The CUDA 5.5 app version 0.07 published just now should work on both Windows and Linux. Validation with previous versions might not work; but I do hope that we at least have a working CUDA version for Windows again.
My host 12496320 (the one with the HD 530) also has a 'NVIDIA GeForce GTX 1050 Ti (4096MB) driver: 442.74', and has been chewing on v0.05 for a while. It was showing a number of validation errors last night, but you've cleared those up. I'll watch it as the v0.07s start to roll in.
has anyone run with cuda-5.5 application any tasks that have "segment_4" in their name? Did any of those tasks succeed?
The segment_4 tasks have smaller padding factor than the previous ones (down from 3.0 to 1.5) and run time is significantly lower as fft_size has gone down (relatively from 4 to 2.5).
My "anonymous platform" CUDA 11.7 app runs them, but does not validate against NVIDIA OpenCL nor ATI OpenCL.
p.s. I got rid of the random lockup (caused by Trap 11) by increasing memory allocation sizes in the source code. (demod_binary.c). EDIT: My stuck tasks were segment_3 and _4 -tasks.
--
petri33
malloc error on "stuck" Linux OpenCL tasks points in the direction of a memory overrun similar to the CUDA error.
Fatal glibc error: malloc assertion failure in _int_malloc: chunk_main_arena (fwd)
The CUDA 5.5 app version 0.07 published just now should work on both Windows and Linux. Validation with previous versions might not work; but I do hope that we at least have a working CUDA version for Windows again.
the first two tasks I got straight away with v0.07 cuda55 hung up. one with sig6 and another with sig11.
but after those two I'm processing some more that are going OK. very easy to spot the stuck task since it only loads about 250MB to the GPU then stops. whereas successful tasks will load up to the ~1GB amount.
The CUDA 5.5 app version 0.07 published just now should work on both Windows and Linux. Validation with previous versions might not work; but I do hope that we at least have a working CUDA version for Windows again.
Doesn't work. I'm reverting back to 0.03 for CUDA to see whether this still works (and validates).
On my stderr output there is a list of found signal candidates and some meaningful data. It would be nice to have them reported by other versions too. The fA is the sorting parameter. n_harm tells how many harmonics are on the sum (power).
// write all candidates with power > 0.0 to file (and stderr)
int counter = 0;
while (counter < uvar.Ncand && candidates_all[0].fA > 0.0)
The CUDA 5.5 app version 0.07
)
The CUDA 5.5 app version 0.07 published just now should work on both Windows and Linux. Validation with previous versions might not work; but I do hope that we at least have a working CUDA version for Windows again.
BM
My host 12496320 (the one
)
My host 12496320 (the one with the HD 530) also has a 'NVIDIA GeForce GTX 1050 Ti (4096MB) driver: 442.74', and has been chewing on v0.05 for a while. It was showing a number of validation errors last night, but you've cleared those up. I'll watch it as the v0.07s start to roll in.
petri33 wrote: Hi, has
)
malloc error on "stuck" Linux OpenCL tasks points in the direction of a memory overrun similar to the CUDA error.
Fatal glibc error: malloc assertion failure in _int_malloc: chunk_main_arena (fwd)Bernd Machenschalk
)
the first two tasks I got straight away with v0.07 cuda55 hung up. one with sig6 and another with sig11.
https://einsteinathome.org/task/1341567606
https://einsteinathome.org/task/1341567520
but after those two I'm processing some more that are going OK. very easy to spot the stuck task since it only loads about 250MB to the GPU then stops. whereas successful tasks will load up to the ~1GB amount.
_________________________________________________________________________
petri33 wrote: My "anonymous
)
One of my tasks (linux OpenCL) just validated against one of yours.
https://einsteinathome.org/workunit/667201142
Bernd Machenschalk
)
Doesn't work. I'm reverting back to 0.03 for CUDA to see whether this still works (and validates).
BM
still seeing stuck tasks with
)
still seeing stuck tasks with v0.03 cuda55
https://einsteinathome.org/task/1345488519
_________________________________________________________________________
So far on my system, it looks
)
So far on my system, it looks like a O% validation rate on AMD Linux hosts validated against any Windows host:
https://einsteinathome.org/host/12883788/tasks/5/57
The sample size is pretty small, but the few tasks that have validated were validated against linux hosts (opencl-ati and cuda55).
https://einsteinathome.org/workunit/667201859
Thanks TICTOC! Nice to
)
Thanks TICTOC!
Nice to know I'm doing something right.
On my stderr output there is a list of found signal candidates and some meaningful data. It would be nice to have them reported by other versions too. The fA is the sorting parameter. n_harm tells how many harmonics are on the sum (power).
{
...
logMessage(info, false, "CAND: %10.6f %11.6f %8.7f %8.7f %9.6f %9.6f %2d\n",
candidates_all[0].f0*res_factor, candidates_all[0].P_b,
candidates_all[0].tau, candidates_all[0].Psi,
candidates_all[0].power, candidates_all[0].fA,
candidates_all[0].n_harm);
Another BRP7 app version
)
Another BRP7 app version release (0.08) - mainly to test the Windows CUDA version whether it works at all.
BM