Parallelizable jpeg-like compression using only DCT, stages of length coding, is compression / performance possible?

Question

We need to compress the thin (monochrome) image data and quickly move it. If you just needed to use the parallelizable jpeg compression stages (DCT and run length coding of quantized results) and run it on the GPU, therefore each block is compressed in parallel. I hope that it will be very fast and still very important compression ratio, for example, full jpeg.

Does anyone with more GPU / image compression experience have an idea of ​​how this compares both compression and performance using libjpeg on the processor? (If this is a stupid idea, do not hesitate to talk about it - I am very new to my knowledge of cuda and the various stages of jpeg compression.) Of course, this will be less compression and I hope (faster), but I do not know how significant these factors can be .

Answer 1

You can hardly get more compression in the GPU - algorithms with a fairly complex algorithm that can use this power of MUCH.

alos, JPEG, , PCI-E ( , DMA).

, DMA, "".

10- , , .