[CP2K-user] [CP2K:16177] Re: The CP2K8.2 GPU version is much slower than CP2K7.1 GPU version

[ChunLin Fang]

Sorry for the disturb, I find out that CP2K7.1 uses openblas and CP2K8.2 
uses lapack, which is the main reason of the performance dropout.

在2021年11月3日星期三 UTC+8 下午5:17:05<Alfio Lazzaro> 写道：

> Please, show the timers at the end of the CP2K run so that we can compare 
> where you are spending time. Then, tell us how you are compiling CP2K. For 
> example, are you considering the GPU grid compilation in 8.2?
>
> Il giorno mercoledì 3 novembre 2021 alle 04:09:31 UTC+1 qiy... at gmail.com 
> ha scritto:
>
>> Hi, Dear CP2K deverlopers
>>         I'm trying to optimizing H2O-256.inp in CP2K7.1, Here is the 
>> benchmark result:
>>   **** **** ******  **  PROGRAM STARTED AT               2021-11-03 
>> 10:05:57.720
>>  ***** ** ***  *** **   PROGRAM STARTED ON                                
>>  hpc42
>>  **    ****   ******    PROGRAM STARTED BY                                
>>   root
>>  ***** **    ** ** **   PROGRAM PROCESS ID                                
>> 877565
>>
>>   **** **** ******  **  PROGRAM ENDED AT                 2021-11-03 
>> 10:14:13.855
>>  ***** ** ***  *** **   PROGRAM RAN ON                                    
>>  hpc42
>>  **    ****   ******    PROGRAM RAN BY                                    
>>   root
>>  ***** **    ** ** **   PROGRAM PROCESS ID                                
>> 877565
>>
>> The whole process takes exactly 9 mins in my 64 core ARM64+NVIDIA A100 
>> machine, The bench command is as following:
>> mpirun --allow-run-as-root -np 64 -x OMP_NUM_THREADS=1 cp2k.psmp 
>> benchmarks/QS/H2O-256.inp > cp2k.H2O-256.inp.log
>>
>> When I switch to the newest CP2K8.2,  The benchmark result shows that 38% 
>> slower compared to the previous result. It takes 13.6 mins!
>> [root hpc42 cp2k-8.2]# time -p mpirun --allow-run-as-root -np 64 -x 
>> OMP_NUM_THREADS=1 cp2k.psmp benchmarks/QS/H2O-256.inp > cp2k.H2O-256.inp.log
>> real 819.39
>> user 45364.02
>> sys 2102.99
>>
>> Can anyone tell me what was going on and how to acclerate H2O-256.inp?
>>
>> PS: CP2K7.1 uses elpa2019(GPU enabled) and CP2K8.2 didn't use elpa 
>> because of an compile error.
>>
>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ab7dd712-f8d9-4b8e-afc1-5b1edfabea70n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20211103/49654354/attachment.htm>