opencl_atiapu_SoG)
Windows/x86 8.09 (opencl_ati_cat132)
Windows/x86 8.09 (opencl_ati_nocal)
Windows/x86 8.09 (opencl_ati_sah)

Which, how much apps (shown on the host apps details overview & project folder?) should get my FuryX VGA cards?

At the host apps details overview & in the project folder should be 11 apps?

Raistmer wrote:

Where you see identical plan classes if they supply different binaries??

It looks like we have a communication problem...

I just would like to know how much apps my FuryX VGA cards will get.

In the project apps overview are 11 apps for AMD/ATI GPUs.

How much apps will be in my project folder and how much apps will be shown on the host apps overview (AFAIK, this value could be different)?

---

I just would like to know how much apps my FuryX VGA cards will get.

Eventually, all that compatible with particular plan classes.
Some of plan classes have mutually exclusive driver version restrictions so, until you start to change driver versions between downloads, to have all binaries is impossible. But PC with modern drivers and BOINC should get all 5 different ATi app flavors: HD5 SoG/nonSoG, APU SoG/nonSoG, non-HD5.
Plan class under wich those binaries arrive doesn't matter for processing.

PS: in particular, host you linked has:

SETI@home v8 8.09 windows_intelx86 (opencl_atiapu_sah)
Число завершённых заданий 41
Максимум заданий в день 37
Число заданий сегодня 0
Правильные задания завершённые подряд 5
Средняя скорость обработки 356.42 GFLOPS
Среднее время обработки 0.05 days
SETI@home v8 8.09 windows_intelx86 (opencl_atiapu_SoG)
Число завершённых заданий 20
Максимум заданий в день 38
Число заданий сегодня 0
Правильные задания завершённые подряд 7
Средняя скорость обработки 378.41 GFLOPS
Среднее время обработки 0.05 days
SETI@home v8 8.09 windows_intelx86 (opencl_ati5_nocal)
Число завершённых заданий 61
Максимум заданий в день 39
Число заданий сегодня 0
Правильные задания завершённые подряд 8
Средняя скорость обработки 530.68 GFLOPS
Среднее время обработки 0.05 days
SETI@home v8 8.09 windows_intelx86 (opencl_ati5_SoG_nocal)
Число завершённых заданий 82
Максимум заданий в день 49
Число заданий сегодня 0
Правильные задания завершённые подряд 17
Средняя скорость обработки 614.79 GFLOPS
Среднее время обработки 0.05 days
SETI@home v8 8.09 windows_intelx86 (opencl_ati_nocal)
Число завершённых заданий 78
Максимум заданий в день 48
Число заданий сегодня 0
Правильные задания завершённые подряд 16
Средняя скорость обработки 555.41 GFLOPS
Среднее время обработки 0.04 days

that is, it has all 5 flavors already downloaded.

And, as expected for discrete GPU, HD5 version is the fastest.
Judging from current statistics, SoG HD5 build the best one for your host. But statistics not big enough to not change this conclusion later.

---

News about SETI opt app releases: https://twitter.com/Raistmer

I had the Crimson 16.3 Hotfix (Beta) software installed.

I guess because of too less free CPU-Cores my PC got (errors):
197 (0xc5) EXIT_TIME_LIMIT_EXCEEDED

But I had:
0.75 CPU for 1 GPU app.
This means 3 CPU-Cores for 4 GPU apps (a few % idle CPUs).
And the other 9 CPU-Cores with tasks.

(The PC have two 6 Cores/12 Threads CPUs, but HT off, so 12 real Cores available.)

I had for all apps here used the cmdline settings like for r3330.


Now I'm back to Crimson 15.12.

Maybe I'll test it this time with 1 CPU for 1 GPU app.

The cmdline settings/usage for/of r3401 is differnet than for/of r3330?

The 'default' settings for/of r3401 'use' the GPU 'better'?

If so, I should test the apps with 'default' settings?
But with at least '-no_cpu_lock -hp'?

Thanks.

(BTW. Still just 1 WU/GPU possible (also with Crimson 16.3 Hotfix (Beta))).

---

The cmdline settings/usage for/of r3401 is differnet than for/of r3330?

yes, new added. -sbs N will act differently so new tuning required for it.

The 'default' settings for/of r3401 'use' the GPU 'better'?

Yes, default settings improved.

If so, I should test the apps with 'default' settings?

Yes, it's preferable for initial testing. Only when baseline established I would recommend to start further optimization.

But with at least '-no_cpu_lock -hp'?

No, most probably -no_cpu_lock resulted in EXIT_TIME_LIMIT_EXCEEDED.
Either keep CPU free or not use this option.

---

News about SETI opt app releases: https://twitter.com/Raistmer

I recall your discussion clearly. This is a different issue, in my opinion ...

And in my opinion, too.

As well as running the tweaked client which eliminates BOINC's 'reassuring' calculation and display of pseudo-progress, I wrote a little batch script which collected some interesting metrics from a live running task every 15 seconds or so.

The script collects:

"wu_name" from init_data.xml in the slot directory
"true angle range" from stderr.txt, also in the slot directory
(both so I can be sure I'm reporting on the right task)

"prog" from the checkpoint file state.sah
"fraction_done" from boinc_task_state.xml

"prog" (for 'progress') is, I think, what is printed into std_err.txt as 'Restarted at xx.xx%' when a task is interrupted.
"fraction_done" matches what I see on screen in the progress% column in BOINC Manager while the task is running.

This is the NV SoG log for task 23317903

wu_name: 24no10ab.14894.5793.7.41.229
WU true angle range is :  9.964068

            <prog>        <fraction_done>
17:22:52
17:23:07    0.00174570    0.000023
17:23:22    0.00584649    0.000023
17:23:37    0.01006975    0.000023
17:23:52    0.01416061    0.000023
17:24:07    0.01825479    0.000023
17:24:22    0.02235558    0.000023
17:24:37    0.02235558    0.000023
17:24:53    0.02657553    0.000023
17:25:08    0.03066970    0.000023
17:25:23    0.03476222    0.000023
17:25:38    0.03899044    0.000023
17:25:53    0.04310282    0.042722
17:26:08    0.04720030    0.042722
17:26:23    0.05142025    0.042722
17:26:38    0.05551277    0.042722
17:26:53    0.05960529    0.042722
17:27:08    0.06382854    0.042722
17:27:23    0.06791941    0.042722
17:27:38    0.07202186    0.042722
17:27:53    0.07202186    0.042722
17:28:08    0.07624346    0.042722
17:28:23    0.08033763    0.042722
17:28:38    0.08443346    0.146864
17:28:53    0.08866210    0.146864
17:29:08    0.09276289    0.146864
17:29:23    0.09685376    0.146864
17:29:38    0.10107701    0.146864
17:29:53    0.10516953    0.146864
17:30:08    0.11434709    0.622830

It's clear that <prog> and <fraction_done> are recording different values, and neither is proceeding smoothly from 0 to 100. The last 85% of progress is reported in big jumps in the last 30 seconds of the run - hard to do justice to with the 15-second sampling I was using.

By comparison, here's a similar log (from another machine) for Main project task 4791417288, run using the AVX/64 CPU application and ~60 second sampling.

wu_name: 18se10ac.27843.12337.6.33.240
WU true angle range is :  8.589667

            <prog>        <fraction_done>
16:42:25
16:43:25    0.02829758    0.028819
16:44:28    0.05665864    0.057073
16:45:31    0.08501804    0.085465
16:46:34    0.11446137    0.114971
16:47:37    0.14387988    0.144348
16:48:40    0.17332404    0.144348
16:49:43    0.17332404    0.173753
16:50:46    0.20168344    0.202050
16:51:49    0.23005650    0.230304
16:52:52    0.25838693    0.258655
16:53:55    0.28672896    0.286721
16:54:58    0.31508836    0.315580
16:56:01    0.34370029    0.343762
16:57:05    0.37205183    0.372415
16:58:08    0.40150344    0.401897
16:59:13    0.43093519    0.431290
17:00:16    0.45928714    0.459446
17:01:19    0.48869572    0.488797
17:02:22    0.51814733    0.517937
17:03:25    0.54649928    0.546169
17:04:28    0.57485289    0.574161
17:05:31    0.60317919    0.602094
17:06:34    0.63155224    0.629994
17:07:36    0.66010122    0.657754
17:08:39    0.68846062    0.685064
17:09:42    0.71677203    0.712603
17:10:45    0.71677203    0.712603
17:11:47    0.74513308    0.739604
17:12:49    0.77457559    0.766697
17:13:52    0.80296850    0.793204
17:14:53    0.83132210    0.818806
17:15:56    0.86074972    0.844610
17:16:58    0.88911243    0.868462
17:18:01    0.91747265    0.891763
17:19:04    0.94604818    0.915093

<prog> and <fraction_done> are in far closer agreement, and while not perfectly linear, are a far closer approximation to what volunteers will expect to see.

The cmdline settings/usage for/of r3401 is differnet than for/of r3330?

yes, new added. -sbs N will act differently so new tuning required for it.

Increasing the -sbs setting, or decreasing?

After looking at the output I had to increase mine significantly. Default of 128 worked pretty well but there were still some sub-optimal array sizes. The largest work group I saw was 1312 I think, so I bumped mine to 1536 and that seems to stopped all the warnings about sub-optimal size. That being said, is likely specific to my setup.

Chris

I am getting quite a few inconclusives. Seeing that on my Mac Pro both on Main and beta, but not on a Nvidia 570 in Windows. In each case I'm missing a gaussian compared to my wingman.

Main:

http://setiathome.berkeley.edu/results.php?hostid=6105482&offset=0&show_names=0&state=3&appid=

Beta:

https://setiweb.ssl.berkeley.edu/beta/results.php?hostid=60016&offset=0&show_names=0&state=3&appid=

Let me know if you need something else.

Thanks,

Chris

The cmdline settings/usage for/of r3401 is differnet than for/of r3330?

yes, new added. -sbs N will act differently so new tuning required for it.

Increasing the -sbs setting, or decreasing?

Try both ways to find optimum. The same number of possible WGs aligned differently versus CUs now. Also, there are 2 new params to use in conjunction with -sbs N. I did some parameters space exploration and Mike did the same on own GPU but had no time to process those results so far so can't say if defaults chosen really optimally or some further improvement possible in this area.

---

News about SETI opt app releases: https://twitter.com/Raistmer

17:30:08 0.11434709 0.622830

Difference is big enough indeed. How long task lasted as whole?
What checkpoint interval was?
What if checkpoint interval will be set to 10 seconds?

---

News about SETI opt app releases: https://twitter.com/Raistmer

After looking at the output I had to increase mine significantly. Default of 128 worked pretty well but there were still some sub-optimal array sizes. The largest work group I saw was 1312 I think, so I bumped mine to 1536 and that seems to stopped all the warnings about sub-optimal size. That being said, is likely specific to my setup.

Chris

Yep, some tasks along with high-numbered CUs can require hige number of workgroups to fill all CUs evenly. But cause even load not the only factor that affects performance benchmarking requiring of real performance, making -v 8 warnings disappear is not enough. Worth to check (with -v 8 disabled) how your GPU performs with high sbs settings and with more moderate ones.

---

News about SETI opt app releases: https://twitter.com/Raistmer

I am getting quite a few inconclusives. Seeing that on my Mac Pro both on Main and beta, but not on a Nvidia 570 in Windows. In each case I'm missing a gaussian compared to my wingman.

Main:

http://setiathome.berkeley.edu/results.php?hostid=6105482&offset=0&show_names=0&state=3&appid=

Beta:

https://setiweb.ssl.berkeley.edu/beta/results.php?hostid=60016&offset=0&show_names=0&state=3&appid=

Let me know if you need something else.

Thanks,

Chris

Worth to check if it's SOG specific or Mac-specific issue.
I could process those beta tasks later with own GPU, maybe Urs could check them on Mac. Urs?

---

News about SETI opt app releases: https://twitter.com/Raistmer