Frauenhofer HEVC / GPU&CPU Encoder

[Easy]

Good evening, I have found this site:

https://hevc.hhi.fraunhofer.de/trac/hevc/browser | Newest

Is this a new distribution of Hevc or a mirror repo?
-and where can I find binaries of it?


A general question: Is there a HEVC encoder that use both GPU and CPU, so in one, the GPU for the not exact algorithms, and the rest with the CPU? Shouldn't the FPS go up? Or not because it takes to long to load variable from GPU after CPU?(The same by Intel GPU?)
And were is the difference between Intel GPU (Quick Sync) and Nvidia GPU (CUDA?) compared in accurate and speed?

Best regards
Easy

[Parabola]

Hi Easy - You've found the reference software - the original HEVC implementation that was used by the people who researched and designed the standards. You probably don't want the binaries as they will run incredibly slowly and lack many practical features you need for real-world use. But this software will have good encode quality and is the reference model for decoding.

On CPU/GPU: unless the GPU has hardware specially designed for HEVC, there are limited benefits to using it to accelerate video encoding or decoding. If a GPU does have dedicated hardware then generally it will be pretty inflexible and only operate in the way that the designers anticipated. Historically quality was dubious too although it looks like this is improving. For hybrid CPU/GPU processing, data bandwidth (between separate CPU and graphics DRAM) can be an issue depending on system architecture.

[Easy]

ok thanks Parabola for the detailed answer

I thought only certain standard post-analysis algorithms would possibly settled faster in the GPU. There is surely something like AVX or SSE for GPU's latest generation.
And for example the intra prediction is computed ahead in parallel, and the CPU performs the actual calculation for the CU's
But since there is no, you cann't do anything

Are there any other Intel QS encoder as the Japanese QSVenc?
What are the advantages of DivX and kvazaar? Respectively are they useful in addition to the X256 version?

[vivan]

Originally Posted by Easy

And were is the difference between Intel GPU (Quick Sync) and Nvidia GPU (CUDA?) compared in accurate and speed?

QuickSync is intel hardware (ASIC, fixed-function) encoder. NVidia's alternative is NVENC, ATI's is VCE.
CUDA is API for running code on GPU, it's alternative is OpenCL.

Hardware encoders are fixed-function, they do all the work and the best you can do is to fiddle with couple of settings. They're fast, but they're far from beating best cpu encoders in quality-per-bit ratio. That's why trying different encoders doesn't make much sence - they still use same fixed-function hardware.

Encoders written solely for CUDA/OpenCL are even worse, and this is the reason why all GPU manufactures decided to include ASIC encoders instead of developing those (they actually tried at first).

There were efforts to use CUDA/OpenCL for partial encoding acceleration, the most notable is OpenCL accelerated lookahead in x264. Still, it wasn't really useful (few percent faster in some cases, while slower in other).

Originally Posted by Easy

What are the advantages of DivX and kvazaar? Respectively are they useful in addition to the X256 version?

In addition to? They are just other HEVC encoders.
If there were any other competitive encoders there would've been at least threads about them. But since there aren't...