0
  • Windows 7 x64
  • Python 3.5.2
  • CUDA Toolkit 8.0.61
  • Tensorflow package: tensorflow-gpu-1.2.0rc0
  • cudnn 8.0 (for CUDA 8.0 toolkit)

Test:

# Creates a graph.
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)
# Creates a session with log_device_placement set to True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
# Runs the op.
print(sess.run(c))

Results:

2017-05-30 13:50:33.021124: I C:\...\gpu_device.cc:906] Found device 0 with properties:
name: NVS 5200M
major: 2 minor: 1 memoryClockRate (GHz) 1.344
pciBusID 0000:01:00.0
Total memory: 1.00GiB
Free memory: 886.41MiB
2017-05-30 13:50:33.022124: I C:\...\gpu_device.cc:927] DMA: 0
2017-05-30 13:50:33.022124: I C:\...\gpu_device.cc:937] 0:   Y
2017-05-30 13:50:33.022124: I C:\...\gpu_device.cc:969] Ignoring visible gpu device (device: 0, name: NVS 5200M, pci bus id: 0000:01:00.0) with Cuda compute capability 2.1. The minimum required Cuda capability is 3.0.
Device mapping: no known devices.
2017-05-30 13:50:33.024124: I C:\...\direct_session.cc:265] Device mappin
g:

MatMul: (MatMul): /job:localhost/replica:0/task:0/cpu:0
2017-05-30 13:50:33.026124: I C:\...\simple_placer.cc:847] MatMul: (MatMul)/job:localhost/replica:0/task:0/cpu:0
b: (Const): /job:localhost/replica:0/task:0/cpu:0
2017-05-30 13:50:33.027124: I C:\...\simple_placer.cc:847] b: (Const)/job:localhost/replica:0/task:0/cpu:0
a: (Const): /job:localhost/replica:0/task:0/cpu:0
2017-05-30 13:50:33.027124: I C:\...\simple_placer.cc:847] a: (Const)/job:localhost/replica:0/task:0/cpu:0
[[ 22.  28.]
 [ 49.  64.]]

I assume my problem is "Ignoring visible gpu device with CUDA compute capability 2.1. The minimum required Cuda capability is 3.0." So it seems I'm limited to CUDA 2.1 by my hardware, but it's not clear were the 3.0 requirement comes from. Is it the CUDA toolkit or the tensorflow libraries?

talonmies
  • 70,661
  • 34
  • 192
  • 269
DrTarr
  • 922
  • 2
  • 15
  • 34
  • 1
    TF was originally published with a requirement for compute capability 3.0 (for GPU acceleration). I'm not able to give you an exact link to this for TF docs. Furthermore, much of TF DNN acceleration on the CUDA GPU comes about through the cudnn library, which [does specifically call out the required GPUs as being Kepler or newer](https://developer.nvidia.com/cudnn), and your cc2.x device is a Fermi device. Fermi GPUs are specifically not supported by cudnn. – Robert Crovella May 30 '17 at 18:13

1 Answers1

0

You can find instructions about GPU support on installation page.

GPU card with CUDA Compute Capability 3.0 or higher. See NVIDIA documentation for a list of supported GPU cards.

Still, there are ways to use GPU with lower Compute Capability. Refer to this.

YLJ
  • 2,940
  • 2
  • 18
  • 29