GTX 780M with 340 driver and CUDA8 ( ? CUDNN ) - installation confusion
Hi, I'm running Ubuntu Zesty 17.04 on an Imac. With GPU GTX 780M I would like to run multiple ML projects and I need a proper installation of CUDA for theano,caffe,tensorflow,anacanda, etc... I've been painstakingly at it for week to get a working setup for my Graphics card. Which Cuda version and Cuddn version als g++ and gcc should I install to make this work. nvidia-smi gives me this with !! 384.90 !! driver, only option other I have is the proprietary 340.** version. I also have cuda 8.0 installed and it's in my .bashrc file Tue Nov 14 17:04:32 2017 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 384.90 Driver Version: 384.90 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 780M Off | 00000000:01:00.0 N/A | N/A | | N/A 50C P0 N/A / N/A | 279MiB / 4035MiB | N/A Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | 0 Not Supported | +-----------------------------------------------------------------------------+ With the 340.102 +------------------------------------------------------+ | NVIDIA-SMI 340.102 Driver Version: 340.102 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 780M Off | 0000:01:00.0 N/A | N/A | | N/A 49C P0 N/A / N/A | 350MiB / 4095MiB | N/A Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Compute processes: GPU Memory | | GPU PID Process name Usage | |=============================================================================| | 0 Not Supported | +-----------------------------------------------------------------------------+
Hi,

I'm running Ubuntu Zesty 17.04 on an Imac.

With GPU GTX 780M

I would like to run multiple ML projects and I need a proper installation of CUDA for theano,caffe,tensorflow,anacanda, etc...

I've been painstakingly at it for week to get a working setup for my Graphics card.

Which Cuda version and Cuddn version als g++ and gcc should I install to make this work.

nvidia-smi gives me this with !! 384.90 !! driver, only option other I have is the proprietary 340.** version.
I also have cuda 8.0 installed and it's in my .bashrc file

Tue Nov 14 17:04:32 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 384.90 Driver Version: 384.90 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 780M Off | 00000000:01:00.0 N/A | N/A |
| N/A 50C P0 N/A / N/A | 279MiB / 4035MiB | N/A Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 Not Supported |
+-----------------------------------------------------------------------------+

With the 340.102


+------------------------------------------------------+
| NVIDIA-SMI 340.102 Driver Version: 340.102 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 780M Off | 0000:01:00.0 N/A | N/A |
| N/A 49C P0 N/A / N/A | 350MiB / 4095MiB | N/A Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Compute processes: GPU Memory |
| GPU PID Process name Usage |
|=============================================================================|
| 0 Not Supported |
+-----------------------------------------------------------------------------+

#1
Posted 11/14/2017 04:18 PM   
384.90 driver will work fine with any CUDA version, including 9 and 8. That is a good choice and you should not need to change it. The specific CUDA version will depend on what you want to do. If you want to run a specific framework (e.g. TF) by downloading compiled binaries for it, then you would need the specific CUDNN and CUDA version that the binaries were compiled against. Right now CUDA 8/CUDNN 6 may be good choices, but over time this will change. If you want to build a framework (e.g. TF) from sources, perhaps so as to take advantage of most recent features or most recent CUDNN version, then you can use the latest. Currently CUDA 9/CUDNN 7 may be good choices, and should work on your hardware. 384.90 can work with any of the above. On linux, it is not difficult to have multiple CUDA and CUDNN versions installed, and switch between them via appropriate settings in PATH and LD_LIBRARY_PATH variables.
384.90 driver will work fine with any CUDA version, including 9 and 8. That is a good choice and you should not need to change it.

The specific CUDA version will depend on what you want to do. If you want to run a specific framework (e.g. TF) by downloading compiled binaries for it, then you would need the specific CUDNN and CUDA version that the binaries were compiled against. Right now CUDA 8/CUDNN 6 may be good choices, but over time this will change.

If you want to build a framework (e.g. TF) from sources, perhaps so as to take advantage of most recent features or most recent CUDNN version, then you can use the latest. Currently CUDA 9/CUDNN 7 may be good choices, and should work on your hardware.

384.90 can work with any of the above. On linux, it is not difficult to have multiple CUDA and CUDNN versions installed, and switch between them via appropriate settings in PATH and LD_LIBRARY_PATH variables.

#2
Posted 11/14/2017 05:55 PM   
Scroll To Top

Add Reply