Strange problem with the new v0.2 ATI app on one of my boxes. Two out of three are fine after updating to v0.2, the third craps out with a failed to create program error.

The error message from boincmanager is

14/09/2009 12:22:54	Milkyway@home	Output file gs_constrainted_82_2s_4_2346758_1252927308_0_0 for task gs_constrainted_82_2s_4_2346758_1252927308_0 absent

This happens on all WU under 0.2: reverting back to v0.19 the box runs fine.

Specs: Boinc 6.6.36, HD4780, 32bit Win7 RC1. This box

Tried running Boinc as Admin: same error.

Any thoughts?

---

Symington weather report and video feed

Do you have the six CAL libraries in your system32 folder? That was my issue when I had your problem. Not sure if it applies to yours as .19 works, but it couldn't hurt.

Yes, all six dll's in place.

It's very repeatable, so something systematic. I also double checked that I have MSCV++ 2005 SP1 installed too.

Just been playing with permissions and tweaks in the app_info but still no joy.

The error says that the output file for task xyz is missing. I'm not entirely sure which file that is referring to. It might be the OUT file in the slot directory, but in anycase, it's too fast for me to check. Boinc whizzes down the cache in a couple of seconds trashing each WU

0.19 ATI runs fine; 0.20 ATI barfs, but only on that machine - the other two boxes are running sweet under XP SP3.

---

Symington weather report and video feed

I'm also having problems with 0.20 on Win 7 x64.
But for me all tasks are failing... Noticed this in the stderr -

<core_client_version>6.10.3</core_client_version>
<![CDATA[
<message>
Unzulässige Funktion. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.20 (Win64) by Gipsel
ignoring unknown input argument in app_info.xml: --device
ignoring unknown input argument in app_info.xml: 0
CPU: Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz (4 cores/threads) 3.18772 GHz (252ms)

CAL Runtime: 1.4.283
Found 1 CAL device

Device 0: ATI Radeon HD 4800 (RV770) 512 MB local RAM (remote 2047 MB cached + 2047 MB uncached)
GPU core clock: 750 MHz, memory clock: 900 MHz
800 shader units organized in 10 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

0 WUs already running on GPU 0
Starting WU on GPU 0

main integral, 320 iterations
predicted runtime per iteration is 104 ms (33.3333 ms are allowed), dividing each iteration in 4 parts
borders of the domains at 0 400 800 1200 1600
0, integration, Kernel Execution : Failed to create Program

</stderr_txt>
]]>

I'm using the normal app_info provided with the win64 0.20 download. Only changed <count> to 0.5. On another host with Vista x64 everything works fine... And yes, I have installed the MSCV++ 2005 SP1 for x64 too...

Any ideas?

[edit]Using BOINC 6.10.3...

---

Time Bandit,

I see the same lines in my successfully running tasks.

So, I do not think that this is for sure the issue.

I have been running over 6 hours now (as far as I know continually) with no error or invalid results. Major difference is that I am running Vista 32 Ultimate ...

Time Bandit,

I see the same lines in my successfully running tasks.

So, I do not think that this is for sure the issue.

I have been running over 6 hours now (as far as I know continually) with no error or invalid results.

It's no issue. It just says that the app ignores the wishes of the BOINC client which GPU should be used ;) The app will obey these wishes in a future version, but it is working well as it is.

Ok, thanks... Do you have any idea why 0.20 is only giving errors and 0.19 f works fine on Windows 7 X64?

---

FWIW, my stderrr for these is as follows:

<core_client_version>6.6.36</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.20 (Win32, SSE2) by Gipsel
setting minimum kernel frequency to 10 Hz
scaling the wait times with 0.8
CPU: Intel(R) Core(TM)2 CPU          6700  @ 2.66GHz (2 cores/threads) 2.6605 GHz (396ms)

CAL Runtime: 1.4.283
Found 1 CAL device

Device 0: ATI Radeon HD 4800 (RV770) 512 MB local RAM (remote 831 MB cached + 831 MB uncached)
GPU core clock: 750 MHz, memory clock: 900 MHz
800 shader units organized in 10 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

3 WUs already running on GPU 0
No free GPU! Waiting ... 0.03125 seconds.
Starting WU on GPU 0

main integral, 320 iterations
predicted runtime per iteration is 104 ms (100 ms are allowed), dividing each iteration in 2 parts
borders of the domains at 0 800 1600
0, integration, Kernel Execution : Failed to create Program

</stderr_txt>
]]>

Notice the same error message in the last line.

---

Symington weather report and video feed

Do you have any idea why 0.20 is only giving errors and 0.19 f works fine on Windows 7 X64?

I'm having no trouble with the new 0.20 ATI app on Win 7 64 - using the new Catalyst 9.9 and BOINC ver 6.10.3

It sounds like CP has a suggestions for trying a different Catalyst version.

FWIW, my stderrr for these is as follows:

<core_client_version>6.6.36</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
[..]

CAL Runtime: 1.4.283

[..]
0, integration, Kernel Execution : Failed to create Program
</stderr_txt>
]]>

Notice the same error message in the last line.

There appears to be a certain combinations of a Win7, Catalyst 9.5/9.6(?) and the 0.20 app not working together. The error you see is indicating that the JIT compiler in the driver (used to translate the GPU code from some intermediate language to the code optimized for the exact GPU type you are using) is unable to compile the GPU code (or bind it to a buffer or whatever). It simply appears to be a bug of that specific driver version. Installing a different one should solve the problem.

The 0.20 ATI app was tested on about 15 different systems (I was not doing this alone) using several combinations of 32 and 64Bit versions of XP, Vista as well as Win7 and also several Catalyst versions and this behaviour was not observed on a single system. But I guess it is clear that with my limited resources I can't test all combinations of 6 different Windows variants and the at least 10 different Catalyst versions (8.12 to 9.9 + several hotfixes in between) which in principle are able to run the MW app.

I also could not get 0.20 running, and none of the normal mistakes were made.. finally I could not return succesfully to 0.19 before copying the entire folder from another machine.

ATI9.5, 4850, win7x64, 6.6.36. Had all the files, removed 6.10.x specific lines in app_info, had the renamed dll's in system32.. just could not get it working. Any ideas from the failed tasks?
<core_client_version>6.6.36</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.19e by Gipsel
allowing 2 concurrent WUs per GPU
CPU: AMD Athlon(tm) 64 Processor 3200+ (1 cores/threads) 2.30077 GHz (425ms)

CAL Runtime: 1.4.283
Found 1 CAL device

Device 0: ATI Radeon HD 4800 (RV770) 1024 MB local RAM (remote 319 MB cached + 319 MB uncached)
GPU core clock: 735 MHz, memory clock: 1000 MHz
800 shader units organized in 10 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

0 WUs already running on GPU 0
Starting WU on GPU 0

main integral, 320 iterations
predicted runtime per iteration is 195 ms (33.3333 ms are allowed), dividing each iteration in 6 parts
borders of the domains at 0 266 533 800 1066 1333 1600
0 Kernel Execution : Failed to create Program

</stderr_txt>
]]>

OK, downloading new driver now......will post back with findings shortly.

I also could not get 0.20 running, and none of the normal mistakes were made.
[..]
ATI 9.5, 4850, win7x64, 6.6.36. Had all the files, removed 6.10.x specific lines in app_info, had the renamed dll's in system32.. just could not get it working. Any ideas from the failed tasks?

[..]
CAL Runtime: 1.4.283
[..]
0 Kernel Execution : Failed to create Program

Same driver/OS combination, same error. Appears to be a pattern.

I would suggest to stay with Catalyst 9.2 even for Win7.
It seems higher versions only degrade stream computing, not improve it...

I just upgraded to Cat9.9 and hey presto it sprung into life!

Most excellent result, thanks very much for your hard work!

---

Symington weather report and video feed

Great! With Cat 9.9 it's working now! :)
Thanks !

---

Mine seems to be working ok under Windows 7, 64 Bit Catalyst 9.9. Tho, I do have one question... in the BOINC manager it still says Milkyway 0.19, is this correct? If so, is it because the official client is still 0.19?

Edit: After downloading new WUs, it changed to 0.20. Am getting a few Computation Errors after 1 second here and there, but most WUs seem to be going thru OK.

---

Am getting a few Computation Errors after 1 second here and there, but most WUs seem to be going thru OK.

This is probably due to your BOINC client 6.10.4 It can leave one workunit in progress and switch to another more recently downloaded one. When it goes back to the interrupted unit it may error out. If you use 6.10.3 this clears up.

Would edit, but it won't let me again... my system is tossing a fair number of Computation Errors now... running 6.10.4 with a Radeon 4850 on afore mention Windows 7 64 bit Catalyst 9.9, any idea what might be wrong, or where I can snag the info that might help?

Edit:

Am getting a few Computation Errors after 1 second here and there, but most WUs seem to be going thru OK.

This is probably due to your BOINC client 6.10.4 It can leave one workunit in progress and switch to another more recently downloaded one. When it goes back to the interrupted unit it may error out. If you use 6.10.3 this clears up.

You responded while I was typing up a new message, I've now reverted to 6.10.3 by your advise to see if that helps, cheers!

---

When rolling back, 6.10.3 may error out on those interrupted units that are "waiting to run", so don't worry if has to clear up your plate a bit. From there on it should be smooth sailing. (As much as BOINC ever is...) :)