V12 Render speeds - CUDA is fast

dxdy wrote on 10/1/2012, 9:16 AM
I have a i7-950 with 8 Gig RAM and a PNY 560ti with factory overclock . Downloaded 12 today, and rendered some tests using 1080p footage from a Panasonic 3 chip camera. Rendering to MC mp4 Internet widescreen template, there is a newer version of hardware selection option at the bottom (used to be encode mode was auto, CPU only, GPU if available). The new version allows choosing CPU only, OpenCL if available, or CUDA if available.

With a layer of legacy text and a layer of 1080p footage, my tests turned out like this:

CPU only - 95% CPU, GPU in the teens, 39 seconds.
With Open CL identical to CPU only.
With CUDA, 70% CPU, 78% GPU, 12 seconds

The CUDA rendered output was a couple hundred bytes bigger than the CPU only. CPU and OpenCL were the same size.

Now I just have to figure out how to get Boris BCC 7 to run with SVP 12.

Comments

megabit wrote on 10/1/2012, 10:26 AM
A very similar picture to what I'm having with my System #1 (a little slower GPU in my System #2). With both 11 and 12.

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)

dxdy wrote on 10/1/2012, 10:56 AM
And another test -

1 track of Sony chroma key above a single track of AVCHD 1080p, a 23 second clip - not only is the preview solid at Best Half, but good CUDA render results:

CPU only: CPU @ 95%, GPU @ 17%, 72 seconds (3 times real time)

CUDA: CPU @ 70%, GPU @ 70%, 31 seconds (1.3 times real time)

While the CUDA time is the same as 11 was, 11 wouldn't preview at full speed at Best Half, I had to reduce it to Best quarter, and then it crashed during preview, the first time that has happened to me.

Edit: BTW Nvidia driver is 296.10