Performance evaluation

[AnonymousYWL]

Hello.
Are there any benchmark results evaluating the performance of AMX-enabled processors for DNN operators, such as convolution and matrix multiplication?

[asirvaiy]

Hi,
You can use Benchdnn to check the performance of each operator. https://github.com/oneapi-src/oneDNN/tree/main/tests/benchdnn
To enable AMX, you need to use data type int8 or bf16.

[asirvaiy]

Hi, You can use Benchdnn to check the performance of each operator. https://github.com/oneapi-src/oneDNN/tree/main/tests/benchdnn To enable AMX, you need to use data type int8 or bf16.

Hope you were able to benchmark. Closing the issue now.