View Full Version : PNY Geforce GTX 660 - Will it Improve Performance?


Kenneth Fisher
February 3rd, 2013, 07:36 PM
I recently bought the PNY Geforce GTX 660 video card, and I think the answer is no it wont accelerate Vegas, or not yet.

However, I am not sure. I have checked both the Sony AVC and Mainconcept AVC rendering templates, and they both detect GPU acceleration.

I tried rendering the same 1:31 length clip 4 times, and here are my results:

1) Mainconcept AVC with GPU acceleration: 111 seconds
2) Mainconcept AVC without GPU acceleration: 104 seconds
3) Sony AVC with GPU acceleration: 72 seconds
4) sony AVC without GPU acceleration: 72 seconds

The project and all of the media is on one dedicated drive, and I render to a separate dedicated drive.

***

Vegas Video Preview Performance:

Either I am doing something wrong or my card doesn't work to accelerate that either. I read about guys running stuff on "Best, Full" realtime but I don't get that performance. When I go to auto either on preview or best I have a much better playback experience with .h264. But I feel it should be better.

***

To fill you in on my system specs:

Sandy bridge Core 17 2600k
16 GB DDR3 Memory
PNY Geforce GTX 660 - 940 CUDA cores, 2 GB ram
6 Western Digital Caviar Black 2TB 7200 RPM drives

***

Premiere Performance is Noticeably Better with GPU on than off:

The performance boost in premiere is very noticeable. The test Clip length is 1:43:06

1) GPU acceleration off 82 seconds
2) GPU acceleration on 58 seconds

Except I don't use Premiere often!

***

Conclusion: GPU acceleration in Vegas 12 with the PNY Geforce GTX 660 video card doesn't work, or the system specs are good enough that the card cannot boost its performance.

Any comments?

Juris Lielpeteris
February 3rd, 2013, 08:32 PM
Is the GPU acceleration gives a real benefit or not, it depends not only on the video card, but the also on processor power.
P.S. if you have not changed anything in Premiere rendering settings, then by default the quality is lower, and the so speed is higher than using Vegas Pro.

Jay Allen
February 4th, 2013, 08:25 AM
ver 12 has a long way to go to be fixed, even though ver 11 had problems, ver 12 is worse.
I can run a clean full preview in cs6, but in vegas 12 forget it...not even close.

I render faster and have better preview rates in ver 11
the main problem is Sony...even when the masses have the same problems....Sonys pad answer is to re- install vegas and windows.

Jeff Harper
February 4th, 2013, 10:50 AM
GPU acceleration with Vegas has seemed like a bad joke to me since it began. I prefer to put my money in the most expensive CPU I can afford, overclock it, and be done with it. Just my 2 Cents.

Some folks do get some great result with it, but the faster your CPU the less effective the GPU will be. The whole things seems to work best with people with slower CPUs, at least that's my take on it.

It's too bad Vegas cannot self detect whether it's better to run the acceleration feature or not, that would be most helpful.

Gene Gajewski
February 8th, 2013, 06:49 PM
The Nvidia 600 series has a different architecture from past designs which affects its usability with OpenCL. Nvidia made a choice with its consumer boards to focus more on gaming rather than general purpose GPU computing in an effort to save money and possibly drive GPU computing users towards its Tesla products.

There's performance reviews on Tom's Hardware describing the issue in detail, lot's of discussion in the OpenCL community (this doesn't affect just Vegas), and discussion of the architecture changes and reasoning.

The latest Radeons run OpenCL orders of magnitude faster than Nvidia's latest. The HD7970 is a reportedly screamer.

Danny Fye
February 9th, 2013, 05:16 AM
"The Nvidia 600 series has a different architecture from past designs which affects its usability with OpenCL. Nvidia made a choice with its consumer boards to focus more on gaming rather than general purpose GPU computing in an effort to save money and possibly drive GPU computing users towards its Tesla products.

There's performance reviews on Tom's Hardware describing the issue in detail, lot's of discussion in the OpenCL community (this doesn't affect just Vegas), and discussion of the architecture changes and reasoning."

This is why I will not upgrade from my 560 Ti to the 600 series. So when the 560 Ti finally dies or if I sell it my GPU days will be done.

Not worth the driver hit and miss hassles.

Hmmm, I might look at the ATI cards now that they can finally scream! LOL!

Danny Fye
www.dannyfye.com - Videos - CCM. (http://www.dannyfye.com/ccm)

Mike Kujbida
February 9th, 2013, 08:06 AM
From user Matt Carlson here's a great explanation as to why the 600 series cards currently won't work with Vegas.

The new Nvidia Keppler cards (i.e. the 600 series) changed their architecture and divided the CUDA cores in to four banks. This change is rumored to be intentional in order to limit performance to specific SDK licensing. The effect of the architecture change for Vegas is this... only the first CUDA bank is recognized. Your 570 had 480 cores. Your 660ti has 1344 cores but Vegas only sees 336 of them which makes it a step down. Once again the rumor is that this was intentional meaning that we will not see a driver update that fixes this as it is not a "problem."

Gene Gajewski
February 9th, 2013, 06:06 PM
I think it might be a little more nuanced than that. My takeaway was that although nVidia increased the number of stream processors, they did nothing with the local memory available to those processors as a whole, thus shrinking the available local memory available to each processor. This causes, of course, an increase in memory contention, thus a delay in memory access. The article went on to mention that this isn't nominally an issue for gaming graphics which usually have tightly bound memory and/or literals in their algorithms, but is a hindrance for *general* purpose GPU computing where variables are specified that aren't necessarily, and usually aren't closely bound. (Think stacks - variables originating from a parent calling function).

Now, you can hack *some* of this in the driver to rearrange things to more optimally match the architecture limitations (ala CUDA), but then you're doing things you didn't have to before, and making portability more difficult.

nVidia simply weighed the costs savings against the hubba from the OpenCL community and said, 'Screw 'em, well take the money'. That's about it.