View Full Version : Is it possible to modify x264 to use GPU power to speed-up encoding ?
St Devious
7th November 2007, 18:55
As the title says ?
Wat r the challenges of doing so ?
Anyone considering to do it in the future ?
Sharktooth
7th November 2007, 18:56
good luck...
Romario
7th November 2007, 18:59
who knows.
audyovydeo
7th November 2007, 19:00
I for one would settle for GPU decoding x264, before asking for it to assist encoding
;-)
cheers
audyovydeo
Dark Shikari
7th November 2007, 19:14
Encoding algorithms generally require loads and loads of branching, which video cards aren't very good at. They'd do a great exhaustive search, but of course x264's exhaustive search is already optimized quite a bit with its SEA algorithm, which would mean the video card would have to be vastly faster than the CPU to be at all useful.
Sagekilla
7th November 2007, 19:43
Encoding algorithms generally require loads and loads of branching, which video cards aren't very good at. They'd do a great exhaustive search, but of course x264's exhaustive search is already optimized quite a bit with its SEA algorithm, which would mean the video card would have to be vastly faster than the CPU to be at all useful.
Fortunately, in the aspect of branching instructions, the Radeon X1K series does very well with branching. 7xxx series does worse (It takes longer to perform a branch, on the order of a few magnitudes I think it was) and I have no idea how well the 8xxxx and HD 2k series does with this. If I find the article, I'll show you, but branching isn't a problem anymore with certain GPUs.
Edit: On the point of a ESA search, that is something that a GPU would be vastly better at than a CPU. In fact, this is the very thing they excel at since it's processing huge amounts of data in parallel, which is the very design goal of a GPU. But like you said, there's the branching issues and it'd have to be done on a fairly specific generation of GPUs to see any improvement. On a side note, I remember a thread started about the "fastest AVC encoder," which ended up being one from AMD/ATI. I wonder how much of the code was actually accelerated on the GPU, and how much of a performance boost it meant. Would be interesting to grab one of the supported cards and underclock it to see how well it performs at lower speeds.
foxyshadis
8th November 2007, 00:16
Okay, we just had one of these about a week ago (https://forum.doom9.org/showthread.php?p=1060890&highlight=gpu#post1060890), this is getting ridiculous. Use search before posting new threads!
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.