Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
Beginner
19 Views

Is the Docker image usable at all? Or at least on Intel GPUs?

Hi,

I seem not to own any Intel CPU with a graphics unit (an NVidia instead), and a Mac with an Intel graphics unit but currently unsupported by OneAPI. So I got the docker DL image on the Mac, tried the CCL Getting Started, and I am getting

 

 mpirun -n 2 ./out/sycl_allreduce_cpp_test cpu

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 0 PID 464 RUNNING AT f39a6870c956
=   KILLED BY SIGNAL: 7 (Bus error)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 1 PID 465 RUNNING AT f39a6870c956
=   KILLED BY SIGNAL: 7 (Bus error)
===================================================================================

 

 

Same for the gpu target.

On the surface this could be anything, but I thought it would be the absence of OpenCL in the container. I installed clinfo and discovered the container shows at least two OpenCL providers, including the IntelFPGA emulator, if this is the correct terminology. I am very new to this toolkit so not really sure how this whole thing is supposed to work if the Getting Started fails, but I thought I should add here a couple of FAQs

1) is cuda as a target coming, or openCL perhaps on NVidia, or even is it already there? A bit surprised the READM only mentions CPU and GPU

2) A couple of years ago I got the impression your "emulated" OpenCL support for CPUs without a graphics unit was discontinued, am I right? Has it made a comeback in the FPGA SDK? Or am I even completely wrong about this whole OpenCL thing and "running on CPU" is supposed to optimize DL by using Intel CPU features and bypassing OpenCL altogether?

 

In short it would help if someone looked into running the getting starteds inside Docker.

 

Thanks!

Tags (1)
0 Kudos
3 Replies
Highlighted
Beginner
19 Views

I am also trying https:/

I am also trying https://github.com/intel/DLFDKit-code-samples , but the FP32 sample with which I started seems to have a dependency on the Intel compiler, and this returns nothing:

find / -name icpc

 

Is this an accident or by design?

Thanks!

0 Kudos
Highlighted
Employee
19 Views

Hi Taha,  I try to answer


  I try to answer your questions one by one. If I missed, please let me know.
 
  1) Error to run: mpirun -n 2 ./out/sycl_allreduce_cpp_test cpu

  Answer:
    oneAPI only support linux and windows by now.
    In Mac, the docker of oneAPI could be installed. But oneAPI in docker on Mac still is not supprted.
    Because the OS kernel of docker is same as Host. It's still Mac, instead of linux or windows.
   
  2) is cuda as a target coming, or openCL perhaps on NVidia, or even is it already there? A bit surprised the READM only mentions CPU and GPU
 
  Answer:
    oneAPI only support Intel CPU and GPU. It's unknown the result of other GPU.
    It could be extended to support more CPU & GPU as an open standard in the future.
    A company Codeplay has allowed oneAPI and SYCL / Data Parallel C++ to run atop NVIDIA GPUs via CUDA. (Nov. 2019)
   
   
  3) A couple of years ago I got the impression your "emulated" OpenCL support for CPUs without a graphics unit was discontinued, am I right? Has it made a comeback in the FPGA SDK? Or am I even completely wrong about this whole OpenCL thing and "running on CPU" is supposed to optimize DL by using Intel CPU features and bypassing OpenCL altogether?
 
  Answer: Yes, oneAPI DL tookits optimize the DL by using Intel CPU features and cooperate with GPU. The toolkits is developed by DPC++ which is based on SYCL & OpenCL, running on GPU and CPU.
 
  4)Issue of 'find / -name icpc'
  Answer: It's a bug in docker image: intel/oneapi-dlfdkit. We will fix it as soon.
  Before it's fixed, please download and install oneAPI DL Framework Developer Toolkit directly.
  Please refer to: https://software.intel.com/en-us/articles/installation-guide-for-intel-oneapi-toolkits
 

0 Kudos
Highlighted
Employee
19 Views

Hi Taha,  For the icpc issue,

Hi Taha,
  For the icpc issue, after check, I think it's not a bug:
  The DLFTKit doesn't include the icpc according to the design.
 
  In the guide of https://github.com/intel/DLFDKit-code-samples/tree/master/oneDNN_CNN_INFERENCE_FP32, it mentions to use Intel C++ Compiler, it is required to install Intel oneAPI HPC Toolkit which provide the icpc.
 
  Please install the Intel oneAPI HPC Toolkit in the environment(docker container) before use icpc.
 
  Thank you!

0 Kudos