{"id":817,"date":"2019-06-01T15:43:50","date_gmt":"2019-06-01T06:43:50","guid":{"rendered":"https:\/\/itips.krsw.biz\/?p=817"},"modified":"2019-06-01T15:43:50","modified_gmt":"2019-06-01T06:43:50","slug":"tensorflow-blas-gemm-launch-failed-error","status":"publish","type":"post","link":"https:\/\/itips.krsw.biz\/en\/tensorflow-blas-gemm-launch-failed-error\/","title":{"rendered":"How to solve \"Blas GEMM launch failed\" in TensorFlow"},"content":{"rendered":"

\"How<\/p>\n

Sometimes you see Blas GEMM launch failed<\/code> error in TensorFlow.
\nIt is a little difficult to understand.
\nSo this introduce how to solve \"Blas GEMM launch failed\" in TensorFlow.<\/p>\n


\n<\/br><\/p>\n

What is TensorFlow ?<\/h2>\n

TensorFlow is very famous machine learning library.
\nIn machine learning field, everyone know it.
\nIt is useful for neural network processing.
\nIn TensorFlow you has detailed settings like hidden layer or activation function.<\/p>\n

\n TensorFlow is a free and open-source software library for dataflow and differentiable programming across a range of tasks. It is a symbolic math library, and is also used for machine learning applications such as neural networks.
\n TensorFlow - Wikipedia<\/a>\n<\/p><\/blockquote>\n

<\/br><\/p>\n

Cause of \"Blas GEMM launch failed\"<\/h2>\n

When you execute TensorFlow process in Python, you may see this error.<\/p>\n

\nInternalError: Blas GEMM launch failed\n<\/pre>\n
A reason of this error is memory overflow.
\nIf you try to use too much memory, the error happens.<\/p>\n
In order to check memory,  it is easy to check GPU memory usage<\/strong>.
\nIn case of Windows10, you can run task manager with using taskmgr<\/code> command.
\nThe you can see GPU  memory usage of GPU0<\/code> or GPU1<\/code> .
\nIf GPU memory usage goes to 100% when you run TensorFlow prrogram, it means memory overflow.<\/p>\n
\n  Check GPU memory usage. In Windows10 you can check it on task manager.
\n  [Tensorflow]InternalError: Blas SGEMM launch failed | CodeLab<\/a>\n<\/p><\/blockquote>\n
<\/br><\/p>\n
Is shared GPU memory available ?<\/h2>\n
On task manager, you can see private GPU memory<\/strong> and shared GPU memory<\/strong>.
\nSo you may estimate that shared GPU memory is available.
\nBut it is not possible to use shared GPU memory for TensorFlow.<\/p>\n
Shared GPU memory is not on GPU. It is PC memory that is shared for GPU.
\nCUDA can use only GPU memory. So it can't use shared GPU memory<\/strong>.<\/p>\n
\n  Shared memory is reserved memory of main RAM.
\n  It is not on your NVIDIA GPU. CUDA can't use it.
\n   Is shared GPU memory available in TensorFlow ?<\/a>\n<\/p><\/blockquote>\n
<\/br><\/p>\n
How to avoid \"Blas GEMM launch failed\"<\/h2>\n
There are some solutions.<\/p>\n