Dear SNPE developers,
Since SNPE 1.12.0 is released, it says that GPU_FLOAT16 runtime is supported.
But when I set GPU_FLOAT16 instead of GPU runtime, the network outputs nothing.
Do I have to convert model to 16-bit model or something else? I have found nothing about how to use GPU_FLOAT16.
The documents introduce about what would happen in GPU_FLOAT16 runtime, but do not say about how to run in GPU_FLOAT16.
Please guide me to the correct way. Any comment is appreciated!
Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.
Hi All,
It is a bug in my app, it turns out that I did not return the correct runtime.
But GPU_FLOAT16 is not much faster than orignal GPU mode, about few miliseconds on 820?
The performance impact depends on the network. Some networks see 10 - 20% improvement in execute time using GPU_FLOAT16.