nvJPEG scaled decoding like IJLs IJL_JFILE_READONEEIGHTH

does nvJPEG support scaled decoding, like IJL with IJL_JFILE_READONEEIGHTH?

In my application case, I would only need 1/64 of the GPU memory if this option were available

Hi lc640, Thanks for your query. nvJPEG currently does not support scaled decoding.
We will evaluate this feature for future releases.

Hi Mandar,

this would be great.

The effort to work with a lot of jpeg-thumbs on the GPU is very high at the moment,
I can only decode the JPeg in full size, then it has to be scaled down, the memory of the Jpeg frame must be released immediately to keep a large number of Thumbs in the GPU memory. Scaled decoding would be very helpful - it saves memory bandwidth and GPU usage.

This procedure has already proven itself on the CPU

Scaled decoding is on one of the features i’m missing in H264 / H265 etc as well.

Working on big Streams (4K 8K) is very expensive regarding GPU-Memory. (I’m missing this feature in Intel Media SDK, too)

NVDEC supports immediately scaling H.264 for you. Set ulTargetWidth and ulTargetHeight in CUVIDDECODECREATEINFO when you create the decoder.