-
CUDA Test Comparisons in CS6 and Adobe Media Encoder
Hey folks,
I was having trouble with my CPU maxing out at 100% and being unable to do anything else and one of things I thought might help was upgrading my video card. There’s a lot of misinformation regarding whether Adobe Media Encoder takes advantage of the GPU acceleration provided by the Mercury Playback Engine when rendering, so I decided to do my own before and after tests.
First off, here’s my current setup:
Windows 7 64-bit, i7-3770K 3.5ghz Quad-Core, 16gb of RAM, SSD C drive, RAID data drive.
I went from the GeForce 210 to the GeForce GTX 750 Ti. Here’s the card comparison:
GeForce 210
RAM: 1gb 64-bit DDR3
Core Clock: 520mhz
CUDA Cores: 16GeForce GTX 750 Ti
RAM: 2gb 128-bit DDR5
Core Clock: 1072mhz
CUDA Cores: 640In order to enable CUDA in either one I needed to modify the acceptable list of video cards, instructions for which you can find here:
https://www.studio1productions.com/Articles/AfterEffects.htm
FWIW, according to the above the bit interface (64 vs. 128) might be more important than trying to win the CUDA cores battle, so take that for what it’s worth.
On to the tests…
I did 3 different tests. The first two were static 1080i/29.97 interviews with some intro/outro graphics + Levels/Contrast/Saturation and Red Giant’s Misfire Vignette. There’s also a logo graphic in the corner of the entire clip.
————————————————————
Render Test 1 (Static Interview):Length: 2:20
Attrib: 1080i/29.97
Filters: L/C/S/V + Graphics
GeForce 210 Encoding Time: 15:32
GeForce GTX 750 Ti Encoding Time: 12:44 — approx. 40% CPU usage
Control 750 w/ MPE set to Software: 15:38 — approx. 65% CPU Usage
————————————————————————————————————————
Render Test 2 (Static Interview):
Length: 3:06
Attrib: 1080i/29.97
Filters: L/C/S/V + Graphics
GeForce 210 Encoding Time: 20:52
GeForce GTX 750 Ti Encoding Time: 17:05
Control 750 w/ MPE set to Software: Skipped
————————————————————Render time seemed to be about 20% faster with the new card and CUDA enabled, and processor usage dropped from 65% to 45%. Disabling CUDA and setting the Mercury Playback Engine to Software Only showed roughly the exact same time as my old card. I didn’t bother running the Control for the second interview since the first was so obviously conclusive.
The other test I ran was on some 720p/29.97 stuff that was the cause of my original peaking at 100% problem. It contains a lot of speeding up/slowing down of the footage, and the second 2/3rds of it have a title graphic in addition to a logo overlayed and contain frame blending. The entire sequence has Levels + Contrast + Misfire Vignette.
————————————————————
Render Test 3 (Action with additional complexity)
Length: 1:48
Attrib: 720p/29.97
Filter: L/C/V + Graphics + Timeshifting/Frame Blending
GeForce 210 Encoding Time: 8:55
GeForce GTX 750 Ti Encoding Time: 5:33 — approx. 45% for 1/3rd, 100% next 2/3rds (gets more complex)
Control 750 w/ MPE set to Software: 8:30 — approx. 90% for 1/3rd, 100% next 2/3rds
————————————————————In this test you can see my time dropped significantly, and my CPU usage for the first third was cut in half from 90 to 45… however the second two thirds are still maxing out my CPUs at 100%. Disabling MPE put it back on par with my old card.
Conclusion: Well, upgrading my card didn’t solve my 100% CPU problem for the speed/frame blend stuff, but it looks like it’s going to improve my overall render time and drop my CPU usage for the majority of the stuff I deal with. Also, it seems pretty clear to me that AME does take advantage of the CUDA cores, at least in regards to the basic filters/graphics stuff I was doing.
Hope you guys find this informative, if a bit long 🙂