You do not even mention your CPU here, and it is important to have a "balanced" system for Premiere Pro since various components are taxed simultaneously and for some workflows your net speed will be determined by the weakest link. A fast 6-core system may be well matched to a GTX 770, whereas an older quad core might be just about as fast with a lesser video card.
Now for some facts:
- No Adobe Premiere Pro version prior to CC supported more than one GPU card [sidebar: there were some special Tesla card configurations supported but those are not really GPU cards]
- You are correct, having lots of GPU firepower past what you need to keep up with your CPU for timeline work is pretty much wasted EXCEPT FOR renders to DVD format which are massively helped by dual GPU cards (for CC users anyway; other versions of Premiere cannot utilize past a single GPU card)
- SLI does not come into play; CC can use more than one card whether or not SLI is enabled. Too, some earlier versions of Premiere Pro (I don't remember exactly what releases) had issues when SLI was enabled
Jim and esr.
I would like to dispel a GPU acceleration myth here and now!
On our 7-layer H.264 timeline with lots of different formats including a few 4K clips but mostly 1920 x 1080 clips and it is a 1920 x 1080 timeline, when you export it with GPU hardware acceleration with CC 2014 (which has new GPU accelerated features) to 1920 x 1080 output on my main editing computer which had a GTX 770 SC, the time to do the export was 101 seconds. When I turned off GPU acceleration and checked "Use Maximum Render Quality" (as required to equal the GPU accelerated quality) the time to export was 908 seconds. Nine times faster!
So therefore the myth of GPU hardware acceleration is only effective for producing DVD's is debunked. I would like to see some others try this for yourself.
In our PPBM6/7/8 benchmark (Harm has yet to update that info) but with that one benchmark located on the ppbm7.com webpages you now can run with either CS6, CC (version 7) or CC 2014 (version 8) just by downloading the one download. In it are three project files and three statistics gathering files. All are identical timelines but the embedded instructions have to be different as the GUI of Premiere has changed with each major release.
Only if you export a lot with resizing, frame rate changes or blurring going on, is a good video card with hardware acceleration helpful to reduce encoding times. Otherwise you will profit more from extra cores, higher clock speeds and bigger L3 cache. Dual video cards are generally a waste of money for Category B users, since they are idle most of the time anyway. If the majority of your work does not use resizing and usually consists of exporting HD to full HD 1920 x 1080, there is no sense to get a top-notch video card, since it will hardly be used, if at all. The CPU carries the burden of the encoding.
I think you need to read both what esr (Original Poster) and I have put into writing in this thread more carefully.
Neither of us ever mentioned anything about not using a decent GPU which seemed to be the focus of your above rather condensending mythbusting post. Our actual words included, "multi gpus help with exporting to DVD", "it seems kinda pointless to use 2 or more gpus unless I am spitting out thousands of DVDs", and "having lots of GPU firepower past what you need to keep up with your CPU for timeline work is pretty much wasted EXCEPT FOR renders to DVD".
It could be interesting and related to this thread if you were to rerun the H.264 export test with two GPUs and compare the results to what you got with one GPU. Based on the many tests that I have run using PPBM7 in the past, I would not expect to see a huge difference on the H.264 export, but who knows (I've done zero PPBM runs since CC 2014 came out).
Looks like you are confusing some technologies there.
GPU affinity is simply locking a GPU process to a specific GPU. That is not the same as scaling. GPU acceleration and SLI are completely different technologies. SLI has to do with drawing frame data on a screen using more than 1 GPU. GPU acceleration is using the GPU processing for compute tasks and processing frame data in the case of media content applications. The GPU acceleration can be used to handle many aspects of the frame and does so with Adobe. However it does not currently handle decoding or encoding of the codec ie compression data for the frame. CC 2014 added the GPU accelerated debayering which is decoding some elements of those codecs but the codec algorithms are still handled by the CPU's.
GPU acceleration across multiple GPU's is application MPI and or driver scheduling. GPU affinity has nothing to do with the capability of scaling GPU acceleration data across more than 1 GPU. That just locks process/threads to 1 GPU or the other. Without affinity scaling will still occur. The reason for the affinity was to prevent data from having to copy from 1 GPU to the other before processing. However GPU direct has resolved that so it really has little purpose with the current Cuda version.
Currently CC 2014 has poor GPU acceleration performance when scaling across more than 1 GPU. The performance is actually lower than just using 1 GPU as the acceleration device. Until that is resolved I would not set both GPU's to be used with the GPU acceleration with Premiere. BTW Speedgrade still uses CPU processing. All decoding/encoding has to be done on the CPU along with all of the memory management. Ray Tracer is basically phasing out since the performance is often slower for many aspects. The linking to C4D is the replacement to the features Ray Tracer was meant for.
Eric from adk,
That was the exact info i was looking for. The rest of you are helpful too, dont worry.
Since i recently looked up gpu affinity, i confused myself as how adobe could take advantage of multiple gpus with out sli or some way to speak to them. your explanation of scaling and how things work is great.
please, take a moment to educate me, i beg you.
What advantages does multiple gpus offer in these specific programs, AE,AI,FL,PR,PS.
i know your company and i know your a busy fellow, you can answer like this,
AE, faster 3d preview.
PS, faster export.
those are not meant to be correct, just and example of an answer so you dont have to waste time. of course you are welcome to elaborate, as i will learn more from your knowledge.
please reference the CS6 CC14 universes as i am not yet upgraded, i know the support in cs6 is almost nil but i keep hearing things like AE will use all cuda on all gpus you throw at it. however, as you know the web is rife with wrong information, please set me straight.
if you could sir. i would be grateful.
AE - Considerably faster if Ray Tracer is used and supported. However only very specific elements http://tv.adobe.com/watch/learn-after-effects-cs6/ray-traced-extruded-text-and-shapes-part -1/
PP - Not supported. However one can output display with 1 GPU and GPU accelerate with another so performance can be gained if lower end cards.
PS - None
AME - None
Speedgrade - Same as PP.