[gmx-users] GROMACS 5.0.5 GPU version on K620
Szilárd Páll
pall.szilard at gmail.com
Wed Jul 22 23:02:38 CEST 2015
Krzysztof,
Thanks for getting back. I was interested in the -version output of the
problematic binary hoping to pinpoint the source of the issues you reported
- especially because that error should not have occurred to begin with
(unless you changed compiler flags manually!).
The poor performance of the card is no surprise, the K620 is a rather
low-end GPU (and currently GROMACS does not implement partial nonbonded
offload).
--
Szilárd
On Wed, Jul 22, 2015 at 12:16 AM, Krzysztof Kuczera <kkuczera at ku.edu> wrote:
> Hi Szilárd ,
>
> The test case worked after I added the flags you recommended, though the
> speed was not as high as I had hoped - equivalent to about 6 CPU cores - I
> guess my GPU is not so hot.
>
> Below is the output from "mdrun -version". Please let me know if you see
> things that might be optimized there.
>
> Thanks for your excellent help!
>
> Krzysztof
> -------------------------------
> Gromacs version: VERSION 5.0.5
> Precision: single
> Memory model: 64 bit
> MPI library: thread_mpi
> OpenMP support: enabled GPU support: enabled invsqrt routine:
> gmx_software_invsqrt(x)
> SIMD instructions: AVX2_256
> FFT library: fftw-3.3.4-fma-sse2-avx RDTSCP usage: enabled C++11
> compilation: disabled TNG support: enabled
> Tracing support: disabled
> Built on: Tue Jul 21 16:25:29 CDT 2015
> Built by: kuczera at lolipop-chem-ku-edu [CMAKE]Build OS/arch:
> Linux 3.10.0-229.4.2.el7.x86_64 x86_64
> Build CPU vendor: GenuineIntel
> Build CPU brand: Intel(R) Xeon(R) CPU E5-2687W v3 @ 3.10GHzBuild CPU
> family: 6 Model: 63 Stepping: 2Build CPU features: aes apic avx avx2
> clfsh cmov cx8 cx16 f16c fma htt lahf_lm m
> mx msr nonstop_tsc pcid pclmuldq pdcm pdpe1gb popcnt pse rdrnd rdtscp sse2
> sse3 sse4.1 sse4.2 ssse3 tdt x2apicC compiler: /usr/bin/cc GNU 4.8.3C
> compiler flags: -march=core-avx2 -Wno-maybe-uninitialized -Wextra
> -Wno-missing-field-initializers -Wno-sign-compare -Wpointer-arith -Wall
> -Wno-unused -Wunused-value -Wunused-parameter -O3 -DNDEBUG
> -fomit-frame-pointer -funroll-all-
> loops -fexcess-precision=fast -Wno-array-bounds
> C++ compiler: /usr/bin/c++ GNU 4.8.3
> C++ compiler flags: -march=core-avx2 -Wextra
> -Wno-missing-field-initializers -Wpointer-arith -Wall -Wno-unused-function
> -O3 -DNDEBUG -fomit-frame-pointer -funroll-all-loops
> -fexcess-precision=fast -Wno-array-bounds
> Boost version: 1.53.0 (external)
> CUDA compiler: /usr/local/cuda-7.0/bin/nvcc nvcc: NVIDIA (R) Cuda
> compiler driver;Copyright (c) 2005-2015 NVIDIA Corporation;Built on
> Mon_Feb_16_22:59:02_CST_2015;Cuda compilation tools, release 7.0, V7.0.27
> CUDA compiler
> flags:-gencode;arch=compute_20,code=sm_20;-gencode;arch=compute_20,code=sm_21;-gencode;arch=compute_30,code=sm_30;-gencode;arch=compute_35,code=sm_35;-gencode;arch=compute_35,code=compute_35;-gencode;arch=compute_50,code=compute_50;-use_fast_math;-Xcompiler;-fPIC
> ;
> ;-march=core-avx2;-Wextra;-Wno-missing-field-initializers;-Wpointer-arith;-Wall;-Wno-unused-function;-O3;-DNDEBUG;-fomit-frame-pointer;-funroll-all-loops;-fexcess-precision=fast;-Wno-array-bounds;
> CUDA driver: 7.0
> CUDA runtime: 7.0
>
>
>
> On 7/17/15 5:39 PM, Szilárd Páll wrote:
>
>> Krzysztof,
>>
>> Whyle GROMACS 5.0.x build system does not explicitly generate options
>> targeting 5.x devices, the binary built should still be compatible with
>> your GPU. You can try adding the device-specific optimization flags with
>> the
>> -DCUDA_NVCC_FLAGS_RELEASE="-gencode;arch=compute_50,code=sm_50"
>> cmake flag, but I'm not sure this fix the issue.
>>
>> What does gmx -version show?
>>
>> Cheers,
>>
>> --
>> Szilárd
>>
>> On Fri, Jul 17, 2015 at 11:49 PM, Krzysztof Kuczera <kkuczera at ku.edu>
>> wrote:
>>
>> Hi Group,
>>>
>>> I am getting a run-time error on my Linux workstation with K620 GPU
>>> for the GPU version of GROMACS 5.0.5
>>> using gcc 4.8.3 and CUDA Toolkit 7.0 I had no problem compiling the
>>> code
>>> but got this error when starting a test case:
>>>
>>> Program mdrun, VERSION 5.0.5
>>> Source code file:
>>> /home/kuczera/prog/gromacs-5.0.5/src/gromacs/mdlib/nbnxn_cuda/
>>> nbnxn_cuda.cu, line: 619
>>>
>>> Fatal error:
>>> cudaStreamSynchronize failed in cu_blockwait_nb: an illegal memory access
>>> was encountered
>>>
>>> Searching the Web, I found that this type of error was already resolved
>>> in
>>> GROMACS 4.6 some time ago.
>>> Could somebody suggest a solution?
>>>
>>> We have been able to compile and run GROMACS 5.0.4 on an older GPU with
>>> compute capability 3.5,
>>> my newer K620 has compute capability 5.0 - could this be the problem ?
>>>
>>> Thanks
>>> Krzysztof
>>>
>>> --
>>> Krzysztof Kuczera
>>> Departments of Chemistry and Molecular Biosciences
>>> The University of Kansas
>>> 1251 Wescoe Hall Drive, 5090 Malott Hall
>>> Lawrence, KS 66045
>>> Tel: 785-864-5060 Fax: 785-864-5396 email: kkuczera at ku.edu
>>> http://oolung.chem.ku.edu/~kuczera/home.html
>>>
>>> --
>>> Gromacs Users mailing list
>>>
>>> * Please search the archive at
>>> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
>>> posting!
>>>
>>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>>
>>> * For (un)subscribe requests visit
>>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
>>> send a mail to gmx-users-request at gromacs.org.
>>>
>>>
>
> --
> Krzysztof Kuczera
> Departments of Chemistry and Molecular Biosciences
> The University of Kansas
> 1251 Wescoe Hall Drive, 5090 Malott Hall
> Lawrence, KS 66045
> Tel: 785-864-5060 Fax: 785-864-5396 email: kkuczera at ku.edu
> http://oolung.chem.ku.edu/~kuczera/home.html
>
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>
More information about the gromacs.org_gmx-users
mailing list