Hi, there appears to be some issue with library compatibility with the PPSieve Mac CUDA App (Version 1.24) - can any Mac users who have tried these WUs please post here if you are willing to carry out a few tests for me. Also, if you have successfully completed any of the WUs also please post here. Please include your OS version, GPU type, CUDA driver version, and the results of running the command "ls /usr/local/cuda/lib" in the Terminal.

Thanks!

- Iain

I'm happily churning out valid cuda ppsieve WUs with 1.24 on my iMac, OSX 10.6.4, GT120 (256MB), cuda driver version 3.1.10, GPU driver version 1.6.16.11 (19.5.8f01).
% ls -l /usr/local/cuda/lib
total 166920
-rwxr-xr-x 1 root wheel 46131124 Mar 11 13:25 libcublas.dylib*
-rwxr-xr-x 1 root wheel 24796928 Mar 11 13:25 libcublasemu.dylib*
-rwxr-xr-x 1 root wheel 32816 Jun 17 15:35 libcuda.dylib*
-rwxr-xr-x 1 root wheel 252424 Mar 11 13:25 libcudart.dylib*
-rwxr-xr-x 1 root wheel 274908 Mar 11 13:25 libcudartemu.dylib*
-rwxr-xr-x 1 root wheel 12273808 Mar 11 13:25 libcufft.dylib*
-rwxr-xr-x 1 root wheel 1673444 Mar 11 13:25 libcufftemu.dylib*
-rwxr-xr-x 1 root wheel 10172 Mar 11 13:25 libtlshook.dylib*

--Gary

I've build a new version of the app that should be CUDA 2.3 compatible, but it was a bit of a hack, so should be tested before release.

Can you please download the following two versions of the app

http://www.pyramid-productions.net/downloads/ppsieve_cuda23.tar.gz
http://www.pyramid-productions.net/downloads/ppsieve_cuda31.tar.gz

Unzip and run them at the command line using:

./primegrid_ppsieve_1.25_i686-apple-darwin__cuda23 -p42070e9 -P42070003e6 -k 1201 -K 9999 -N 2000000 -z normal (or __cuda31).

Please post the version of CUDA you have installed (BOINC reports it when it starts in the 'Messages' Log, and the results of both the apps.

I hope that if you have the CUDA 2.3 or 3.0 driver installed, the first app will work, but if you have the 3.1 driver then both apps will work!

Thanks

- Iain

I should also add, that in order to make this test exactly like the BOINC distribution please also download the required libraries from
http://www.pyramid-productions.net/downloads/libs_23.tar.gz
http://www.pyramid-productions.net/downloads/libs_31.tar.gz

The libraries should be extracted into the same directory as the executable, and then do

export DYLD_LIBRARY_PATH=.:$DYLD_LIBRARY_PATH before running the app as above.

Thanks

- Iain

Hmmm... not so much luck with the 2.3 libraries/app; 3.1 seems fine. I completely shut down BOINC before running anything. BOINC client reports "CUDA version 3010" at startup; the CUDA preferences app reports "CUDA Driver Version: 3.1.10 GPU Driver Version: 1.6.16.11 (19.5.8f01)".

--Gary

% echo $DYLD_LIBRARY_PATH
.
% gunzip -c libs_23.tar.gz | tar xvf -
x libcudart.dylib
x libtlshook.dylib
% ./primegrid_ppsieve_1.25_i686-apple-darwin__cuda23 -p42070e9 -P42070003e6 -k 1201 -K 9999 -N 2000000 -z normal
ppsieve version cuda-0.1.5 (testing)
Compiled Aug 12 2010 with GCC 4.2.1 (Apple Inc. build 5664)
nstart=76, nstep=32, gpu_nstep=32
ppsieve initialized: 1201 <= k <= 9999, 76 <= n <= 2000000
% cat stderr.txt
shmget in attach_shmem: Invalid argument
12:54:21 (6730): Can't set up shared mem: -1. Will run in standalone mode.
Sieve started: 42070000000000 <= p < 42070003000000
Thread 0 starting
Detected GPU 0: Device Emulation (CPU)
Detected compute capability: 9999.9999
Detected 16 multiprocessors.
Insufficient available memory on GPU 0.
12:54:21 (6730): called boinc_finish
% rm stderr.txt
% gunzip -c libs_31.tar.gz | tar xvf -
x libcudart.dylib
x libtlshook.dylib
% ./primegrid_ppsieve_1.25_i686-apple-darwin__cuda31 -p42070e9 -P42070003e6 -k 1201 -K 9999 -N 2000000 -z normal
ppsieve version cuda-0.1.5 (testing)
Compiled Aug 12 2010 with GCC 4.2.1 (Apple Inc. build 5664)
nstart=76, nstep=32, gpu_nstep=32
ppsieve initialized: 1201 <= k <= 9999, 76 <= n <= 2000000
42070000070587 | 9475*2^197534+1
42070000198537 | 3373*2^1046686+1
Found 2 factors
% cat stderr.txt
shmget in attach_shmem: Invalid argument
12:55:27 (6735): Can't set up shared mem: -1. Will run in standalone mode.
Sieve started: 42070000000000 <= p < 42070003000000
Thread 0 starting
Detected GPU 0: GeForce GT 120
Detected compute capability: 1.1
Detected 4 multiprocessors.

Thread 0 completed
Sieve complete: 42070000000000 <= p < 42070003000000
count=95668,sum=0x37dacb7121ccffe4
Elapsed time: 26.76 sec. (0.02 init + 26.74 sieve) at 117642 p/sec.
Processor time: 0.33 sec. (0.03 init + 0.30 sieve) at 10492405 p/sec.
Average processor utilization: 1.49 (init), 0.01 (sieve)
12:55:54 (6735): called boinc_finish
%

It looks like with the 2.3 drivers it detected the device emulation capability. Try 2.3 again with "--device 1".
____________

No joy:

% echo $DYLD_LIBRARY_PATH
.
% gunzip -c libs_23.tar.gz | tar xvf -
x libcudart.dylib
x libtlshook.dylib
% ./primegrid_ppsieve_1.25_i686-apple-darwin__cuda23 -p42070e9 -P42070003e6 -k 1201 -K 9999 -N 2000000 -z normal --device 1
ppsieve version cuda-0.1.5 (testing)
Compiled Aug 12 2010 with GCC 4.2.1 (Apple Inc. build 5664)
nstart=76, nstep=32, gpu_nstep=32
ppsieve initialized: 1201 <= k <= 9999, 76 <= n <= 2000000
% cat stderr.txt
shmget in attach_shmem: Invalid argument
17:58:33 (18607): Can't set up shared mem: -1. Will run in standalone mode.
Sieve started: 42070000000000 <= p < 42070003000000
Thread 0 starting
GPU 1 not compute-capable.
Cuda error: getting device properties: invalid device ordinal
17:58:33 (18607): called boinc_finish
%

--Gary

I've had at least four CUDA WUs error out over the last few days, for example: http://www.primegrid.com/result.php?resultid=188733631. So far as I can tell, this started when we switched to the 1.29 version. It seems to run to, or nearly to, completion based on the run time. Other 1.29 WUs over the last few days have completed and validated OK. I don't think I had this problem on the prior version; was happily crunching on GPU and CPU.
--Gary

The new 1.30 version for OS X is running great; 100% of WUs are completing successfully; I was having about a 50% failure rate with 1.29. Run time on my (relatively lame) GTX 120 has dropped from about 1:47:00 to about 1:02:00. Cool! Thanks Iain and Rytis and anyone else involved.
--Gary