Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Charles Determan
    @cdeterman
    I have tested a GeForce GTX 970 GPU using CUDA 7.5
    Karl Rupp
    @karlrupp
    hi
    Charles Determan
    @cdeterman
    Hey @karlrupp I'm hoping to use this to consolidate those who use the package successfully. Be good to begin making a list of GPUs that are working successfully with the package.
    Karl Rupp
    @karlrupp
    ah, I see. I thought you wanted to chat about something specific
    Sorry for the noise then
    Charles Determan
    @cdeterman
    nah, was just inviting you to this chat room in case you were curious
    Colin Rundel
    @rundel
    I was able to get things running on a nvidia K20X system running Centos 7 + CUDA 7.5. Only issue is libOpenCL is in /usr/lib64/nvidia on Centos which requires adding an additional -L flag to configure.ac
    Charles Determan
    @cdeterman
    Thanks @rundel I will probably make a wiki page for CentOS systems (and subsequently different Linux flavors as needed).
    vsmaier
    @vsmaier
    Currently testing with (two) AMD FirePro D300 2048 MB on a Mac Pro. So far smooth sailing.
    Charles Determan
    @cdeterman
    User confirmed GeForce GTX Titan X on Windows
    Oshane Thomas
    @oothomas
    Dr. Determan is likely referring to my gpuR installation on a GeForce GTX Titan X, the package continues to work well.
    Navarro-msr
    @Navarro-msr
    I have tested a R9 390 GPU in a Windows 10, worked. In ubuntu 16 I can't install, several warnings of "deprecated declarations", in Windows this happened too but it was surpassed.
    Charles Determan
    @cdeterman
    @Navarro-msr thanks for the report, the deprecated declarations are likely because the OpenCL headers are for version >= 2.0. I actually just recently started messing with Ubuntu 16 and it has discontinued support for the fglrx driver for AMD cards. I am currently looking into how to get it to work there.
    Bruno Hanzen
    @Bruno-Hanzen
    Works well on an AMD A10
    Floris Padt
    @Fpadt
    I have 2 NVidia cards to test (Geforce 940MX and GeForce 660 Ti) dual boot windows/Ubuntu. How do I test on windows? can this be done via Rstudio, I have devtools and Rtools 3.3, all installed well and examples did run
    Charles Determan
    @cdeterman
    @Fpadt You should be able to test on windows the same as any other OS. Is the install failing for you in some way?
    Floris Padt
    @Fpadt
    @Charles: everything is installed and I can run the short script from the pdf. However can't get the R CMD check gpu_version running. I executed the check in Rstudio in the build tab and this end with an error I must be doing something wrong.
    • preparing 'gpuR':
    • checking DESCRIPTION meta-information ... OK
      ERROR
    Charles Determan
    @cdeterman
    @Fpadt Strange, I would export more information than that. Which branch of code are you using? master or develop?
    Floris Padt
    @Fpadt

    installed via cran so that must e master then?
    ==> Rcpp::compileAttributes()

    • Updated

    ==> devtools::check(document = FALSE)

    Setting env vars --------------------------------------------------------------
    CFLAGS : -Wall -pedantic
    CXXFLAGS: -Wall -pedantic
    Building gpuR -----------------------------------------------------------------
    "C:/PROGRA~1/R/R-34~1.1/bin/x64/R" --no-site-file --no-environ --no-save \
    --no-restore --quiet CMD build \
    "C:\Users\Floris\Documents\R\win-library\3.4\gpuR" --no-resave-data \
    --no-manual

    • checking for file 'C:\Users\Floris\Documents\R\win-library\3.4\gpuR/DESCRIPTION' ... OK
    • preparing 'gpuR':
    • checking DESCRIPTION meta-information ... OK
      ERROR
      computing Rd index failed:cannot open the connection
      Warning in file(con, "r") :
      cannot open file 'man': No such file or directory
      Error: Command failed (1)
      Execution halted

    Exited with status 1.

    Charles Determan
    @cdeterman
    @Fpadt Are you able to load the package manually and use any of the functions directly?
    Floris Padt
    @Fpadt
    yes, and the two pieces of code from your vignette (pag 3 & 4)seem to run fine. It is the R CMD check gpuR_version which not giving the expected results (so probably just ignore)
    Daniel
    @danielw2904
    Works on Ubuntu 17.10:
    library(gpuR)
    Number of platforms: 1
    • platform: NVIDIA Corporation: OpenCL 1.2 CUDA 9.0.282
      • context device index: 0
        • Quadro K620M
          checked all devices
          completed initialization
          gpuR 2.0.0
    Charles Determan
    @cdeterman
    @danielw2904 thanks for the report. Glad it is working on 17.10. That is the first I've heard of anyone trying that.
    Daniel
    @danielw2904
    No problem. Also since that has been a struggle for me for years: my laptop has optimus and this was the first time Nvidia drivers and cuda worked with decent performance just by installing from the non-free repos. Tested before on fedora 27 and had terrible performance and Debian 9 where I could only get Nvidia to work with bumblebee.
    Daniel
    @danielw2904
    No problem. Also since that has been a struggle for me for years: my laptop has optimus and this was the first time Nvidia drivers and cuda worked with decent performance just by installing from the non-free repos. Tested before on fedora 27 and had terrible performance and Debian 9 where I could only get Nvidia to work with bumblebee.
    Sicheng Hao
    @SichengH
    Hello,
    I tested some matrix operations on AMD Radeon Pro 450(OpenCL 1.2), but the speed seems favor the regular matrix. My guess is my GPU was not being used. Did anyone have the same issue?
    Charles Determan
    @cdeterman
    @sichenghao1992 hello, what version of gpuR are you using? Also, what operations are you testing and what are the matrix dimensions? You may wish to open a github issue to troubleshoot this.
    olepett
    @olepett

    Tested: gpuR v 2.0.3

    OS: Windows 10 Home
    platformName NVIDIA CUDA
    platformVendor NVIDIA Corporation
    platformVersion OpenCL 1.2 CUDA 9.2.147
    GPU Geforce 1080 Ti

    Works like a charm! Note this is a single precision GPU.

    Thank you for a great package.

    nadal8701
    @nadal8701_twitter
    Hi, I have 4 rtx 2080 ti gpu cards installed on a computer with 48 cpu cores and 128GB ram, pls contact me if you need.