Nero gets CUDA support

[Wreckage]

http://www.beyond3d.com/content/news/723

Ahead announced support for NVIDIA's CUDA in Nero's Move it application today at CeBIT. The encoder is targetted at people wanting to create AV for mobile devices like the PSP, iPhone and T-Mobile G1, and for low-def online use on places like YouTube, but it can also do Full HD stuff as well. Ahead seemingly use CUDA to offload a portion of the encode process to the GPU, much like Badaboom does, and the speedups they tout are as impressive as expected. Availability isn't until April for the CUDA-enabled version, but you can go ahead and order it now from nero.com. There's support for most popular codecs, including H.264 (Advanced Profile to boot) included in the 50 Euro price.

 

[exar333]

I saw this article too - has anyone here used Nero's "Move It" application? If so, is it any good? How does it compare to say Badaboom or other CUDA-supported encoders?

 

[Keysplayr]

Nero has a 15 day try before you buy for Nero Move it.
Trial limitation does not support MPEG-4.

http://www.nero.com/enu/moveit-introduction.html

I haven't used it as I'm not that big on handhelds and watching movies on a ipod or other device.

But, if you can try it for free to see what it's like without the CUDA enhancement, and then see what it can do after, It might be of interest to some.

 

[Idontcare]

I'd like to add to this thread my experience with CUDA and encoding, but not with Nero, rather using TMPGEnc 4.0.

Initial thread here: http://forums.anandtech.com/me...id=28&threadid=2259708

Since that thread I have used the TMPG CUDA encoding and it is pretty slick. Basically you enable the option of allowing CUDA at the encoder's leisure, it runs an initialization test where it determines which is best for your specific rig (CPU and GPU) for any given filter and encoding option.

From that point onwards when you run an encoding job, depending which filters/options you used in your job, the encoding chooses the most optimal CPU/GPU processing mix to minimize the encoding time while adhering to the user's quality specifications.

It's not CPU or GPU exclusive, both are used, just in varying ratios. I love it.

(hope this isn't viewed as a thread derail, just wanting to add value to the CUDA encoding theme)

 

[soccerballtux]

Wow that's great!

 

[josh6079]

Is there anyway to take a normal DVD movie and upscale it to 1080p? Would CUDA be involved for that?

 

[nosfe]

programs cannot make up detail, they can only fake it

 

[Keysplayr]

Originally posted by: josh6079
Is there anyway to take a normal DVD movie and upscale it to 1080p? Would CUDA be involved for that?

Download the demo and try. You'll be stuck with CPU only until April AFAIK. I have no idea, yet, of the capabilities of Nero Move It. I've downloaded the demo myself, but haven't had the time to try it yet. Maybe by the weekend.

 

[KutterMax]

Interesting stuff.

I've been playing around with Badaboom this past week and have been really impressed with how much faster I can encode video using my GTX280.

The developers seem to be continually working on Badaboom, trying to improve it. I've encoded a number of DVD's that will stream from my PC to PS3 and this has worked great. Unfortunately it is still crashing when I try and encode video from a ripped Blu-ray. Apparently there is a known bug and it is being worked on.

I am glad other developers are getting on board with CUDA.

 

[Andrew1990]

Originally posted by: KutterMax
Interesting stuff.

I've been playing around with Badaboom this past week and have been really impressed with how much faster I can encode video using my GTX280.

The developers seem to be continually working on Badaboom, trying to improve it. I've encoded a number of DVD's that will stream from my PC to PS3 and this has worked great. Unfortunately it is still crashing when I try and encode video from a ripped Blu-ray. Apparently there is a known bug and it is being worked on.

I am glad other developers are getting on board with CUDA.

How fast are the encodings on that GTX280? I should be getting my GTX285 soon and was wondering how fast it could encode a DVD.

 

[Denithor]

Originally posted by: Idontcare
I'd like to add to this thread my experience with CUDA and encoding, but not with Nero, rather using TMPGEnc 4.0.

Now, that looks like a slick software package, well-integrated with your hardware for sure. What's the IQ like? How comprehensive are the settings for resolution, etc?

 

[Idontcare]

Originally posted by: Denithor

Originally posted by: Idontcare
I'd like to add to this thread my experience with CUDA and encoding, but not with Nero, rather using TMPGEnc 4.0.

Now, that looks like a slick software package, well-integrated with your hardware for sure. What's the IQ like? How comprehensive are the settings for resolution, etc?

It's actually seamless with the package and is CUDA oblivious.

You specify the compression levels in the software just as you would if you were relying on the CPU without CUDA.

The program knows what it can task to the GPU without sacrificing IQ, etc. I get identical output whether I use Cuda or not. Not all features will use cuda though I have noticed, some encoding jobs for me run 100% on CPU.

 

[Dadofamunky]

Originally posted by: Idontcare

Originally posted by: Denithor

Originally posted by: Idontcare
I'd like to add to this thread my experience with CUDA and encoding, but not with Nero, rather using TMPGEnc 4.0.

Now, that looks like a slick software package, well-integrated with your hardware for sure. What's the IQ like? How comprehensive are the settings for resolution, etc?

It's actually seamless with the package and is CUDA oblivious.

You specify the compression levels in the software just as you would if you were relying on the CPU without CUDA.

The program knows what it can task to the GPU without sacrificing IQ, etc. I get identical output whether I use Cuda or not. Not all features will use cuda though I have noticed, some encoding jobs for me run 100% on CPU.

In optimal cases, roughly how much time does CUDA shave off your encoding jobs?

 

[ViRGE]

Originally posted by: Dadofamunky

Originally posted by: Idontcare

Originally posted by: Denithor

Originally posted by: Idontcare
I'd like to add to this thread my experience with CUDA and encoding, but not with Nero, rather using TMPGEnc 4.0.

Now, that looks like a slick software package, well-integrated with your hardware for sure. What's the IQ like? How comprehensive are the settings for resolution, etc?

It's actually seamless with the package and is CUDA oblivious.

You specify the compression levels in the software just as you would if you were relying on the CPU without CUDA.

The program knows what it can task to the GPU without sacrificing IQ, etc. I get identical output whether I use Cuda or not. Not all features will use cuda though I have noticed, some encoding jobs for me run 100% on CPU.

In optimal cases, roughly how much time does CUDA shave off your encoding jobs?

This is a bit off-topic, but if you want a truly optimal case, you have to take a look at not just using the GPU to encode, but using the GPU to process the material too. I just had a situation the other day where I was working on a fully-interlaced MPEG-2 file. The best deinterlacer I have for such material is MCBob for Avisynth, which bobs the video and then runs a bunch of searching for motion compensation. It takes a 30fps interlaced video and spits out a 60fps video (1 frame for each set of interlaced fields). Even on a quad core processor, the entire processing chain (decoding -> resizing -> bobbing -> losslessly encoding) took something around 16 hours; virtually all of that time is spent on MCBob.

In comparison, the PureVideo deinterlacer can do the same thing in real time. Unfortunately I don't have a GPU with the VP2 decoder (the only software package out there right now that interface PV + CUDA with avisynth requires VP2), but with the CPU freed from everything but lossless encoding, the specifications are that it should be able to do the whole process in real time (it may be able to do faster than RT, but I'd need to test it). Since this was a 90 minute video, this is a 10x speedup over using the CPU.

This is an optimal edge case, most people will not be encoding video that requires such heavy processing (I would expect most stuff needs simple IVTC at best). None the less, in those edge cases the speed improvement is immense.

 

[s44]

Is MCBob really better than (the much faster) yadif?

 

[ViRGE]

Originally posted by: s44
Is MCBob really better than (the much faster) yadif?

I have no idea. This was the first time I've ever used MCBob.

 

[Keysplayr]

Originally posted by: Dadofamunky

Originally posted by: Idontcare

Originally posted by: Denithor

Originally posted by: Idontcare
I'd like to add to this thread my experience with CUDA and encoding, but not with Nero, rather using TMPGEnc 4.0.

Now, that looks like a slick software package, well-integrated with your hardware for sure. What's the IQ like? How comprehensive are the settings for resolution, etc?

It's actually seamless with the package and is CUDA oblivious.

You specify the compression levels in the software just as you would if you were relying on the CPU without CUDA.

The program knows what it can task to the GPU without sacrificing IQ, etc. I get identical output whether I use Cuda or not. Not all features will use cuda though I have noticed, some encoding jobs for me run 100% on CPU.

In optimal cases, roughly how much time does CUDA shave off your encoding jobs?

We are probably talking in the area of "hours to minutes" or "minutes to seconds". Of course it will depend on the task at hand, like what kind of job you're doing, what kind of files and what you're converting/encoding/transcoding to.

 

[Idontcare]

Originally posted by: Dadofamunky
In optimal cases, roughly how much time does CUDA shave off your encoding jobs?

I haven't done any homework to compare encoding times with and without Cuda, I just assumed at worst it would not be any slower having CUDA enabled so most of my "characterization" efforts were dedicated to ensuring my output video wasn't fubar. (which none of it has been to date)

TMPGenc claims a bunch of speedup under various filter setups, I took them for their word (knowing they were absolutely showing best cases on their website no doubt) and figured my mileage will vary between having zero speedup (but no harm done) to their claimed speedup (also with no harm done).

Checkout their self-published benchmarks: http://tmpgenc.pegasys-inc.com...roduct/te4xp.html#tabs

 

[rbV5]

Originally posted by: josh6079
Is there anyway to take a normal DVD movie and upscale it to 1080p? Would CUDA be involved for that?

Your graphics card already handles scalng in real-time, you don't want to encode to a higher resolution.