[QUOTE=TheJudger;255085]limited by the speed of your CPU (average wait is very small)[/QUOTE]I'm happy(?) to report I don't have that problem... I wish I did :wink:[code] class | candidates | time | avg. rate | SievePrimes | ETA | avg. wait
1348/4620 | 18.79G | 589.96s | 31.85M/s | 100000 | 4d15h | 1640us[/code]

I merely report the speed.Another more expansive solution would be to get a better CPU.. ;p

I get this error with 0.16:
running a simple selftest...
ERROR: selftest failed for M49635893
no factor found
ERROR: selftest failed for M51375383
no factor found
ERROR: selftest failed for M47644171
no factor found
ERROR: selftest failed for M51038681
no factor found
ERROR: selftest failed for M53076719

It's reproducable. It always fails on those 5 candidates. GPU is a bit old: GeForce 9800 GTX/9800 GTX+. Works great with 0.14.

Any ideas?

[QUOTE=opyrt;255127]I get this error with 0.16:
running a simple selftest...
ERROR: selftest failed for M49635893
no factor found
ERROR: selftest failed for M51375383
no factor found
ERROR: selftest failed for M47644171
no factor found
ERROR: selftest failed for M51038681
no factor found
ERROR: selftest failed for M53076719

It's reproducable. It always fails on those 5 candidates. GPU is a bit old: GeForce 9800 GTX/9800 GTX+. Works great with 0.14.

Any ideas?[/QUOTE]

The same is happening to me, GTX275, CUDA 3.0 and Linux 64.

We are investigating...

Luigi

opyrt: which OS / driver version / CUDA version / ...? Precompiled binary or build from source?

Good to know it is reproduceable for you, this lowers the chance for a race condition...

Quicky on a crappy Quadro NVS 290 (G86 chip), similar to a Geforce 8400GS (1st rev) or 8500GT:

[CODE]mfaktc v0.16
...
CUDA device info
name Quadro NVS 290
compute capability 1.1
maximum threads per block 512
number of multiprocessors 2 (16 shader cores)
clock rate 918MHz

CUDA version info
binary compiled for CUDA 3.20
CUDA driver version 3.20
CUDA runtime version 3.20
...
running a simple selftest...

Selftest statistics
number of tests 31
successfull tests 31

selftest PASSED!
[/CODE]

To be honest: right now I've no clue whats wrong. During betatest James Heinrich reported no issues on his 8800GT (cc 1.1, too) using my precompiled binaries.

For now we've to collect data on which systems it works and on which not. :sad:

---
Edit: reproduced by downgrading the CUDA toolkit from 3.2 to 3.0 on Linux.

Oliver

[QUOTE=TheJudger;255136]During betatest James Heinrich reported no issues on his 8800GT (cc 1.1, too) using my precompiled binaries.[/QUOTE]
On Windows, mind you. This may be a *nix-specific issue? I haven't started running v0.16-final yet (93h left to go on current assignment before I can switch over), but I did fire it up once and it passed the self-test just fine.

This is my CUDA info:
[CODE]CUDA device info
name GeForce 9800 GTX/9800 GTX+
compute capability 1.1
maximum threads per block 512
number of multiprocessors 16 (128 shader cores)
clock rate 1850MHz

CUDA version info
binary compiled for CUDA 3.10
CUDA driver version 3.20
CUDA runtime version 3.20
[/CODE]OS is:
[CODE]Linux 2.6.32-29-server #58-Ubuntu SMP Fri Feb 11 21:06:51 UTC 2011 x86_64 GNU/Linux
Ubuntu 10.04.2 LTS
[/CODE]Let me know if you need me to test anything.

I think I got it.

I installed CUDA 3.2 on my computer, did a "make clean; make" to recompile mfaktc. Info at start-up:[CODE]CUDA device info
name GeForce 9800 GTX/9800 GTX+
compute capability 1.1
maximum threads per block 512
number of multiprocessors 16 (128 shader cores)
clock rate 1850MHz

CUDA version info
binary compiled for CUDA 3.20
CUDA driver version 3.20
CUDA runtime version 3.20

Automatic parameters
threads per grid 1048576

running a simple selftest...
Selftest statistics
number of tests 31
successfull tests 31

selftest PASSED!
[/CODE]It seems this is an issue when mfaktc is compiled on CUDA 3.1 computers.

Impressive
 

mfaktc 0.16: Overall Throughput > 200M/s (see attachment)

Intel Core i5-750 @ 3600 MHz (4 physical/logical cores)
Nvidia Geforce 560 Ti @ 823 MHz (384 CUDA cores)

Danke Oliver

somewhat less impressive ...
 

mfaktc 0.16: Overall Throughput ~ 26 M/s (see attachment)

Brain's i5 / Geforce 560 TI is much faster, runs hotter, too, though!

I ran the same exponent as Brain's second instance, obtaining half the throughput for that instance (only took one mfaktc instance to fully load my GPU).

Intel Core i7-840QM @ 1.87 GHz (nominal) (4 cores, 8 logical CPUs)
Nvidia Quadro FX 2800M @ 600 MHz GPU core clock (96 CUDA cores), 999 MHz GPU memory clock, 1500 MHz GPU shader clock

+++++

Thank you to James Heinrich for providing me a copy of "cudart64_32_16.dll" (per post 588, [URL]http://www.mersenneforum.org/showpost.php?p=253371&postcount=588[/URL]).

+++++

drh ([URL="http://www.mersenneforum.org/showpost.php?p=255082&postcount=636%29:"]http://www.mersenneforum.org/showpost.php?p=255082&postcount=636):[/URL]

[QUOTE=drh;255082]I get it right after I try to run mfaktc -st ... mfaktc -tt runs fine, as far as I can tell. I get cudaSetDevice(0) failed. I also get it if I try to run the program with a -tf ...

Thanks,
Doug[/QUOTE]

Had same problem myself this morning. Solution:

Go here ([URL]http://developer.nvidia.com/object/cuda_3_2_downloads.html[/URL]) and get the developer drivers (notebook drivers 260.99) for your operating system, turn off your antivirus, install driver, turn antivirus back on.

[QUOTE=S34960zz;255187]Thank you to James Heinrich for providing me a copy of "cudart64_32_16.dll" (per post 588, [URL]http://www.mersenneforum.org/showpost.php?p=253371&postcount=588[/URL]).[/QUOTE]Actually, look at [url=http://www.mersenneforum.org/showpost.php?p=255068&postcount=632]my more recent post[/url] instead, where I actually attached the file; no need to PM/email me if you need the DLL.