10d: GPU rendering...

megabit wrote on 5/7/2011, 1:44 AM
Let's put here our experience with the speed and quality of the Sony AVC encoding with GPU "acceleration".

As to CUDA, at least on my computer nothing has changed, i.e. using GPU is actually slower than CPU alone.

How does it work with supported AMD cards?

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)

Comments

Sunflux wrote on 5/7/2011, 1:57 AM
I'd test it, but according to the release notes it requires driver version 11.2 (February) and won't work with 11.3 (which I'm using) or 11.4, and 11.5 hasn't been released yet. :-/
PLS wrote on 5/7/2011, 2:08 AM
No improvement in rendering time here with Cuda enabled card (GTX 260) on quad core Q6600.
edenilson wrote on 5/7/2011, 2:58 AM
amd 4dcore
8giga de ram
avc render fast..
very good

nvidia xfx 460 evga....1 giga ddr5
rique wrote on 5/7/2011, 4:44 AM
I had the same result in that the CPU only render was faster than the GPU render. Also, the resulting file sizes were different.

Running a GeForce GT 240 with latest drivers (270.61)
and an i7 quad core 920 2.67GHz CPU.

Windows 7 Pro 64bit.
3DFrog wrote on 5/7/2011, 5:05 AM
Turn off Video Preview, and try render again.
im.away wrote on 5/7/2011, 7:04 AM
I think the key to this is in the release notes for 10d.

"GPU-accelerated rendering performance will vary depending on your specific hardware configuration. If you have an older CPU and a newer GPU, rendering using the GPU may improve render times."

Perhaps the advances in CPU technology have made GPU rendering superfluous? I went to the results page for John Cline's 2010 Render Test and it is pretty obvious that each new generation of CPU is outperforming the previous generation, and in many cases, very significantly. Maybe all the CUDA cores in the World can't compete with a Core i7 2600K?

I did couple of tests and here are the results...I was rendering 2 minutes of HDV m2t clip (with some colour correction throughout the event) to 1280 x 720 ACV (MP4). Being in PAL land it was at 50i.

CPU only..............................1:23
GPU only..............................2:02
Auto.......................................2:02

Then I did the same render and immediately that it started, I ran John's Rendertest 2010 in another instance of Vegas. The object of this was to see if the GPU rendering freed up enough CPU resources to be significant.

CPU only.............................3:27 (my render)...............2:19 (rendertest)
GPU only.............................4:00 (my render)...............2:16 (rendertest)

It is pretty clear that my PC's configuration is such that GPU rendering is of little or no benefit. It freed up enough CPU resources to perform a simultaneous second render just three seconds quicker at the expense of taking 33 seconds more to do the primary render.

About the only thing I gained is the knowledge that I could perform at least two simultaneous renders and theyd probably be both finished in the morning when I woke up. With my previous renderbox I wouldn't have contemplated simultaneous renders.

Cheers

Russ
im.away wrote on 5/7/2011, 7:04 AM
deleted by user - duplicated post
3DFrog wrote on 5/7/2011, 7:19 AM
im.away

What GPU card do you have?

im.away wrote on 5/7/2011, 7:27 AM
Hi

NVidia GTX460 (2 GB Mem)

Russ
Jumping Rascal wrote on 5/7/2011, 4:26 PM
I compared the 10c CPU rendering time of a project to rendering the same project using 10d in the auto and and GPU only modes to see how the rendering times compared.

The project source was AVCHD HD 1080-60i (1920x1080, 29.970 fps) rendered with Sony AVC to .m2ts. The project was 5.5 minutes with many color corrections and several effects. My PC has a core i7 930 processor with 12G RAM. The video card is an ATI Radeon 5870 (11.2 Catalyst 8.82 driver) so seems to fit the required GPU specs.

The rendering time comparison result was as follows:
10c.... (CPU) 26.15 minutes
10d.....auto mode ....26.14 minutes
10d...GPU only....25.07 minutes.
The last number is a bit of a surprise although the difference is small. I'll have to update my driver to 11.4 eventually so presumably SCS is working on the 11.3 and 11.4 reported issues. I encountered one of the 10c bugs and was glad to see a fix but overall I was very happy with 10c for use with AVCHD projectcs directly (no transcoding to other formats necessary at all) and what appeared to be excellent real time previewing on complex HD projects. I hope that 10d is as good in terms of overall performance.

PLS wrote on 5/8/2011, 2:37 AM
.... "Perhaps the advances in CPU technology have made GPU rendering superfluous?"

I am (well my daughter actually) is running a system with PPro on a 3.06GHz i7 processor and with GPU enabled the render time is between 3 and 4 times faster then when it is disabled. It would be good if Vegas could at least show some small improvement.
im.away wrote on 5/8/2011, 6:37 AM
Hello,

I find your observations to be quite interesting. You weren't specific, but I'm assuming that when you state that the render time was three to four times faster using GPU rendering, you meant that it was faster than using the CPU in the same program ie, PPro?

Have you tried to render exactly the same footage in Vegas to see how the times compare? Not having PPro, I'm wondering whether it could be that CPU rendering in that program is inferior to Vegas? I'm not trying to be smart-arse, I'm just wondering whether we are comparing apples with apples in this instance?

Cheers

Russ
PLS wrote on 5/8/2011, 7:53 AM
Yes, 3x to 4x faster than using the CPU alone.

And you make a good point, I wonder how PPro would compare to Vegas Pro rendering the same footage. I will do some trials.

Paul.
Randy Brown wrote on 5/8/2011, 8:00 AM
PLS:
PLEASE do and let us know your results.
Thanks,
Randy
PLS wrote on 5/8/2011, 8:59 AM
Well I loaded Vegas Pro onto the computer that has PPro (3.06GHz i7, GTX 260) and did some very quick tests as below:

1) Exporting an AVCHD HD clip:
- Sony AVC (1440x1080, Dolby Stereo, TS): CPU: 1:58, GPU: 1:37
- PPro H.264 (1440x1080, Dolby Stereo, TS): CPU: 0:45, GPU: 0:45

2) Exporting picture-in-picture, same AVCHD HD clip with pip effect:
- Sony AVC (1440x1080, Dolby Stereo, TS): CPU: 2:11, GPU: 1:50
- PPro H.264 (1440x1080, Dolby Stereo, TS): CPU: 2:20, GPU: 0:45

Huge improvement for PPro when the GPU kicks in to render the PIP effect. Although Vegas and PPro take about the same time to render the effect with CPU. Although PPro is quicker with no effect.

I will do some more the tests a little later as the CPU utilization seems rather low for the Vegas AVC tests... only about 30%. That said, rendering the same clip to HD MPEG2 takes 37 seconds and has 80%+ CPU utilization, so perhaps it is just the SONY AVC export format.