Huawei's Bolt 1.5 Adds AVX-VNNI, Intel Desktop GPU Support

Huawei's Bolt project is a deep learning library focused on high performance and heterogeneous flexibility and supporting a variety of neural networks. Bolt claims to outperform other deep learning acceleration libraries while supporting models from TensorFlow, ONNX, Caffe, and more.

With today's Bolt 1.5 release there is now support for AVX-VNNI instructions as well as formally adding support for the Armv9 architecture. Also on the new hardware support side, Intel desktop GPUs (Arc Graphics) are now supported by the Bolt library when using the Float16 or Float32 data types.

Bolt 1.5 also introduces a Python API, adds support for Windows on Arm builds, supports more neural networks and operators, model file compatibility enhancements, and improves the multi-threaded parallel inference performance on CPUs.

Downloads and more details on the Bolt 1.5 deep learning library via GitHub.