Many people see an error reading “Your CPU Supports Instructions That This Tensorflow Binary Was Not Compiled to Use: AVX2” while attempting to install or run TensorFlow. Here’s an actual post from the GitHub discussion board that illustrates the point:
AVX: What Does It Mean?
You need to understand what AVX is in order to fix the “Your CPU Supports Instructions That This Tensorflow Binary Was Not Compiled to Use: AVX2” problem.
AVX is an abbreviation for “Advanced Vector Extensions,” which are updates to the x86 instruction set architecture for modern microprocessors from Intel and AMD. It introduces a new set of capabilities, protocols, and encoding method.
With AVX2, also known as Haswell New Instructions, most integer operations can now use the full 256 bits of memory available to them. This feature was introduced with Intel’s Haswell processors. The FMA was introduced by AVX and uses the single-step floating-point multiply-add operation to speed up a variety of calculations.
Algebra computations are accelerated and simplified using AVX, too. It’s possible that AVX and FMA-enabled CPUs will outperform their non-supporting predecessors. Yet, this warning may appear if your CPU isn’t AVX-capable.
Explaination: Your CPU Supports Instructions That This Tensorflow Binary Was Not Compiled to Use: AVX2
If you see an error message saying “your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA,” it’s because your computer’s processor is capable of running programmes that require a different set of instructions than those included in the TensorFlow shared library. It’s because from TensorFlow 1.6, the binaries begin using AVX instructions, which can cause problems for older CPUs.
On newer CPUs, you need to construct TensorFlow from source to the CPU, but on older CPUs, the AVX instructions simply won’t run. To what end is the AVX not the de facto standard? This is because the problem is triggered since the default TensorFlow release is not created with the CPU extensions.
Fix: Your CPU Supports Instructions That This Tensorflow Binary Was Not Compiled to Use: AVX2
This TensorFlow binary was not compiled to use certain CPU instructions (AVX, AVX2); how can I solve this? The caution notice was actually delivered so you’d know you were using a source-built version of TensorFlow.
These AVX support warnings are unnecessary if you have a GPU installed in your PC and can bypass them. You can prevent this error message from appearing again by including the code below. Specifically, your main programme code needs to import the OS module and configure the mapping object.
os import
Setting os.environ[‘TF CPP MIN LOG LEVEL’] to 2
The following is an example of an export command that may be used in the bash shell on Unix systems.
It’s possible to minimise log levels in TF-CPP by exporting TF CPP MIN LOG LEVEL=2.
To get the most out of your CPU if you don’t have a GPU, you should compile TensorFlow from source with AVX, AVX2, and FMA support enabled.
Just what is the Root of this Caution?
Binaries built with TensorFlow 1.6 use AVX instructions, which may cause compatibility issues with older CPUs. Therefore, the AVX will not work on older CPUs, and the user of a newer CPU will need to compile tensorflow from source code for their CPU.
