search

KhronosGroup / Vulkan-Ecosystem

Public repository for Vulkan Ecosystem issues
Apache License 2.0
132 stars 15 forks source link

About machine learning compute extension #45

Closed nihui closed 3 years ago

nihui commented 5 years ago

Hello, I saw a new effort on the "Vulkan state of the Union GDC Mar19" about that the Vulkan ecosystem is working on exposing machine learning computing capabilities on modern GPU. At the moment, our open-source project "ncnn" uses the Vulkan API to implement nerual network inference on GPU through writing lots of compute shaders. It has been working well on various platforms. We've achieved better memory management and used extensions such as VK_KHR_16bit_storage and VK_KHR_shader_float16_int8 to get really good compute acceleration. I think it would be better and run faster if we could use the new machine learning compute extension for the nerual network operator computing. I am very interested in this and would like to get some more details and usage design of the extended interface. Whether a separate pre-defined-command/pipeline (similar to CUDNN) has been designed for each nerual network operator ? Whether an operator definition based on the current existing will be used for the operator parameter convention(such as ONNX or NNEF) ? the ncnn community, as a potential target user, thank you very much for your fantastic efforts :)

project link: https://github.com/Tencent/ncnn

bhack commented 5 years ago

https://www.phoronix.com/scan.php?page=news_item&px=SIGGRAPH-2019-Vulkan-ML

marty-johnson59 commented 3 years ago

This repository is being archived as it has been replaced with the vulkan.org website and is no longer being maintained (i.e., issues posted here are no longer being addressed ). After reviewing issues posted here, most (if not all) have been resolved or have already been re-opened in Vulkan-Docs (https://github.com/KhronosGroup/Vulkan-Docs) or other repositories for further consideration. Therefore, all issues in this repository will be closed. If you believe your issue has not yet been resolved, please re-open in Vulkan-Docs. Thanks!