Hi ,alexhiggins732

CUDALucas.cu:
//# define RINT(x) (floor(x+0.5))
# define RINT(x) (((x) + A ) - B)

Please replace RINT(). floor is slow but safety.

[QUOTE=msft;253478]Hi ,alexhiggins732

CUDALucas.cu:
//# define RINT(x) (floor(x+0.5))
# define RINT(x) (((x) + A ) - B)

Please replace RINT(). floor is slow but safety.[/QUOTE]

Thanks for your work and the fast reply.

Ok, now the hex string is gone for 216149. Is this correct?


[QUOTE]CUDALucas.1.2>cudalucas 216149 -t
Iteration 10000 M( 216149 )P, n = 524288, CUDALucas v1.2[/QUOTE]Good thing is 18760031 matches your check file.:smile:
[QUOTE]CUDALucas.1.2>cudalucas 18760031
Iteration 10000 M( 18760031 )C, 0x6f3b0ca04650a35d, n = 1048576, CUDALucas v1.2[/QUOTE]How can I tell if the final result is good? It should match the LL result on the exponent page?


Again thanks for your great work!

Edit.

I modifed CUDALucas.cu to
[CODE]#if defined(__x86_32__)
#define RINT(x) (floor(x+0.5))
#else
#define RINT(x) (((x) + A ) - B)
#endif
[/CODE]

I have attached the source code with make files for win32, win64, and Linux and added a short README.txt with basic instructions on how to compile on each of the systems. Feel free to include these in your distribution.

The attachment also contains win32 binaries.

Please advise how to get this to work.

mfaktc is working so the cuda drivers are there.

cufft64 and cudart64 are both in the directory.

Thanks

D:\Data\Mersenne Primes\cudalucas.1.0b.winx64>cudalucas 26826383
CUDALucas.cu(526) : cufftSafeCall() CUFFT error.

D:\Data\Mersenne Primes\cudalucas.1.0b.winx64>cudalucas -c10000 26826383
CUDALucas.cu(526) : cufftSafeCall() CUFFT error.

what gpu do you have?
It probably isn't possible to run primality tests on it since it doesn't support double precision(same as mine:smile:).
But you can still do sieving/TF on it, things for which the cpu/gpu speedup is much higher than for primality tests.

Geforce GTX 460 768MB. Don't tell me that can't do DP.

Of course you can. Only < sm_13 cant do DP FP.

[QUOTE=vsuite;253488]Please advise how to get this to work.

mfaktc is working so the cuda drivers are there.

cufft64 and cudart64 are both in the directory.

Thanks

D:\Data\Mersenne Primes\cudalucas.1.0b.winx64>cudalucas 26826383
CUDALucas.cu(526) : cufftSafeCall() CUFFT error.

D:\Data\Mersenne Primes\cudalucas.1.0b.winx64>cudalucas -c10000 26826383
CUDALucas.cu(526) : cufftSafeCall() CUFFT error.[/QUOTE]


Looking at this post [url]http://forums.nvidia.com/index.php?showtopic=100557[/url]

the problem was caused by an invalid SDK version/driver version combo.

Please run mfaktc and post your cuda info as I did in this thread above. [url]http://www.mersenneforum.org/showpost.php?p=253477&postcount=429[/url]

I think I get it. The machine is elsewhere, so I'll check in the morning, but what CUDA version/toolkit do I need for Windows 7 64-bit with CudaLucas1.0b, and what do I need for Windows XP 32 bit with CudaLucas1.2?

I upped the 32-bit machine to 32_32_16 to get mfaktc1.5 Win 32 to work.

Thanks.

To solve it, I installed both the 32_32_16 toolkit (needed for the latest mfaktc) and 32_31_9 toolkit (needed for CudaLucas).

I copied both cudart and cufft files to the mfaktc and cudalucas folders, though iirc cudalucas also worked without copying the files.

CudaLucas seems to use very little CPU, like 2%. So I have 1 thread of a Athlon 64 X2 running mfaktc, one threadlet running cudalucas, and one thread running prime95.

New question. Can CudaLucas1.2.Win32 be reasonably used on Windows 7, 64-bit? (seems to be running, will see if/when it finishes the current assignment.)

Can someone who has 3.1 toolkit and <260.xx drivers please compile the latest CUDALucas for windows ?
Both 64 and 32 bit binaries will do fine.

GPU/CPU Usage
 

A single instance of mfaktc uses about 53% of the GPU (and a full core, ie 25%), and I can run two instances, to increase the overall M/sec, pushing the GPU to 100%, but reducing the M/sec for each instance.

CudaLucas appears to use minimal CPU resources eg 2% of a Q6600 quad-core, but maxes out (~99%) the GPU.

Does it make sense to run multiple instances of CudaLucas to increase overall throughput, or is the GPU really maxed out?

Thanks.