Problems with MCX/CUDA

103 views
Skip to first unread message

Demian Augusto Vera

unread,
Sep 8, 2021, 11:18:50 AM9/8/21
to mcx-users
HI! 
I'm having some troubles running MCXLab (in Matlab 2017a, Windows 10, 64 bits, NVidia GeForce GTX 480). When I run any of the examples, this occurs:

- variant name: [Fermi] compiled by nvcc [7.5] with CUDA [7050]
- compiled with: RNG [xorshift128+] with Seed Length [4]

GPU=1 (GeForce GTX 480) threadph=4340 extra=6400 np=100000000 nthread=23040 maxgate=1 repetition=1
initializing streams ... MCXLAB ERROR -13 in unit mcx_core.cu:2079: invalid device symbol
Error from thread (0): invalid device symbol
C++ Error: MCXLAB Terminated due to an exception!


If I run  gpuinfo=mcxlab('gpuinfo'):

=============================   GPU Infomation  ================================
Device 1 of 1: GeForce GTX 480
Compute Capability: 2.0
Global Memory: 1610612736 B
Constant Memory: 65536 B
Shared Memory: 49152 B
Registers: 32768
Clock Speed: 1.40 GHz
Number of SMs: 15
Number of Cores: 480
Auto-thread: 23040
Auto-block: 192

I've never get this error in other PCs that I use. I also tried different binaries (several compilation, nightly build, etc). 
Any idea of what's happening?

Thanks!
Best,
Demián.

Fang, Qianqian

unread,
Sep 8, 2021, 11:29:10 AM9/8/21
to mcx-...@googlegroups.com, Demian Augusto Vera
On 9/8/21 11:18 AM, Demian Augusto Vera wrote:
HI! 
I'm having some troubles running MCXLab (in Matlab 2017a, Windows 10, 64 bits, NVidia GeForce GTX 480). When I run any of the examples, this occurs:


GTX 480 (fermi architecture) is too old for most of the newly released binaries. As a matter of fact, CUDA 9 dropped the support for fermi and the lowest architecture it supports is Kepler. We have switched to CUDA 9 over a year ago.


there are possibly two ways to get around.

first, try one of our older releases, say 2018 or earlier.

alternatively, download cuda toolkit 7.5 on your own machine and build mcx locally. The mcx's code should still compatible with fermi, but I can't be certain because we stopped testing on fermi several years ago.




- variant name: [Fermi] compiled by nvcc [7.5] with CUDA [7050]
- compiled with: RNG [xorshift128+] with Seed Length [4]

GPU=1 (GeForce GTX 480) threadph=4340 extra=6400 np=100000000 nthread=23040 maxgate=1 repetition=1
initializing streams ... MCXLAB ERROR -13 in unit mcx_core.cu:2079: invalid device symbol
Error from thread (0): invalid device symbol
C++ Error: MCXLAB Terminated due to an exception!


If I run  gpuinfo=mcxlab('gpuinfo'):

=============================   GPU Infomation  ================================
Device 1 of 1: GeForce GTX 480
Compute Capability: 2.0
Global Memory: 1610612736 B
Constant Memory: 65536 B
Shared Memory: 49152 B
Registers: 32768
Clock Speed: 1.40 GHz
Number of SMs: 15
Number of Cores: 480
Auto-thread: 23040
Auto-block: 192

I've never get this error in other PCs that I use. I also tried different binaries (several compilation, nightly build, etc). 
Any idea of what's happening?

Thanks!
Best,
Demián.
--
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/430e76dd-50a6-4750-8e5e-b36059dc6552n%40googlegroups.com.


Reply all
Reply to author
Forward
0 new messages