Need a new video card for Premiere CS6

[AndrewThomas]

Hey guys, I'm looking to upgrade my video card and was wondering if a GTX 670 will work with the "hack" method of adding non-official nVidia cards. Is it better to just spend the extra $100 and get a 680 which is officially supported?

[AndrewThomas]

This is making me so sad :p

[Paul Watt]

I've read that 5xx cards are faster than 6xx when cuda rendering. Something to do with nvidia crippling the 6 series to protect quadro. GTX 580 prob your best bet.
I've hacked in a few different cards, and they've worked very well. you have to re-insert the graphic card model name after updates usually.

here's where I read it:
http://www.tomshardware.com/reviews/...k,3161-15.html

[Jeff Kilgroe]

Has nothing to do with nVidia intentionally crippling anything and actually the 6xx cards are similar in speed, even a bit faster, for single-precision calculations, which Premiere seems to be. The issue is the new GPU design with the Kepler architecture.

The Quadro cards still have an edge in some regards, or I should say the Quadro 6000 and its Tesla C2075 counterpart (same card less the video ports) have an advantage. And it's mostly due to the extra onboard RAM.

The big question I would ask yourself first is about budget and how long you want this card (and system you're installing it in) to continue working for you. You really can't go wrong with a GTX580 card with 3GB. The EVGA CLASSIFIED ULTRA model is amazing, but just be warned it doesn't physically fit into some systems/cases, it's oversized. A GTX 680 card is not a bad option either, but the drivers still need to mature a bit more and due to the newer and different architecture and first real model release on the new architecture, all those little factors, it's not a whole lot faster. For some tasks, it is indeed slower. But it does accelerate the MPE quite well. If you are running Resolve or AfterEffects CS6, you should consider a GTX690 as it will use both onboard GPUs -- it's the equivalent of two GTX680s in SLI on a single card.

nVidia has done a pretty good job of not crippling CUDA performance on the GeForce line compared to Quadro. OpenGL differs between the two lines and it's not so much a crippling, but a difference in optimizations for intended purpose. GeForce cards are optimized for gaming and performance visuals where the Quadros are optimized for precision and stable performance.

[Nick Pasquariello]

Kilgroe, I've seen a lot of talk (been doing a bit of reading all over the web for the past week or so) about how the Quadros are slower, but more precise and less prone to errors.

What does this translate to in terms of real-world use? Are we talking bad frames? Failed renders? Artifacts? Are these issues centered around Maya/Mudbox/3DSMax/Nuke use? Or do the issues with the non-Quadros affect Premiere and other programs?

[David Granger]

I just picked up the Galaxy GTX 680 GC Edition. This one came with 4GB of video memory. I added it to the hack list for CS6 and it seems to be a fantastic solution.

[Jaakko Rinne]

Can CS6 use one card for display and another for processing?

[Jeff Kilgroe]

Kilgroe, I've seen a lot of talk (been doing a bit of reading all over the web for the past week or so) about how the Quadros are slower, but more precise and less prone to errors.

What does this translate to in terms of real-world use? Are we talking bad frames? Failed renders? Artifacts? Are these issues centered around Maya/Mudbox/3DSMax/Nuke use? Or do the issues with the non-Quadros affect Premiere and other programs?

Quadro cards have increased precision for OpenGL and real-time render operations. It has no direct bearing on CUDA calculation performance -- it affects precision visualization apps like Maya, Pro/E, SolidWorks, etc.. You get smoother and more accurate edges and surface unions on your display and for simulation purposes within OpenGL.

When it comes to CUDA acceleration or general visual performance, the Quadro cards are more consistent in performance and tend to run smoother, but at a slower pace. With a Quadro you don't gain any more reliability or accuracy for CUDA acceleration. They are clocked a bit slower and have larger quantities of RAM with ECC capability. With the exception of the more sensible clock speeds and power consumption, the only real difference between current generation Quadro cards and their GeForce counterparts is the amount of RAM and, in some cases, the video connectors offered on the cards. Most of the real differences occur within the drivers.

[Nick Pasquariello]

So, if one was mostly editing R3D files in Premiere, and not doing much or any Maya/Solidworks/etc., go for a GTX card, even though it runs a little hotter and uses a little more power? It would be about half the cost for a equivalent or even a few more CUDA cores, and it doesn't sound like it would handicap anything (as long as display options are adequate for the intended use).