CUDALucas (a.k.a. MaclucasFFTW/CUDA 2.3/CUFFTW)

[flashjh]

Ideas?

[owftheevil]

Originally Posted by flashjh

r46: -memtest 56 1
.
.
.
I timed the last group. CUDALucas says 2:48 elapsed, real time was 2:37.9
.
.
.
CUDALucas Time: 5 sec, Timed: 5.8 sec

Its not quite reporting the elapsed time, but rather, how long to finish the test if the rest is done at the same rate as what has already been done, so you will see some differences here, especially at the beginning.

[owftheevil]

New version up at sourceforge. The memory allocations are split into two chunks. On Windows, ~15% slowdown occurs only at the first report after switching chunks.

Also, the problem with gaps in the *fft.txt files is fixed. There should not be anymore "fft to large" errors.

[Manpowre]

Originally Posted by owftheevil

New version up at sourceforge. The memory allocations are split into two chunks. On Windows, ~15% slowdown occurs only at the first report after switching chunks.

Also, the problem with gaps in the *fft.txt files is fixed. There should not be anymore "fft to large" errors.

you are amazing!

my titan node is back up. corsair 860w modular PSU.. pretty good one.

I found a screw in the old psu. no wonder why it stopped working! lol

Ill download the latest code and give it a good go tonight!

[flashjh]

I'm having a problem with CUDALucas stopping:

 
C:/CUDA/CuLu/src/CUDALucas.cu(372) : cudaSafeCall() Runtime API error 30: unknow
n error.

No specific error. Sometimes it goes for hours, sometimes minutes. Any ideas?

[Manpowre]

Originally Posted by flashjh

I'm having a problem with CUDALucas stopping:

 
C:/CUDA/CuLu/src/CUDALucas.cu(372) : cudaSafeCall() Runtime API error 30: unknow
n error.

No specific error. Sometimes it goes for hours, sometimes minutes. Any ideas?

On my new node, which had heat issues (Not in production yet), when the temperature gets close to 90 degrees or above 90 degrees, cuda just dont want to calculate anymore. could it be that ? checked your gpu temperature recently ? dust ? just ideas, as I cleaned my titans and 590s during weekend to prevent this from happening.

( I have put the new codebase on test on the new node)..

[flashjh]

Everything is clean. Temps stay at 72 under full load.

With MSVS, the debug build won't run from MSVS because it doesn't find the worktodo.txt file. Have you run debug from MSVS?

Since it doesn't actually crash I haven't had the option to debug. I know it says unknown error, but what line can we add to output more error info from safecall?

[Manpowre]

Originally Posted by flashjh

Everything is clean. Temps stay at 72 under full load.

With MSVS, the debug build won't run from MSVS because it doesn't find the worktodo.txt file. Have you run debug from MSVS?

Since it doesn't actually crash I haven't had the option to debug. I know it says unknown error, but what line can we add to output more error info from safecall?

Yes, there is a parameter you either need to change to your directory with worktodo, or put worktodo in the project directory, same dir as the sln file I guess.. just try the different directories.. also the .ini file needs to go there..

but there is a setting for default working directory...

[flashjh]

I put the files there, but it still doesn't find them... something I'm doing wrong. My classes all end this week as the students take finals, hopefully I can find some time to look into it.

[Manpowre]

Originally Posted by flashjh

I put the files there, but it still doesn't find them... something I'm doing wrong. My classes all end this week as the students take finals, hopefully I can find some time to look into it.

Ill check it when I get home later today..

[owftheevil]

Originally Posted by flashjh

I'm having a problem with CUDALucas stopping:

 
C:/CUDA/CuLu/src/CUDALucas.cu(372) : cudaSafeCall() Runtime API error 30: unknow
n error.

No specific error. Sometimes it goes for hours, sometimes minutes. Any ideas?

Thats the memcopy timing out because one of the previous cufft calls has hung. A problem since 300++ drivers came about.