singa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Muhammad Hamdan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SINGA-308) CPU-GPU parallelism
Date Sat, 08 Apr 2017 18:37:42 GMT

    [ https://issues.apache.org/jira/browse/SINGA-308?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15961915#comment-15961915
] 

Muhammad Hamdan commented on SINGA-308:
---------------------------------------

I tried that, but got SegFault! 


     MemPoolConf mem_conf;
     mem_conf.add_device(0);
    // mem_conf.add_device(1);
   // std::shared_ptr<DeviceMemPool> mem_pool(new CnMemPool(mem_conf));
   // std::shared_ptr<CudaGPU> dev_1(new CudaGPU(0, mem_pool));
  //std::shared_ptr<CudaGPU> dev_2(new CudaGPU(1, mem_pool));

Does this code allocate memory between the two GPUs? 
I assume that the model configuration should not matter on which platform to be trained (Cudnn_Conv
or SingaConv) these are just identifiers right? 


> CPU-GPU parallelism 
> --------------------
>
>                 Key: SINGA-308
>                 URL: https://issues.apache.org/jira/browse/SINGA-308
>             Project: Singa
>          Issue Type: Test
>          Components: Core, PySINGA
>         Environment: Ubuntu 16.04
> CPU-GPU of the same machine
>            Reporter: Muhammad Hamdan
>              Labels: test
>
> Is it possible to parallelize the alexnet model for the cifar10 example on a CPU and
GPU instead of 2-GPUs ? Assuming asynchronous communication between the two components



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message