Device detected by clinfo but not by tensorflow

[InonS]

I'm not sure if this is a "supported architectures" issue, or if there are more details I should give. What do you think?

$ ./run_tests.sh
+ cd examples
+ pushd 2_BasicModels
/.../TensorFlow-Examples/examples/2_BasicModels /.../TensorFlow-Examples/examples
+ python linear_regression.py
gpu_manager->VisibleDeviceCount is 0
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/client/session.py", line 972, in _do_call
    return fn(*args)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/client/session.py", line 950, in _run_fn
    self._extend_graph()
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/client/session.py", line 999, in _extend_graph
    self._session, graph_def.SerializeToString(), status)
  File "/usr/lib/python3.5/contextlib.py", line 66, in __exit__
    next(self.gen)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/framework/errors.py", line 463, in raise_exception_on_not_ok_status
    pywrap_tensorflow.TF_GetCode(status))
tensorflow.python.framework.errors.InvalidArgumentError: Cannot assign a device to node 'GradientDescent/learning_rate': Could not satisfy explicit device specification '/device:GPU:0' because no devices matching that specification are registered in this process; available devices: /job:localhost/replica:0/task:0/cpu:0
         [[Node: GradientDescent/learning_rate = Const[dtype=DT_FLOAT, value=Tensor<type: float shape: [] values: 0.01>, _device="/device:GPU:0"]()]]

$ clinfo
Number of platforms                               1
  Platform Name                                   Intel(R) OpenCL
  Platform Vendor                                 Intel(R) Corporation
  Platform Version                                OpenCL 1.2 LINUX
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_depth_images cl_khr_3d_image_writes cl_intel_exec_by_local_thread cl_khr_spir cl_khr_fp64
  Platform Extensions function suffix             INTEL

  Platform Name                                   Intel(R) OpenCL
Number of devices                                 1
  Device Name                                     Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 1.2 (Build 25)
  Driver Version                                  1.2.0.25
  Device OpenCL C Version                         OpenCL C 1.2
  Device Type                                     CPU
  Device Profile                                  FULL_PROFILE
  Max compute units                               2
  Max clock frequency                             2500MHz
  Device Partition                                (core)
    Max number of sub-devices                     2
    Supported partition types                     by counts, equally, by names (Intel)
  Max work item dimensions                        3
  Max work item sizes                             8192x8192x8192
  Max work group size                             8192
  Preferred work group size multiple              128
  Preferred / native vector sizes
    char                                                 1 / 16
    short                                                1 / 8
    int                                                  1 / 4
    long                                                 1 / 2
    half                                                 0 / 0        (n/a)
    float                                                1 / 8
    double                                               1 / 4        (cl_khr_fp64)
  Half-precision Floating-point support           (n/a)
  Single-precision Floating-point support         (core)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Address bits                                    64, Little-Endian
  Global memory size                              4142542848 (3.858GiB)
  Error Correction support                        No
  Max memory allocation                           1035635712 (987.7MiB)
  Unified memory for Host and Device              Yes
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        262144
  Global Memory cache line                        64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             480
    Max size for 1D images from buffer            64727232 pixels
    Max 1D or 2D image array size                 2048 images
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             2048x2048x2048 pixels
    Max number of read image args                 480
    Max number of write image args                480
  Local memory type                               Global
  Local memory size                               32768 (32KiB)
  Max constant buffer size                        131072 (128KiB)
  Max number of constant args                     480
  Max size of kernel argument                     3840 (3.75KiB)
  Queue properties
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Local thread execution (Intel)                Yes
  Prefer user sync for interop                    No
  Profiling timer resolution                      1ns
  Execution capabilities
    Run OpenCL kernels                            Yes
    Run native kernels                            Yes
    SPIR versions                                 1.2
  printf() buffer size                            1048576 (1024KiB)
  Built-in kernels
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Device Extensions                               cl_khr_icd cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_depth_images cl_khr_3d_image_writes cl_intel_exec_by_local_thread cl_khr_spir cl_khr_fp64

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  No platform
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   No platform
  clCreateContext(NULL, ...) [default]            No platform
  clCreateContext(NULL, ...) [other]              Success [INTEL]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  No platform

[hughperkins]

You need a device with Device Type set to GPU. Yours says CPU?

[InonS]

It says: CL_DEVICE_TYPE_CPU ...

[hughperkins]

Oh. Youre the guy that posted a LinkedIn message that was "like"d by thousands :)

Welcome :)

Ok, so, modern Intel CPUs often contain GPUs. Which is not the 'CPU' bit of the CPU itself, but an extra component, inside the CPU. It's a bit confusing :-P

So, it turns out that your cpu does in fact have a GPU inside it, it's an HD4600: https://ark.intel.com/products/78930/Intel-Core-i7-4710HQ-Processor-6M-Cache-up-to-3_50-GHz

Then the next question is: does your HD4600 GPU, inside your 4710HQ CPU, support OpenCL? The page above doesnt say. But it sounds modernish: I used to have an HD4000, and that supported OpenCL, so let's see...

Googling for 'wikipedia hd', we get https://en.wikipedia.org/wiki/Intel_HD_and_Iris_Graphics#Capabilities . This shows that Haswell cpus have OpenCL 1.2 GPUs:

At this point, I conclude: you're missing the driver :) . You probably need to install a driver from the Intel website, eg something like ...hmmm... I can only find for windows https://downloadcenter.intel.com/product/97501/Graphics-for-5th-Generation-Intel-Processors . oh right: you need Beignet :)

• https://www.freedesktop.org/wiki/Software/Beignet/
• https://github.com/intel/beignet

[hughperkins]

(Note that you're very late to the party though; I'm working on other things at the moment; meanwhile tf-coriander is missing a cudnn replacement. I started working on one at https://github.com/hughperkins/coriander-dnn , but never quite got round to plugging it in to tf-coriander. I dont think it's tons of work. If I did have a moment, it'd probably take me ~40-80 hours. But of course if someone else does it, they have to learn everything from scratch, so could easily be 4-8 times that.

If you can find someone who could be interested in helping with that, I'm happy to assist them with knowledge acquisition, meet them in Hangouts etc.

The impact of not having cudnn currently is that convolutions run on CPU (not the GPU part of the CPU, the CPU bit). )

[InonS]

Wow, thanks Hugh!

Yeah, the response to that post surprised me too. I was so upset that TensorFlow doesn't support an open standard for GPGPU!

As for my particular case, I do have an AMD GPU in addition to the Intel one which is built into my motherboard. It's a Radeon R9 M265X. Pretty low-scale by today's standards, but It supports OpenCL, and it's frustrating that I can't use it with TensorFlow out of the box (as opposed to an nVidia card, I expect). I'm not even sure how much help it would be, considering I went and paid for a quad-core CPU.

I guess I need to re-check my driver installation. I tried using Ubuntu-based dockers, which is where I got the stacktrace and clinfo output in my original post. Maybe I should address AMD support?

[hughperkins]

Yeah, the response to that post surprised me too. I was so upset that TensorFlow doesn't support an open standard for GPGPU!

:)

Yes, impressive to write something that went so viral :)

I guess I need to re-check my driver installation. I tried using Ubuntu-based dockers, which is where I got the stacktrace and clinfo output in my original post.

Docker is not very OpenCL/GPU friendly. Docker does work ... with NVIDIA GPUs :-P . I'm not saying Docker cant be tweaked to work with AMD GPUs, but I've never heard of that being possible. For NVIDIA GPUs, you need some special additional drivers, eg https://github.com/NVIDIA/nvidia-docker , or at least pass the drivers through, using --device option to Docker, like https://hub.docker.com/r/hughperkins/cltorch-nvidia/ But I've never heard of this being possible for AMD GPUs.

Your easiest options for AMD GPUs will probably be to use the AMD GPU directly from your OS, so one of:

• install an OS that supports the AMD GPU drivers, and /or
• find AMD GPU drivers for your current main OS

If it was me, well .... so... I got into OpenCL, since I had a laptop with an Intel CPU, with an HD4000 inside, and I thought it was so cool that the CPU had a GPU inside, and wanted to play, and of course it wont work with CUDA, so I wrote https://github.com/hughperkins/DeepCL from scratch, incrementally, over ~6 months, so that I could play with using the HD4000 GPU :)

Later on though, I found that the Intel GPU, whilst fun, is not something I'd ever train an ml model on: aws works well for that, or at least, an NVIDIA GPU. There are no AMD cloud-enabled GPUs around that I can find.

Currently, I think that whilst it'd be good to have competition for NVDIA GPUs, to keep them on their toes, I'm not sure that AMD will be that competition, at least, not in a big way. I think that something like the Nervana TPUs might be a more realistic competition possibly? https://cloud.google.com/blog/big-data/2017/05/an-in-depth-look-at-googles-first-tensor-processing-unit-tpu

[InonS]

Agreed. On all points ;-)