Vegas Pro 10 - GPU Rendering Results

MelvinG wrote on 10/11/2010, 2:16 PM
I just tried out Vegas 10's GPU-accelerated AVC render and it was slower than my Vegas 9 rendering. The simple test project file was a series of 720p AVC clips crossfading into each other and on Vegas 9 the AVC render took 18:20 while with Vegas 10 on "Main Concept AVC it took 24:25 and with the Sony AVC setting it took 19:30.

My system is an Intel Quadcore Extreme, 8GB RAM, 15Krpm SCSI Drives, nVidia GTX 480 (1.5GB DDR5 VRAM), running Win 7 64 Ultimate with the 64 bit version of Vegas Pro 9 & the 64 bit Vegas Pro 10 demo version.

I called tech support to see what's wrong and if I have to download the Sony AVC plug-in separately and enable it or whether it's already included in the v10 Package. The lady told me that it's already included but she was at a loss to explain how things can be now slower when I'm running nVidia's top of the line GTX 480. She said that I should be seeing about a 5% improvement in speed and that she was going to pass on my info and findings to the dev team. Supposedly they're working on improving more GPU functionality for future updates/versions.

Comments

Wolfgang S. wrote on 10/11/2010, 2:34 PM
Well, with my Q6600, win7 64bit, 6 gb ram and an old nVidia 9800 GT I see an improvement of 26% for rendering AVCHD 1080 50i footage (most crossfading), when I enable GPU rendering versus CPU only. But I have not tested 720p nor compared that with the mc avc encoder yet.

Desktop: PC AMD 3960X, 24x3,8 Mhz * RTX 3080 Ti (12 GB)* Blackmagic Extreme 4K 12G * QNAP Max8 10 Gb Lan * Resolve Studio 18 * Edius X* Blackmagic Pocket 6K/6K Pro, EVA1, FS7

Laptop: ProArt Studiobook 16 OLED * internal HDR preview * i9 12900H with i-GPU Iris XE * 32 GB Ram) * Geforce RTX 3070 TI 8GB * internal HDR preview on the laptop monitor * Blackmagic Ultrastudio 4K mini

HDR monitor: ProArt Monitor PA32 UCG-K 1600 nits, Atomos Sumo

Others: Edius NX (Canopus NX)-card in an old XP-System. Edius 4.6 and other systems

Rob Franks wrote on 10/11/2010, 2:36 PM
I just tried out Vegas 10's GPU-accelerated AVC render and it was slower than my Vegas 9 rendering.

I'm running a GTX285 and I can report pretty much the same thing. I'm not sure if it's SLOWER... but there is absolutely no speed increase that's for sure. The avchd is much quicker on the time line.... but rendering... no.
Rob Franks wrote on 10/11/2010, 2:38 PM
, when I enable GPU rendering versus CPU only

Where do I find this switch???
Rob Franks wrote on 10/11/2010, 3:04 PM
Never mind... found the switch.

Comparatively speaking between the cpu and gpu rendering... the gpu certainly is faster.... but it sure doesn't 'feel' much faster than vegas 9. Maybe just my imagination?
farss wrote on 10/11/2010, 3:15 PM
Remember it's ONLY accelerating the encoding. That's generally not what the CPU would spend most of its time doing anyway. Getting the uncompressed frames of vision to the encoder (whatever it is) can involve way more work than what the encoder itself has to do.

Its also not impossible to imagine that sending those frames to the GPU and getting the encoded data back may be slower than using the CPU itself.


Bob.
Rob Franks wrote on 10/11/2010, 3:43 PM
"Comparatively speaking between the cpu and gpu rendering... the gpu certainly is faster.... but it sure doesn't 'feel' much faster than vegas 9. Maybe just my imagination?"

Well.... maybe I lied.

With a given time line I get;

Vegas 10
cpu render: 6:13
gpu render: 4:27

Vegas 9 (Same t/l same settings): 4:49

So with gpu it appears to be a bit faster than 9.... but with cpu it is most definitely SLOWER than 9
jabloomf1230 wrote on 10/11/2010, 4:31 PM
You have to understand one thing, before you prejudge the CUDA-based h.264 encoder that is included with Vegas. Getting the most out of it requires a balance of performance among all the major components of your system, including:

1) The GPU.
2) The CPU
3) The RAM
4) The disk subsystem.

One of those items will always be the limiting factor in performance, especially, when the software relies on CUDA. For example, if you have a slow, dual core CPU, buying a GTX 480, may only marginally improve encoding performance, if at all. If you have a system performance program , like Everest, for example, that uses an on-screen display that can run in RT with your software:

http://www.lavalys.com/products/everest-pc-diagnostics

you can see how much the GPU and CPU are used for any given task.

Maybe it's time for John Cline to update his benchmark, so we can test performance in a similar manner that Premiere Pro CS5 users utilize their benchmark, PPBM5:

http://ppbm5.com/
MelvinG wrote on 10/11/2010, 5:03 PM
Rob, where did you find this switch to enable GPU rendering in Vegas 10? I browsed through the preferences section and the help section but did not see the switch and tech support did not indicate there was a switch when I inquired with them.
jabloomf1230 wrote on 10/11/2010, 6:07 PM
Select Sony AVC for rendering, Then click the Custom button. Down at the bottom of the Custom Rendering dialog box, you will see an option "Encode mode". There's three options:

Automatic - Lets Vegas decide whether to use the GPU and CPU, or just the CPU.
Render using CPU only
Render using GPU if available
LSHorwitz wrote on 10/11/2010, 6:39 PM
I have also been doing time trials with the new Vegas 10 64 bit for AVCHD, using a 6 core Intel 980X Extreme, 12 GB of RAM, and a very fast nVidia 470 card.

I am entirely disappointed to see about 6% speed improvement when the GPU is enabled versus the "CPU only" mode. Some other programs I own show a profound benefit from the same GPU versus the CPU only.

I have to believe, based on using other programs, that AVC encoding can be accelerated with some parallel processing on the GPU to a much greater extent than Sony has achieved. A lack of 'smart rendering' and good GPU support for those of us doing AVCHD editing and rendering is a huge disappointment for this new version of Vegas.

Larry
kkolbo wrote on 10/11/2010, 6:51 PM


It is in the render dialog box when you choose custom from a Sony AVC render.

MelvinG wrote on 10/11/2010, 9:05 PM
I was led to believe by Sony tech support that the devs know about this and they're committed to keep working on getting more and more functionality from the GPU in future updates and versions. With Adobe as a leader in this GPU-applied NLE field Sony will hopefully be pressed hard to move more quickly in this development for its customers.

I've only rendered WMV-HD 720P in the past but in my test between that format and AVCHD at 720P with the same project file, there was no discernible difference in quality.
A. Grandt wrote on 10/12/2010, 2:12 AM
Melvin, hopefully Sony don't move TOO quickly with the development; in VP9 we saw what happens when they do.
megabit wrote on 10/12/2010, 3:19 AM
If you take a look at this page:

http://www.nvidia.com/object/cuda_gpus.html

- you will see the "compute capability" index of various nVidia cards.

While for the specialized CUDA units (like Tesla, or some Quadro cards) the index is 2.0, it's only 1.0 for some lower-end Quadro FX cards (like the FX 4600).

For gaming cards, even though they do support CUDA, no compute capability index is even listed. This might explain why those of you using such a card, especially with very powerful CPUs, do not see any speed advantages when rendering H.264 in VP 10...

Piotr

AMD TR 2990WX CPU | MSI X399 CARBON AC | 64GB RAM@XMP2933  | 2x RTX 2080Ti GPU | 4x 3TB WD Black RAID0 media drive | 3x 1TB NVMe RAID0 cache drive | SSD SATA system drive | AX1600i PSU | Decklink 12G Extreme | Samsung UHD reference monitor (calibrated)

farss wrote on 10/12/2010, 4:33 AM
Think about this.
If you apply an FXs or scaling to a frame it has to go through the CPU. Vegas generally does not use the GPU to accelerate FXs. Now to encode the frame you have two choices.
1) Do the encode in the CPU.
2) Send the frame to the GPU and read it back.

Not too hard to image 1) being faster than 2).Of course H.264 is a variable length GOP codec so whatever is processing it, it needs to buffer quite a few frames. That's quite apart from muxing in the audio etc, etc and no doubt other wrapping required.
So far the only people expecting miracles from using GPU accelerated H.264 encoding are Vegas users. Worse when it fails to materialise they're complaining. Adobe don't think it's worth much, SCS never promised a miracle. We're starting to sound like the guys from Reduser.net a couple of years ago :)

Bob.
Fotis_Greece wrote on 10/12/2010, 4:57 AM
When doing GPU rendering use the gpuz utility and note the sensors. Check the gpu load sensor to see if it is actually using gpu to render
http://www.techpowerup.com/downloads/1864/TechPowerUp_GPU-Z_v0.4.6.html
megabit wrote on 10/12/2010, 11:23 AM
Bob,

Yet, when I was testing Cyberlink PowerDirector with one of those Quadro cards, the H.264 transcode was a breeze when compared to CPU only...

Piotr

AMD TR 2990WX CPU | MSI X399 CARBON AC | 64GB RAM@XMP2933  | 2x RTX 2080Ti GPU | 4x 3TB WD Black RAID0 media drive | 3x 1TB NVMe RAID0 cache drive | SSD SATA system drive | AX1600i PSU | Decklink 12G Extreme | Samsung UHD reference monitor (calibrated)

rmack350 wrote on 10/12/2010, 11:33 AM
Larry, which other programs are you comparing things to?

Rob Mack
reberclark wrote on 10/12/2010, 11:40 AM
Anybody tried this with an ATI card? Is it possible?

I have an ATI Radeon 5830HD.

Win7 Pro
I7 X980 3.33GhZ
LSHorwitz wrote on 10/12/2010, 2:33 PM
Rob,
Three programs which benefit most dramatically from GPU acceleration when rendering AVCHD which I often use are Cyberlink Power Director 8, TMPGExpress 4, and Aimersoft DVD to iPhone Converter. I can also add Badaboom to the list, but it was not yet updated to support the newer "Fermi" GPUs from nVidia such as the GTX470 I presently own. My prior nVidia 8800GT made a big difference when rendering with Badaboom, and I would expect that it will make an even bigger difference with the Fermi GTX470 given that there are hundreds of cores in the new card and only a couple dozen in the old card. Also, although I do not personally own them, the entire suite of Aimersoft transcoders for other platforms beyond the iPhone have the same speed-up advantage, but I personally have no use for transcoding to other mobile devices.

In addition to the "favorite" programs I mentioned above, there are MANY GPU-accelerated programs, some of which I also occasionally use such as DVDFab Platinum, and ALL of them get a significant boost from the GPU even on this very high performance (6 core i7 980X Extreme) CPU.

In fact, Vegas 10 is the ONLY program which shows such a miniscule (5 to 6%) gain, making me believe that the programmers at Sony have NOT implemented the CUDA code in any remotely optimal fashion.

The alternative NLE I like to use, and often use, is Edius Neo 2 with AVCHD Booster. I have to say that almost everybody doing AVCHD would see a huge benefit from using the products I mentioned above, since they all use the CUDA GPU and some, in addition, add support for the Leadtek/GrassValley SpursEngine cards. Vegas 10 is still pathetically slow in rendering AVCHD, and has not made any progress relative to all of the other programs out there which offer BOTH CUDA boost as well as Smart Rendering of AVCHD, another feature entirely absent from the new Vegas.

Oh well......maybe when Vegas 11 Pro comes out................

Larry
jabloomf1230 wrote on 10/12/2010, 3:26 PM
There's some mixing apples and oranges in this thread. CUDA is utilized in different ways in the various pieces of software that are being discussed:

1) Badaboom - This is a free-standing h.264 encoder that uses CUDA. It does shorten encoding times much more than the SONY AVC codec used in VP 10. Unfortunately, BB is compatible with only certain input video file formats and besides, unless you are going to output uncompressed from your NLE to load into BB, you lose detail, by creating an extra re-encoded generation. Even if you go uncompressed, the time gained by using the BB encoder is lost by outputting the uncompressed intermediate. BB neither supports Cineform nor any frame serving into it. Some of us here have already tried these options.

2) TMPGenc 4.0 Xpress. This bare-bones NLE uses CUDA for transitions and FX. It does not use a built-in h.264 CUDA-based encoder.

3) Premiere Pro CS5. The "famous" Mercury Playback Engine (MPE) uses CUDA in manner that is identical to TMPGenc 4.0 Xpress. with the MPE, CUDA only speeds up specially designed (not all) transitions and FX. Hence, previewing of the timeline is substantially improved, but encoding (rendering) is only improved a small amount over non-MPE rendering. It surprisingly does not use a built-in h.264 CUDA-based codec for rendering out the timeline (see # 4 below).

4) MainConcept CUDA-based h.264 codec. MC has been trying to get the major NLE vendors to license their codec, but thus far without success. There is no evidence that the SONY AVC codec is somehow derived from the MC codec, although SCS is very clever in disguising where some of it's licensed 3rd-party plug-ins come from.

The other point that I want to make is that generally both consumer and workstation (Quadro) nVidia video cards of recent vintage support CUDA. (ATI cards do not, but have their own implementation of a CUDA-like programming language & SDK) In some cases, there is very little hardware difference (except price) between an nVidia consumer card and the comparable Quadro. For example your can start with a GTX 275 and load the BIOS from a Quadro FX 4800 and you will have a perfectly good FX 4800, for a lot less money:

http://aquamac.proboards.com/index.cgi?board=graphics1&action=display&thread=1043

Finally, it is unknown, whether SCS has built in some limitations to the Sony AVC codec that prevent certain nVidia cards from using the CUDA features of the codec. That may be the difference between the custom rendering setting Automatic vs. Use GPU. P Pro CS5 ships with a configuration file of all the "supported" video cards. You can add an entry for your card, if missing, but don't expect to be able to call up Adobe Support and complain about a bug of any kind. The Sony AVC codec has a DLL that appears to check the GPU's capabilities. What it does exactly is known only to SCS.

As such, the latter two CUDA-based NLEs cannot be directly compared to the VP 10 CUDA implementation. VP 10 can be fairly compared to Badaboom and there does seem to be room for improvement in the SONY AVC codec.

I apologize for this long "rant", but I think that SCS is being criticized a bit prematurely about the whole CUDA thing.
Xander wrote on 10/12/2010, 4:15 PM
I took a 30s HDV clip and rendered it using the default Sony AVCHD 1920X1080 template - 1920X1080@29.97 project. No effects. All that had to be done was decode, scale and encode. I have a Quadro FX 4800. My results are:
Vegas 9 64-bit (CPU): 1:41 (~33% CPU)
Vegas 10 64-bit (CPU): 1:50 (~21% CPU)
Vegas 10 64-bit (GPU): 1:31 (~20% CPU)
LSHorwitz wrote on 10/12/2010, 7:57 PM
I provided a short list of some of the CUDA-enabled programs I am using in my prior post not to try to suggest that they replace Vegas 10, but rather to illustrate that properly implemented CUDA acceleration can and does permit dramatic improvements, vastly better than the meager 5 to 6% that several of us have noted with the new Vegas 10.

As an AVCHD user since its inception several years ago, and an HDV user since its inception several years prior to AVCHD, I have always struggled with the inability of desktop computers to adequately handle these newer high definition formats. Much to the credit of many of the software vendors, NLE and video compression / encoding software has been cleverly optimized / hand coded for the ability to get very fast and tight execution, and have leveraged GPUs, SpursEngines, multicores, etc., and have implemented strategies to only re-compress / re-render when neccesesary.

As a result I have some programs which can cut through AVCHD like a hot knife through butter.

I merely observe, based on the comment made by the original poster of this thread as well as many others, that Sony, even in this very latest version of Vegas 10 Pro, refuses to take full advantage or even any significant advantage of the GPU and Smart Rendering, and is thus an exceptionally slow program when it comes to rendering. I just think it is entirely unfortunate that many other programs have taken advantage of these techniques for several years and several versions, and I was hopeful that this version of Vegas would finally make the leap.

This is not intended to bash Sony, but rather to point out that the limitations being discussed are not technical, but rather reflect management and labor cost decisions which, in my view, have compromised a very basic function of an otherwise excellent product.

Larry
Jeff9329 wrote on 10/13/2010, 6:31 AM
Why is V10 showing slower renders than earlier versions when CPU rendering?

Is it a different (updated maybe) codec being used or just a change in Vegas architecture slowing down V10?