GPU Section Criteria

gnu_B wrote on 3/21/2013, 11:33 AM
I have GPU acceleration working on my MS12 and it's working to my satisfaction. I'm looking for another GPU card that might give me even better performance, at a price I can afford. I know where Sony's recommended card list is, but I realize that they can't test all cards, so I'm searching within their criterion, but outside their list of cards.

I have searched this site: http://www.videocardbenchmark.net/

...and have noticed that they list cards under video performance but there is also a listing for 'directcompute' performance, which would seem to be a closer measure of what I am seeking. In some places the two scores seem to match, in others they don't.

So I'm trying to determine the best performance criterion upon which to search.

Does anyone have any ideas or information on this?

-gnu_B

Comments

NormanPCN wrote on 3/22/2013, 4:03 PM
Good instincts on DirectCompute. Your best benchmarks to compare to Vegas GPU use would be DirectCompute and OpenCL. Vegas uses OpenCL internally, but it's algorithms are certainly different that those used in benchmarks you are likely to see online. At least these two both leave out the game display rendering which is not applicable to Vegas use of GPU.
Remyx wrote on 4/1/2013, 3:34 PM
It is actually very difficult to estimate GPU performance for a given card in the context of video encoding. There are some many variables, including internal GPU architecture, drivers and software implementation and so on.

Few months ago I spent some time benchmarking video rendering with several different hardware configurations. I used 3 different systems, a Core2 Duo 3.0GHz (2 cores), an i5 2.5GHz laptop (2 cores) and a dual socket Xeon 2.8GHz workstation (8 cores). All systems running Windows 7 64 bits. On the Xeon workstation, I tried 2 different GPU cards. On each systems, I rendered the same clip using CPU only, then GPU.

Interesting to note the following observations:
1. Using the same HW config, upgrading from VMS 11 to 12 lead to some noticeable improvements.
2. On the latest Intel chipset, the Intel QuickSync provided very good performance, however the output quality was significantly lower than using CPU only or real GPU. Numbers are provided for information only, but quality was so low than this is not really an option.
3. The 2 GPU cards used on the Xeon workstation are from the same area, similar price point and score very close in game benchmarks. They also scored very close with VMS 11. Didn't get a chance to try the Nvidia 470 with VMS 12 unfortunately.
4. The 8 core Xeon system did not do very well using CPU only. With 4 times the number of cores as the other systems, it only performed about twice as fast.

I'd like to try more recent GPUs, but don't want to spend >$300 on a new card to only get a small performance increase. When everything has been said, it still looks like rendering is a batch processing job :-)

Appreciate any feedback or additional data about other real world performance benchmark using VMS.

System CPU GPU SW Version Render settings Fps
Lenovo D20 Dual Xeon X5560 2.8GHz 8 cores AMD 5850 Studio 12 CPU only 14.2
Lenovo D20 Dual Xeon X5560 2.8GHz 8 cores AMD 5850 Studio 12 GPU if avail 20.3
Lenovo D20 Dual Xeon X5560 2.8GHz 8 cores AMD 5850 Studio 11 CPU only 12.3
Lenovo D20 Dual Xeon X5560 2.8GHz 8 cores AMD 5850 Studio 11 GPU if avail 17.0
Lenovo D20 Wks Dual Xeon X5560 2.8GHz 8 cores Nvidia 470 Studio 11 CPU only 11.8
Lenovo D20 Wks Dual Xeon X5560 2.8GHz 8 cores Nvidia 470 Studio 11 GPU if avail 17.4
Lenovo Laptop i5 2 cores Nvidia GT630M Studio 12 CPU only 9.1
Lenovo Laptop i5 2 cores Nvidia GT630M Studio 12 GPU if avail 9.9
Lenovo Laptop i5 2 cores Nvidia GT630M Studio 12 QuickSync Fast 20.5
Lenovo Laptop i5 2 cores Nvidia GT630M Studio 12 QuickSync Quality 13.5
Home Desktop Core 2 Duo 3.0Ghz 2 cores Nvidia GF 9800GT Studio 11 CPU only 5.8
Home Desktop Core 2 Duo 3.0Ghz 2 cores Nvidia GF 9800GT Studio 11 GPU if avail 7.5


Remy