GPU transcodes four 1080p video files in real time
 

I've been an avid gamer for some time, and recently there have been some developments in video gaming GPU stream processing that look like they may have a huge impact on desktop editing. The potential processing power within gaming video cards now days has past the teraflop range. Both ATI and NVidia are beginning to promote open or semi open standards which expose this power for any application to use, and some programs are beginning to appear. Cyberlink have just released an app that can take four mpeg2 streams and re-encode into mpeg4 in realtime simultaneously on what is essentially a $200US video card.
The possibilities for this power to be used within a decent editing app seem pretty impressive, enough to encourage my first post anyway :D Add to this the ability to chain up to four of these cards together in an off the shelf PC, and you begin to get the idea. ATI have also just announced a real-time full HD raytracing render pathway for these same cards. Hopefully Adobe, Grassvalley, et al, produce plugins to leverage these cards in the not too distant future!

http://www.tomshardware.com/news/Lar...cing,5769.html

http://www.tgdaily.com/content/view/38117/140/

This is one step towards faster batch processing of ingest formats into intermediate formats like ProRes422 or DNxHD or Cineform without CPU power.

This would be very closely related:

http://www.elementaltechnologies.com/index.php

That's kind of what the FXPlug architecture does in FCP. Plug-ins that use FXPlug, tap into the system GPU for more real time efx processing.

-gb-

And just think Pinnacle Liquid was actually one of the first video editing programs to use the gpu long before Apple did. That is why Liquid can handle realtime editing with a software only product that can only be matched with a dedicated hardware based editing card. This is why Liquid can handle native mpeg2 editing so well with multiple streams. It doesn't just use the gpu for effects but for video processing as well.

Pinnacle really gets no credit at all for being one of the first to tap the gpu.

well, it's about time! =D. now they just have to make it available for ati and other NLEs

Anything that speeds things up is fine by me - not that the existence of such high-end graphics cards means much when when your workstation is a Mac!

It will probably be up to the NLE's to figure it out.
Getting back to the original topic, it appears the Cyberlink folks already have for ATI.

http://www.cyberlink.com/eng/press_room/view_1756.html

I wonder if running a couple of HD4850s (Crossfire) would double transcoding performance.

A bit of an update on this subject. Things are looking great for digital processing. Reducing a 3 hour 1080p mpeg2 encode on a quad core to 12 minutes on a 4850.

AMD counters CUDA with renewed ATI Stream initiative - The Tech Report

i was just about to post this, lol.

this is AWESOME!!! it's about time, this should've been done before the rise of GPUs lol. it's one of those "why didn't they do that before?" type of an idea =). hopefully PCs adopt it too not just max.. otherwise, it'll become exclusive =)

There is an early app in the new ATI 8.12 drivers. Pretty simple, but pretty quick. I just re-encoded a 4.7GB DVD VOB to a 700MB DIVX in 13 minutes on a cheap AMD 5200+ dual core with an ATI 4850. Now the HAL is open I hope other vendors make use of it quickly.

Some test indicates that some GPU acceleration are great on speed , but delivers lower quality encoding.

The quality, good or bad, is entirely due to the software using the GPU. The math of 1+1 doesn't create a good quality 2 or a low quality 2 :) When proper implementations of transcoders appear, the GPU will be a boon to real time HD editing.

nVidia has been pushing this only for a few years. Graphics cards in the past didn't offer enough precision and processing "pipelines" to make this worthwhile. Also, the cards were designed for graphics, which basically means that you send higher level descriptions (lower bandwidth) of the graphics and the cards convert them to pixels without a fast data path back to the main processor.

I'd really like to see GPUs used more for JPG decompression. Right now, GPUs just sit idle while feasting on electricity. The computer sales chain is responsible for burning a lot of electricity for no benefit other than heating your room.