CS5: CUDA acceleration on unsupported nvidia GPUs possible!

[Cattykit]

http://www.dvinfo.net/forum/adobe-c...e-cs5-work-gtx-295-possibly-all-200-gpus.html

http://www.cinema5d.com/viewtopic.p...k=t&sd=a&sid=e9568886c78d5b9da1f876c7aeaa7679

http://forums.adobe.com/thread/629557?tstart=0

http://forums.adobe.com/thread/632143?start=0&tstart=0

With my GT 240 DDR3 1024mb model, I saw no hope of being able to use Mercury engine. With such a simple trick, it now works damn well on my cheap GPU.

Below are simple tests I've done:

CPU: i7 860 @ 3.6 ghz
RAM : 8gb DDR3
GPU : GT 240 DDR2 1024mb

4:12. 30fps, 40mbps, h. 264, 5d mk2 video files
about 7 color correction and unsharpen filters applies.

Preview performance (100% resolution)
Off: Jerky previews that make editing near impossbile or pain in the ass.
CPU usuage : 94%
On: Smooth.
CPU usuage : 92%

Encoding performance (VBR 1 pass. Target rate: 10mbps, Max. rate 15mbps)
Off: 30 minutes
On: 15 minutes

Simple conclusion:
It made full resolution editing possible.

2x increase in encoding performance is huge!
Previously, for that 2x performance, I had to upgrade the whole damn computer (From Q6600 @ 3.2ghz to i7 860 @ 3.6ghz).

Such huge improvement was brought by cheap GT 240 that only has 12 multi processors and 96 cuda cores. I can't imagine what Fermi can do.

 

[BTRY B 529th FA BN]

Apparently this is from the Mercury Playback Engine only with Premiere Pro CS5!

I can't wait to get my upgrade from CS4!

 

[MarkLuvsCS]

I'd love to try this with my new gtx470. I'm just not entirely sure what features would be able to stress the GPU more than anything. I tried some basic encoding and it didn't seem to load the gpu according to GPU-Z. I'm not sure how to check the GPU otherwise to see if it's being stressed with CUDA applications.

So far the card seems to OC decently. stable 800gpu/1600shader/1000x4 memory @ 1v. Stable temps at ~70C under full load in Furmark but typically only ~60Cish ingames. Just a Galaxy GTX470 seems to handle the abuse OK

If you'd like to see what fermi could do with same setup just pass me along the settings and any video u want me to play with .

 

[tweakboy]

These are horrible stats,, now I know I wont be playing this one, as I demand hi IQ ,,, hope all is well there chief

 

[NoQuarter]

is tweakboy just a script?

 

[faxon]

were not really sure what he is yet. i think that it depends on the last time he took his meds. sometimes he is really lucid, but other times it's just.....what?

 

[BTRY B 529th FA BN]

I say a temporary b*n

 

[Ben90]

I say a temporary b*n

Concurred!

 

[0roo0roo]

is adobes encode actually high quality? all previous h264 cuda encoders were gimped in quality, good enough for ipod encodes, but not much else.

 

[BenSkywalker]

all previous h264 cuda encoders were gimped in quality, good enough for ipod encodes, but not much else.

What is wrong with Badaboom precisely? I use it regularly and haven't had any quality issues noticeable on my 46" display, I think that qualifies as above iPod. Or are you talking about a couple of years ago when it was still a new product?

On the video side, there was no discernible difference between the original input and output produced by either Badaboom or Handbrake. Both also worked well for upscaling a 480p DVD to 720p and had no problems downscaling for smaller devices like the iPhone.

http://www.overclockersclub.com/reviews/badaboom_1_1_1/4.htm

That's from over a year ago.

 

[0roo0roo]

What is wrong with Badaboom precisely? I use it regularly and haven't had any quality issues noticeable on my 46" display, I think that qualifies as above iPod. Or are you talking about a couple of years ago when it was still a new product?



http://www.overclockersclub.com/reviews/badaboom_1_1_1/4.htm

That's from over a year ago.

they weren't looking very closely ...every review i've seen on the cuda encoders has always pointed to inferior quality. the different encoders were consistently inferior, so i figured there was something inherent in the gpu encode scheme as implemented that made that so. sure at a high enough bitrate it won'tn be that noticeable, but bit for bit the more detailed reviews found that encodes were consistently inferior. it wasn't free speed.

 

[0roo0roo]

here's anadtechs own review
http://www.anandtech.com/show/2586/4
maybe they fixed it, but i dunno about that.

 

[Voo]

here's anadtechs own review
http://www.anandtech.com/show/2586/4
maybe they fixed it, but i dunno about that.

No I'm sure they didn't do anything with the "near-final build" that AT tested compared to the 1.1 version..

Would be interested in some other reviews as well, but it'd be nice if those would be testing new versions.

 

[MarkLuvsCS]

I purchased Bababoom to test the video encoding but with windows 7 64bit + GTX470 its a no go. Some posts seem to indicate there should be another release Q3/Q4 this year to add support for the new cards. The forums are pretty sparse with any "recent" information being weeks old at best.

 

[0roo0roo]

No I'm sure they didn't do anything with the "near-final build" that AT tested compared to the 1.1 version..

Would be interested in some other reviews as well, but it'd be nice if those would be testing new versions.

i remember asking about it on handbrake forums, they claimed there were issues inherent to the process, which is why they weren't going to bother implementing it in handbrake. doom9 type video forums where they nitpick such stuff also seemed to agree that such encodes were inferior last i checked. other implementations other than badobooms also have had history of lower quality as well. its just a lot of evidence that its not the same/up to snuff as regular encoding.

 

[manimal]

These are horrible stats,, now I know I wont be playing this one, as I demand hi IQ ,,, hope all is well there chief

I think its performance art. Andy Kaufman-esque!

 

[Zoeff]

I purchased Bababoom to test the video encoding but with windows 7 64bit + GTX470 its a no go. Some posts seem to indicate there should be another release Q3/Q4 this year to add support for the new cards. The forums are pretty sparse with any "recent" information being weeks old at best.

I've been using MediaCoder for a while now with a 8800 GTX and it gives me about 3 to 4 times the speed of my i7 920 @ 4GHz. There are a few on their forums that appear to be using a GTX470 as well.

It's also free software... ;x

 

[Voo]

i remember asking about it on handbrake forums, they claimed there were issues inherent to the process, which is why they weren't going to bother implementing it in handbrake. doom9 type video forums where they nitpick such stuff also seemed to agree that such encodes were inferior last i checked. other implementations other than badobooms also have had history of lower quality as well. its just a lot of evidence that its not the same/up to snuff as regular encoding.

Yeah everything I've seen/read so far also seems to support the inferior theory. The only thing is: I haven't seen any reviews with current versions of badaboom but the one linked and it seems like they got rid of a lot of problems.
After all there's no inherent reason that GPU accelerated encoding should be inferior, getting it right is just much more complicated than the usual stuff, so it's not too far off I think to assume that there will be lots of quality improvements while developers get better with Cuda and Co.

 

[Zoeff]

Yeah everything I've seen/read so far also seems to support the inferior theory. The only thing is: I haven't seen any reviews with current versions of badaboom but the one linked and it seems like they got rid of a lot of problems.
After all there's no inherent reason that GPU accelerated encoding should be inferior, getting it right is just much more complicated than the usual stuff, so it's not too far off I think to assume that there will be lots of quality improvements while developers get better with Cuda and Co.

I never noticed any quality differences between x264 and mediacoder. Then again I didn't realize that there might be some so I never checked either.

Just looked at some 1080p footage before and after transcoding and it looks fine to me.

 

[Idontcare]

is tweakboy just a script?

I'm 100% convinced of this, have been for a while, every new post just continues to confirm. The syntax is waaaay to consistent for it to be just some 12yr old operating at a 3rd grade level for grammar. Prolly some college kid's computer sci project to punk AT forums. Would not be the first, wouldn't even make it into the top 100.

 

[0roo0roo]

I never noticed any quality differences between x264 and mediacoder. Then again I didn't realize that there might be some so I never checked either.

Just looked at some 1080p footage before and after transcoding and it looks fine to me.

the results were never so terrible that it would jump out at you, but for a given bitrate the cuda encodes were always slightly inferior. so if your goal is to get the best quality possible i have doubts its the best route. if you are encoding for an ipod or such, then its fine.

 

[Zoeff]

the results were never so terrible that it would jump out at you, but for a given bitrate the cuda encodes were always slightly inferior. so if your goal is to get the best quality possible i have doubts its the best route. if you are encoding for an ipod or such, then its fine.

Just did another frame by frame comparison this time fully awake. There does seem to be just a little bit less detail in area's with the least amount of contrast however I'm not entirely sure if these are perceptible in any way though...