Dear David,
Thanks for your comment. I'm going to try installation on a different machine.
Here,
"By default, typing "make test.cuda.serial" tests the DPFP version. If you need more fine grained testing, go to the "test" folder, and examine the test_amber_cuda_serial.sh script, where you can specify which precision model you want."
According to the section 20.6.4 Accuracy Considerations of AMBER20 manual. The default precision model used when running on GPUs is the hybrid SPFP.
The test.cuda.serial test DPFP, instead.
I would like to double check the default GPU runnings use SPFP or DPFP?
I tested SPFP, similar large errors reported on igb 7 and 8 on EPtot values. (there are other a few errors at e-04 to e-07 on runnings like NVT, NVE, etc..)
I will avoid CUDA running on these calculations. I will try on other cluster to check results...I agree there are not too much I can do with current settings.. Thank you David for answering my questions and commenting the reports. Appreciate that!
Best,
Guqin
-----邮件原件-----
发件人: David A Case [mailto:david.case.rutgers.edu]
发送时间: 2021年5月7日 20:31
收件人: AMBER Mailing List <amber.ambermd.org>
主题: [ambermd.org代发]Re: [AMBER] Amber20 CUDA.serial tests comparison failed with gb calculation: differences at e-01 level
On Fri, May 07, 2021, 石谷沁 wrote:
>By the way, I tried “make test.cuda.serial.SPFP”, but program told me
>there is no rule to follow:
>make: *** No rule to make target 'test.cuda.serial.SPFP'. Stop.
If you set -DCUDA=TRUE, cmake will build both the SPFP and DPFP versions.
By default, typing "make test.cuda.serial" tests the DPFP version. If you need more fine grained testing, go to the "test" folder, and examine the test_amber_cuda_serial.sh script, where you can specify which precision model you want.
Have you been able to try intalling and testing on a different machine?
We have successful GPU results on CentOS7 with CUDA SDK 10.1, but I don't know what GPU hardware was involved. (cc-ing to Gerald).
I have tested a quadro RTX5000 with CUDA STK 10.1 on an Ubuntu machine.
Until we can reproduce the problem, we are unlikely to make much progress.
If anyone on the list is using CentOS7 with GPU, please report your CUDA SDK version, GPU hardware, and success/failure on the cuda test suite. Maybe that will help narrow down what combination of OS, CUDA SDK, and GPU hardware is involved here.
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
***********免责声明*************
本电子邮件中包含的信息仅供指定的或授权的个人或团体使用。本电子邮件及附件中提到的信息可能是保密信息或者法律特许保密的信息。如果你不是指定收件人，对于邮件内容的任何披露、复制、散布或者任何针对邮件内容进行的行为都是违法行为，需要严格禁止。如果您误收该电子邮件，请立即通知本公司并从您的系统中删除全部原始信息。该邮件可能会对您的系统或者数据造成损坏，对此我公司不承担任何责任。除非与公司业务有关，否则本邮件中的观点、结论、或者其它包含在邮件中的信息均为发件人个人行为，并不代表我公司。我公司有权保留对收发邮件的监控权利。
***********Business Email Disclaimer**************
This e-mail and any attachments are meant for the intended recipient only and may contain information belonging to Qilu Pharma that is privileged, confidential, proprietary, and/or otherwise protected or prohibited from disclosure. If you are not the correct recipient or received this e-mail erroneously, please inform the sender immediately and delete this mail from your system. Qilu Pharma state no liability for any damage to your system and data caused by this email. Unless this email is related to the business with the company, otherwise any views or opinions presented in this email are solely from the sender. Qilu Pharma has the right to monitor the sending and receiving of the e-mail.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri May 07 2021 - 20:30:02 PDT
