Hi Kanin,
Jason is right here in terms of a number of jobs dying because the card doesn't have enough memory. That said the issue of different hardware giving abnormally large differences on different hardware should have been addressed in AMBER 16. If you are seeing things work on other GPUs my first suspicion would be that your GTX480 is malfunctioning.
download the following
https://www.dropbox.com/s/wmnse8yhyypq28q/GPU_Validation_Test.tar.gz <
https://www.dropbox.com/s/wmnse8yhyypq28q/GPU_Validation_Test.tar.gz>
untar it, edit run_test.x set gpu_count=1, test_count=40 then nohup ./run_test.x &
It will run for about 36 hours and produce a GPU_0.log file. Keep an eye on that log file as it runs - it should end up printing 40 lines where the reported energy is the same on every line. If you see variation between lines then your GPU is toast from a simulation perspective. Donate it to someone who wants to use it to play games - or as a cool doorstop. If all lines match then let me know and we can look more closely at the differences you see in the tests.
All the best
Ross
> On Mar 20, 2017, at 14:29, Jason Swails <jason.swails.gmail.com> wrote:
>
> The CUDA tests do not all pass on all devices, I think (it used to be the case that they all only passed on a single card model, but I'm not sure if that is still the case). Some of the errors make it seem like the 480 doesn't have enough memory for some of the test systems (it is an old mid-level gaming card in the Fermi line IIRC).
>
> The big differences usually correspond to stochastic tests because each card has a different stream of random numbers. As a result the tests seem to differ by a lot, but it doesn't indicate anything is actually broken.
>
> HTH,
> Jason
>
> --
> Jason M. Swails
>
>> On Mar 20, 2017, at 9:46 AM, Kanin Wichapong <kanin.wichapong.gmail.com> wrote:
>>
>> Dear Amber Users,
>>
>> I tried to install Amber 16 on my computer;
>>
>>
>> - GeForce GTX480
>>
>> - NVIDIA Driver Version = 375.39
>>
>> - Cuda version 8.0
>>
>> - gcc (Ubuntu 5.4.0-6ubuntu1~16.04.4) 5.4.0
>>
>>
>> I successfully complied and installed the program.
>>
>> *./configure -cuda gnu*
>>
>> Cleaning the src directories. This may take a few moments.
>>
>> *Configure complete*
>>
>>
>> *source amber.sh*
>>
>> *make install*
>>
>> make[4]: Leaving directory
>> '/home/kanin/Softwares/AMBER16/amber16/src/pmemd/src'
>>
>> make[3]: Leaving directory
>> '/home/kanin/Softwares/AMBER16/amber16/src/pmemd/src'
>>
>> *Installation of pmemd.cuda complete*
>>
>> make[2]: Leaving directory '/home/kanin/Softwares/AMBER16/amber16/src/pmemd'
>>
>> make[1]: Leaving directory '/home/kanin/Softwares/AMBER16/amber16/src'
>>
>>
>> *make test *
>>
>> However, during testing the programs I got several errors; for example,
>>
>> cd cellulose/ && ./Run.cellulose_nvt_256_128_128 DPFP
>> /home/kanin/Softwares/AMBER16/amber16/include/netcdf.mod
>>
>> *cudaMalloc GpuBuffer::Allocate failed out of memory*
>>
>> *./Run.cellulose_nvt_256_128_128: Program error*
>>
>> *Makefile:166: recipe for target 'test.pmemd.cuda.pme.big' failed*
>>
>> *make[3]: [test.pmemd.cuda.pme.big] Error 1 (ignored)*
>>
>> cd cellulose/ && ./Run.cellulose_nvt DPFP
>> /home/kanin/Softwares/AMBER16/amber16/include/netcdf.mod
>>
>> cudaMalloc GpuBuffer::Allocate failed out of memory
>>
>> ./Run.cellulose_nvt: Program error
>>
>> Makefile:166: recipe for target 'test.pmemd.cuda.pme.big' failed
>>
>> make[3]: [test.pmemd.cuda.pme.big] Error 1 (ignored)
>>
>> cd cellulose/ && ./Run.cellulose_npt DPFP
>> /home/kanin/Softwares/AMBER16/amber16/include/netcdf.mod
>>
>> cudaMalloc GpuBuffer::Allocate failed out of memory
>>
>> ./Run.cellulose_npt: Program error
>>
>> Makefile:166: recipe for target 'test.pmemd.cuda.pme.big' failed
>>
>> make[3]: [test.pmemd.cuda.pme.big] Error 1 (ignored)
>>
>> make[3]: Leaving directory '/home/kanin/Softwares/AMBER16/amber16/test/cuda'
>>
>> make[2]: Leaving directory '/home/kanin/Softwares/AMBER16/amber16/test'
>>
>> 124 file comparisons passed
>>
>> 8 file comparisons failed
>>
>> 15 tests experienced errors
>>
>>
>> and when I checked the file.diff there are a big difference in the values;
>> for example,
>>
>> 471c471
>>
>> < EKCMT = 8798.1084 VIRIAL = 19449.8548 VOLUME = 366869.6118
>>
>>> EKCMT = 8793.6574 VIRIAL = 19292.8185 VOLUME = 366873.2544
>>
>> 475c475
>>
>> < NSTEP = 40 TIME(PS) = 50.040 TEMP(K) = 2.27 PRESS = 40.2
>>
>>> NSTEP = 40 TIME(PS) = 50.040 TEMP(K) = 2.11 PRESS = 32.6
>>
>> 476c476
>>
>> < Etot = 19.4103 EKtot = 149.8506 EPtot = 137.3725
>>
>>> Etot = 25.6525 EKtot = 139.3085 EPtot = 126.1566
>>
>> 477c477
>>
>> < BOND = 19.8977 ANGLE = 38.3546 DIHED = 11.8730
>>
>>> BOND = 20.3404 ANGLE = 36.5091 DIHED = 12.9929
>>
>> 478c478
>>
>> < 1-4 NB = 6.2284 1-4 EEL = 12.2396 VDWAALS = 46.4874
>>
>>> 1-4 NB = 7.9763 1-4 EEL = 12.9424 VDWAALS = 50.6061
>>
>> 479c479
>>
>> < EELEC = 87.3490 EHBOND = 0. RESTRAINT = 170.2276
>>
>>> EELEC = 79.7127 EHBOND = 0. RESTRAINT = 169.9625
>>
>> 480c480
>>
>> < EAMBER (non-restraint) = -32.8551
>>
>>> EAMBER (non-restraint) = -43.8060
>>
>> 481c481
>>
>> < EKCMT = 23.0457 VIRIAL = 327.6438 VOLUME = 251.4357
>>
>>> EKCMT = 30.6251 VIRIAL = 272.3865 VOLUME = 248.0353
>>
>>
>> I successfully installed AMBER16 by using the same compiler, same version
>> of NVIDIA driver, CUDA, and Ubuntu but on different GPUs; Tesla C2075,
>> Quadro 6000, and GeForce GTX 1080 without any problems. So I think, it
>> might be because of GTX480 or not. Any suggestions to solve this problems.
>> Thank you in advance.
>>
>>
>> Best Regards,
>>
>> Kanin Wichapong
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Mar 21 2017 - 17:30:03 PDT
