Environment Variables

MXNet 有几个设置可以通过环境变量来修改.
一般情况下, 用户是不需要修改这些设置的, 我们列出来作为参考, 以备不时之需.

• MXNET_GPU_WORKER_NTHREADS (default=2)
  • 每个 GPU 上用来做计算工作的线程的最大数目.
• MXNET_GPU_COPY_NTHREADS (default=1)
  • 每个 GPU 上用来做内存拷贝工作的线程的最大数目.
• MXNET_CPU_WORKER_NTHREADS (default=1)
  • 做 CPU 计算工作的线程的最大数目.
• MXNET_CPU_PRIORITY_NTHREADS (default=4)
  • 用来做优先的 CPU 工作的线程的数目.
• MXNET_EXEC_ENABLE_INPLACE (default=true)
  • 在符号计算中是否能做原地优化.
• MXNET_EXEC_MATCH_RANGE (default=10)
  • The rough matching scale in symbolic execution memory allocator.
  • Set this to 0 if we do not want to enable memory sharing between graph nodes(for debug purpose).
• MXNET_EXEC_NUM_TEMP (default=1)
  • Maximum number of temp workspace we can allocate to each device.
  • Set this to small number can save GPU memory.
  • It will also likely to decrease level of parallelism, which is usually OK.
• MXNET_ENGINE_TYPE (default=ThreadedEnginePerDevice)
  • The type of underlying execution engine of MXNet.
  • List of choices
    • NaiveEngine: very simple engine that use master thread to do computation.
    • ThreadedEngine: a threaded engine that uses global thread pool to schedule jobs.
    • ThreadedEnginePerDevice: a threaded engine that allocates thread per GPU.
• MXNET_KVSTORE_REDUCTION_NTHREADS (default=4)
  • Number of threads used for summing of big arrays.
• MXNET_KVSTORE_BIGARRAY_BOUND (default=1e6)
  • The minimum size of "big array".
  • When the array size is bigger than this threshold, MXNET_KVSTORE_REDUCTION_NTHREADS threads will be used for reduction.

Settings for Minimum Memory Usage

• Make sure min(MXNET_EXEC_NUM_TEMP, MXNET_GPU_WORKER_NTHREADS) = 1
  • The default setting satisfies this.

Settings for More GPU Parallelism

• Set MXNET_GPU_WORKER_NTHREADS to larger number (e.g. 2)
  • You may want to set MXNET_EXEC_NUM_TEMP to reduce memory usage.
• This may not speed things up, especially for image applications, because GPU is usually fully utilized even with serialized jobs.