I have uploaded a new acemdbeta application for Linux and some workunits to test.

Mainly bug fixes.

gdf

Does this new app resolve the Cuda4/downclocking bug discussed in http://www.gpugrid.net/forum_thread.php?id=2534?

If not, may I refer you to http://boinc.berkeley.edu/trac/changeset/23649/, and the new paragraph in http://boinc.berkeley.edu/trac/wiki/AppCoprocessor:

Cleanup on premature exit
The BOINC client may kill your application in the middle. This may leave the GPU in a bad state. To prevent this, call

boinc_begin_critical_section();

before using the GPU, and between GPU kernels do

if (boinc_status.quit_request || boinc_status.abort_request) {
    // cudaThreadSynchronize(); or whatever is needed
    boinc_end_critical_section();
    while (1) boinc_sleep(1);
}

No. This is a cuda3.1 app.

yet I don't understand what that means. In the middle of what?

gdf

No. This is a cuda3.1 app.

yet I don't understand what that means. In the middle of what?

gdf

The BOINC API library code is not threadsafe. If BOINC calls for the application to quit or suspend during computation, BOINC may terminate threads in an unsafe way. The new nVidia drivers which can handle Cuda4 apps are much more sensitive to this behaviour, even if the app that's running is only using a lower CUDA level. In self-protection, nVidia has written the new drivers - eveything strictly *later than* 266.58, from memory - to down-clock the card into a protective state when the abnormal thread termination is detected.

That's my layman's interpretation - I'll try and get you the full report quickly.

Ooops, it's just been pointed out to me that this is a Linux beta app, and my remarks have been concentrating on the Windows API - so probably not important in this case.

But, since the new API code was only posted last night, it's still worth you knowing about it in preparation for the next Windows application test.

The first results are fine. We are going to produce today a beta for Windows.
This applications will substitute all production apps already this week, if all goes well
gdf

Windows application is out.

gdf

Windows application is out.

gdf

I got two of these beta wus. Both of them failed immediately with exit code -1073741819 (0xc0000005). Maybe they can't stand overclocking?

acemdbeta_6.38_windows_intelx86__cuda31 - Application Error

The exception unknown software exception (0xc0000005) occurred in the application at location 0x0040258c.

Click OK to terminate the program
Click on CANCEL to debug the program

System: 2003 Server x64 i7-2600K, 8GB DDR3, 2TB, GTX470 (native clocks, increased fan speeds)

Task ran for 1h45min but stayed at 0% complete, apparently going through a loop.

My only other two Betas also failed after 7 and 15sec.

Worth noting that if there is a pop-up error message and you don't select to end the task it will continue running indefinately.
So if anyone has such a message do something about it or you will just keep running the same erroneuos Beta task.

I got two more of these 6.38 beta, both of them failed immediately just like the previous ones. There were a pop-up application error message.

Application Failure acemdbeta_6.38_windows_intelx86__cuda31 0.0.0.0 in acemdbeta_6.38_windows_intelx86__cuda31 0.0.0.0 at offset 00002c58

These 6.38 beta WUs seem to fail on every computer, so I guess I shouldn't blame the overclocking. :)

No, I am trying with a quick change in few minutes to see if it works. Otherwise, it will take more time to debug it.

gdf

The same here
Run time 66.171875
CPU time 0
Interesting, Swan_sync set and works with standard/long run tasks.

<core_client_version>6.10.60</core_client_version>
<![CDATA[
<message>
 - exit code -1073741819 (0xc0000005)
</message>
]]>

Win XP 64bit, GTX 560.

acemdbeta_6.39 substitutes acemdbeta_6.38 hopefully with better results...

gdf

This 6.39 task reached 10% complete in 17min 43sec on a stock GTX470 (System), so estimated run time is 3h.
98% GPU Utilization, 315MB video memory usage.

Looks good so far... 20%

Strange application that 6.39 beta one.
6.39 CPU process started with high priority. The system GUI was sluggish, 1 - 2 minutes response. The Boinc GUI freezed, the Boinc core restarted (CPU tasks without checkpoint have started from zero progress). After setting the priority of 6.39 CPU process to low by Process Tamer - response of PT GUI was about 2 minutes (high priority set for PT process!) - the Boinc core restarted again and now things seems to be OK.
After 12 min 8% progress.
Win XP 64bit, GTX 560Ti, 925MHz.

4069069 2520078 9 Jun 2011 17:12:15 UTC 9 Jun 2011 20:10:53 UTC Completed and validated 10,510.59 10,444.27 7,491.18 11,236.77 ACEMD beta version v6.39 (cuda31)

Well, this task ran and finished (2h 55min) without error on the 6.39 app. Hopefully other results are just as positive.

Any problem?

gdf

I got one 6.39 beta WU, it's running for 16 minutes now, 15% completed, 98% GPU usage (i7-950 @ 3.56GHz, GTX 580 @ 890MHz, WinXP, Above average priority).

Edit:

I put this WU to my GTX 590 @ 700MHz (my monitor is connected to this card) tried on both GPU of the 590, and I don't experience sluggish Windows GUI.

After 32 minutes 28.2% completed.

This 6.39 beta WU completed fine in 6438s (1h47m) 3.219 ms/step.
I got another one, it's processing will begin in 3 hours, because I'm going to sleep now, so I can't micromanage the processing order of the WUs. :)