MIC Problems

Front page Forums Installation MIC Problems

This topic contains 5 replies, has 2 voices, and was last updated by  ZeroCool 3 years, 8 months ago.

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
    Posts
  • #734

    ZeroCool
    Participant

    I am gettings this errors and i don’t know what to do.
    OPENCL ERROR: CL_OUT_OF_RESOURCES
    I was already able to run paralution 0.6.1 on Host Processors with but i was not able to run it on the MIC.

    best retards

    OpenCL Platform: Intel(R) OpenCL
    CL_DEVICE_NAME: Intel(R) Many Integrated Core Acceleration Card
    CL_DEVICE_VENDOR: Intel(R) Corporation
    CL_DEVICE_TYPE: ACCELERATOR
    CL_DEVICE_MAX_COMPUTE_UNITS: 240
    CL_DEVICE_MAX_WORK_GROUP_SIZE: 8192
    CL_DEVICE_GLOBAL_MEM_SIZE: 11634 MByte
    CL_DEVICE_LOCAL_MEM_SIZE: 32 KByte
    —————————————————-
    ReadFileMTX: filename=fidap031.mtx; reading…
    ReadFileMTX: filename=fidap031.mtx; done
    *** warning: LocalMatrix::Permute() is performed on the host
    *** warning: LocalMatrix::ILU0Factorize() is performed on the host
    LocalMatrix name=fidap031.mtx; rows=3909; cols=3909; nnz=115299; prec=64bit; format=CSR; host backend={CPU(OpenMP)}; accelerator backend={OpenCL}; current=OpenCL
    PCG solver starts, with preconditioner:
    Multicolored ILU preconditioner (power(q)-pattern method), ILU(0,1)
    number of colors = 21; ILU nnz = 115299
    IterationControl criteria: abs tol=1e-15; rel tol=1e-06; div tol=1e+08; max iter=1000000
    OPENCL ERROR: CL_OUT_OF_RESOURCES

    #735

    Dimitar
    Member

    Hi,

    The OpenCL is kind of bad for the MIC/Xeon Phi (very bad performance). It is not recommended to be used. Do you get the same error with other matrices (like http://math.nist.gov/MatrixMarket/data/Harwell-Boeing/laplace/gr_30_30.html)?

    Please, switch to the OpenMP/MIC backend for the MIC/Xeon Phi- check the user manual for details on the compilation.

    Best,
    Dimitar

    #736

    ZeroCool
    Participant

    Ok thx i will try it,the aim is just to compare it on the base on OpenCl , but nevertheless i will take what i get at the moment.

    best regards

    #737

    ZeroCool
    Participant

    Yeah well really nice framework,but i am having some questions i am searching for applications where Xeon Phi is just superior against nvidia cards, so i just found some documents where the xeon phi is performing really well against nvidia cards because of using sparse matrices. Maybe any hints where Xeon Phi is just faster than the other cards. The problem is still the high latency of the card and not the hardware itself.

    How can i switch the acceleretors in the programming code , from opencl to to openmp and vise versa, on the cpu and gpu ?

    best regards

    #738

    Dimitar
    Member

    So far our MIC/Xeon Phi backend is slower compared to the GPUS – you can check our benchmarks in the user manual and/or our video at the SC13. But we are working on improving it…(do not know when this will happen).

    You need to compile the library for each backend (accelerator) and then you can switch between the CPU and the accelerator (GPU, MIC, etc). Of course, you can compile them all and since this is a dynamic library load it at run time.

    If you want to find an application where the Xeon Phi is faster you might want to look into some programs with a lot of branching and cache utilization.

    Best,
    Dimitar

    #740

    ZeroCool
    Participant

    Because i just found this paper and i am trying also to compare it with Xeon Phi,Xeon and Nvidia Cards

    http://arxiv.org/pdf/1302.1078v1.pdf

    Any examples ?

    best regards

Viewing 6 posts - 1 through 6 (of 6 total)

You must be logged in to reply to this topic.