I would like to ask for advice.

My new GPU card (http://www.primegrid.com/forum_thread.php?id=2677) from time to time completes WU with error.
I wonder if it is possible to do something about it. I have checked error code is always the same:

Cuda error: getting factors found: unspecified launch failure

So far the card has completed 126 WU: 104 WU successfully, 22 WU were discarded as error (which is about 20%).
Card is running with default settings, card is not overclocked.
As for run time I see no pattern here, some WU ended after 20sec other almost completed about 2100sec.
Card drivers were downloaded from nvidia webpage: NVIDIA-Linux-x86_64-256.53.run

BOINC detects card as:
03-Sep-2010 14:01:10 [---] Starting BOINC client version 6.10.56 for x86_64-pc-linux-gnu
03-Sep-2010 14:01:10 [---] Config: GUI RPC allowed from any host
03-Sep-2010 14:01:10 [---] log flags: file_xfer, sched_ops, task
03-Sep-2010 14:01:10 [---] Libraries: libcurl/7.18.0 OpenSSL/0.9.8g zlib/1.2.3.3 c-ares/1.5.1
03-Sep-2010 14:01:10 [---] Data directory: /home/boinc/BOINC
03-Sep-2010 14:01:10 [---] Processor: 2 GenuineIntel Pentium(R) Dual-Core CPU E5200 @ 2.50GHz [Family 6 Model 23 Stepping 10]
03-Sep-2010 14:01:10 [---] Processor: 2.00 MB cache
03-Sep-2010 14:01:10 [---] Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm constant_tsc arch_perfmon pebs bts rep_good nopl pni monitor ds_cpl est tm2 ssse3 cx16 xtpr lahf_lm
03-Sep-2010 14:01:10 [---] OS: Linux: 2.6.27-17-generic
03-Sep-2010 14:01:10 [---] Memory: 1.96 GB physical, 227.44 MB virtual
03-Sep-2010 14:01:10 [---] Disk: 3.74 GB total, 987.09 MB free
03-Sep-2010 14:01:10 [---] Local time is UTC +2 hours
03-Sep-2010 14:01:10 [---] NVIDIA GPU 0: GeForce GT 240 (driver version unknown, CUDA version 3010, compute capability 1.2, 511MB, 280 GFLOPS peak)


What is the reason for "getting factors found: unspecified launch failure" error message?
Should I suspect hardware or software problem?
____________

This is not a virtual machine.

I wanted to install Dotsch/UX linux distribution at pendrive, but I have found this drive in my garage and decided to try it.
It has worked as you see...and now it is my "vintage" pendrive :-)

The hard drive is: MPC3043A 4.3GB U/ATA FUJITSU HARD DRIVE

Other PC components are: E5200 @ 2.50GHz running Gigabyte GA-P43-ES3G with 2GB RAM, powered by 400W power supply.

I have checked syslog/messages logs for any suspicious information, but found nothing.
Besides this PC is also crunching other BOINC projects for CPU like WEP-M+2 and WCG.
No problems with WU for those projects...so if that would be hardware issue (memory failure or CPU overheating) those projects should be affected too.
____________

Computers at my account are now visible, the one having problem is hostid=157777.

Is there any other way to test the card, besides running Collatz?
Some other application under Windows? I mean I have nothing against it, just asking for other possibilities.

I could also move the card to another host, for example hostid=147474.
____________

I am not saying it is not a hardware, but after reviewing results it looks like I am not the only one having problem. I have checked what were the results for my task run by other computers to reach quorum. Below are few examples, outcome of three attempts plus additionally at round brackets GPU details.

WU=129821754
Cuda error: getting factors found: unspecified launch failure
Insufficient available memory on GPU 0. (NVIDIA Quadro NVS 160M (255MB))
In progress

WU=129819839
Cuda error: getting factors found: unspecified launch failure
Insufficient available memory on GPU 0. (NVIDIA GeForce 9600 GT (499MB) driver: 25896)
Insufficient available memory on GPU 0. (NVIDIA Quadro NVS 160M (255MB))

WU=129819591
Cuda error: getting factors found: unspecified launch failure
Insufficient available memory on GPU 0. (NVIDIA GeForce 9600 GT (499MB) driver: 25896)
In progress >>without cuda<<

WU=129816514
Cuda error: getting factors found: unspecified launch failure
CreateProcess() failed - (NVIDIA GeForce GTS 250 (512MB) driver: 19107)
CreateProcess() failed - (0x36b1) >>without cuda23<<

WU=129806137
Cuda error: getting factors found: unspecified launch failure
Cuda error: cudaEventCreate: out of memory (NVIDIA GeForce 9500 GT (511MB) driver: 19062)
OK (NVIDIA GeForce 8800 GT (497MB) driver: 25896)

WU=129806440
Cuda error: getting factors found: unspecified launch failure
Insufficient available memory on GPU 0. ([2] NVIDIA GeForce 9600M GT (255MB))
In progress

WU=129690184
Cuda error: getting factors found: unspecified launch failure
Cuda error: getting device properties: CUDA driver version is insufficient for CUDA runtime version (NVIDIA GeForce 8800 GS (511MB))
Detected emulator! We can't use that! ( NVIDIA GeForce GTX 260 (895MB))

WU=129692478
Cuda error: getting factors found: unspecified launch failure
CreateProcess() failed - (0x36b1) (NVIDIA GeForce GTS 250 (512MB) driver: 19107)
OK (NVIDIA GeForce 8500 GT (255MB) driver: 25896)

I see frequently message about insufficient memory available, also at the other task I have not pasted above. What is the minimal amount of memory required to run the PPS CUDA? Does it change by WU, or this value could grow and exceed initial value during calculation? Maybe my cards has the same problem, but due to drivers reports it with other message?

As for my case I will move today the card to another host running Windows to exclude drivers and operating system impact.
____________

Thank you very much for explanation for error codes! I may also help somebody in the future.
I will keep monitoring my card and continue to search other people reporting "Cuda error: getting factors found: unspecified launch failure" error.

P.S. So far 108 out of 142 WU completed with success, which sets completion rate at 76%...:-(


____________

After moving GT240 card from hostID:157777 (Ubuntu x64) to hostID:147474 (Windows XP x86) I was able to successfully complete 46 out of 48 tasks. This is 95% comparing to 75% at previous computer. Moreover at new host I have increased shrader speed from 1460MHz to 1650MHz, while meantime lowering memory speed from 4000MHz to 3400MHz. Those two invalid WU were generated with even higher shader speed, so I think that could be the issues. After those I lowered the speed to stable 1650MHz and all WU passed since that time.

I will keep it there running until I complete 100WU to have reasonable representative sample.
So far it does not look like faulty GPU, but to diagnose the root cause there is a still long way to go....
____________

Yes, it did. At factory settings I finished about 4 WU, then decided to increase shrader speed and did another few. Then I crossed 1700MHz range and failed, but not two times in a row. I moved back to 1650MHz and did another 10 or so and finally decided to lower memory speed (since it should not affect the performance) to keep temperature down. Temperature dropped from 64 to 63 Celsius, so not much really...
____________

Is there any possibility to independently change speed of core, shrader and memory under Linux? I would like to move card back to old host, keeping the same settings as under Windows. Zotec does not provide similar software for Linux and Windows. I checked Nvidia X Server after activating Coolbits, only GPU and memory speed could be changed. I also checked nvclock and it does not support GT 240.
____________