Makefile and -arch=sm_30

[max...@protonmail.com]

Hello,

I am actually trying to compile mcx on a remote server that we use to do heavy computation. 

I never had any problem doing that before, but yesterday I got this error

nvcc -c -g -lineinfo -Xcompiler -Wall -Xcompiler -fopenmp -m64 -DSAVE_DETECTORS -use_fast_math -arch=sm_30 -DMCX_TARGET_NAME='"Fermi MCX"' -DUSE_ATOMIC -use_fast_math -DSAVE_DETECTORS -o mcx_core.o  mcx_core.cu

nvcc fatal   : Value 'sm_30' is not defined for option 'gpu-architecture'

I took a look around and didn't really find any good informations. Only that maybe it can be because of CUDA 11.0

I tried to take a look at the Makefile to make the compilation not using the arch option but I'm a bit lost.

So I was wondering if it could be a "bug" from the Makefile and it need to be updated or if it's on my side and I have to contact the server support.

Also There some hardware informations

• Server (Don't Work)
  • Cuda compilation tools, release 11.0, V11.0.194
  • NVIDIA T4 
• Personal PC (Where it work)
  • Cuda compilation tools, release 9.1, V9.1.85
  • GPU: NVIDIA GeForce GTX 1060 3GB

Thank you in advance (If I found a solution will comeback and Update my post)

[Shijie Yan]

I had the same issue with CUDA 11. I think it is because sm_30 is no longer supported in the latest CUDA version. Please see https://docs.nvidia.com/cuda/archive/11.0/cuda-toolkit-release-notes/, "Support for Kepler sm_30 and sm_32 architecture based products is dropped." I modified the Makefile and replace "sm_30" with "sm_35" and I can still compile and run.

Hope it helps.

Shijie

Release Notes :: CUDA Toolkit Documentation This section summarizes the changes in CUDA 11.0 Update 1 since the 11.0 GA release. New Features. General CUDA. CUDA 11.0 Update 1 is a minor update that is binary compatible with CUDA 11.0. docs.nvidia.com

---

From: 'max...@protonmail.com' via mcx-users <mcx-...@googlegroups.com>
Sent: Thursday, May 20, 2021 9:27 AM
To: mcx-users <mcx-...@googlegroups.com>
Subject: [mcx-users] Makefile and -arch=sm_30

 

--
You received this message because you are subscribed to the Google Groups "mcx-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mcx-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mcx-users/95c876f2-dbea-4ff5-9482-361d0116bfa4n%40googlegroups.com.

[Qianqian Fang]

if you don't want to manually edit Makefile, you can run the below command

make CUGENCODE=-arch=sm_35

if your CUDA toolkit requires a higher architecture version.

To view this discussion on the web visit https://groups.google.com/d/msgid/mcx-users/PH0PR06MB71414F31DEF25062F6B6B3A49C2A9%40PH0PR06MB7141.namprd06.prod.outlook.com.

[Maxime Baillot]

I did what Shijie Yan say and it compile now

Also thank for the make option that is what I was looking for at first

Thanks to both of you !