Was just looking at the specs for the new Nvidia GTX 680, the 580 and the 470 which I own currently.
GTX 680 - 1536
GTX 580 - 512
GTX 470 - 448
First off, - wow ! ... the 680 specs are amazing.
Second thought - I was really thinking of jumping to the GTX 580 instead of the 680 to save money ( 580 pricing will be dropping dramatically in the coming months )
However.... after seeing the CUDA cores with my own eyes, it seems like a less than brilliant choice to even put money into the 580 at this point.
The 470 could do 448 cuda cores, and the 580 only jumps to 512 ... hardly a jump at all. Going from 448 CUDA cores to 1536 is just well, mind popping.
What do you all say ? thoughts ?
Thanks. I am awaiting to see the results. (Though based on another test, I do not expect its performance to be anywhere near that much faster than the GTX 580 given its immense CUDA core count.) The worst part about this is that it's an upper-mid-range card at a high-end price (as far as consumer gaming cards are concerned).
Been doing a bit of a search looking for an answer regarding recycling my GTX 470 and came across this thread. I've just bought a GTX 680 and now wondering if it's worth using the 470 for some extra CUDA cores. However, one of the reasons that prompted the ugrade was the need for Compute 2.1 in certain apps. The 470 being a first gen Fermi is 2.0. So does anyone know if the cores will still be usable provided the 680 is the primary card? Thanks in advance.
And sorry about the hijack of sorts so to give some help to the OP, I can say that I am very happy with the upgrade decision. I think it's early days yet to see just how well (or badly) the Kepler range do in compute performance but it is definately a step up from the 470. So far it's been performing extremely well for me in PS and molecular simulation. The latter being a blend of compute and graphics makes this a reasonable choice in my case. It's also one of the quietest and power efficient cards I've ever used, in stark contrast to the 470. Then again you could save some money with the prices of the 580 crashing right now. I think either card would serve you well enough.
Premiere Pro does not use the second card for GPU acceleration at all. MPE supports only one card. Period. And if a second card is required to drive more than four monitors (as the GTX 680 can do), that second card should be of the same generation series (this means that a 600 series card should be paired with another 600-series card, while a 400-series card should be paired with another 400-series card).
Yes, I am aware that when I'm running PS (or Premiere as you say above) that only one GPU is utilised. I think my post above was probably unclear on that point. I use the card for non-Adobe CUDA applications also. Thanks for the info regarding other uses. It's not a major issue. The 680 is more than fast enough but figured if I could put the 470 to some use it would be a bonus.
Be aware that although the new 680 GTX has 3x as many cuda cores, its only 2x as far as performance is concerned.
They're not the same type of cuda core from the last generation, fermi. They got rid of the shader clock which was 1.4ghz+ in the past, and now the cuda cores run at the 'core' clock, which is significantly less. Because they dropped the clock, they tripled the cuda cores and in turn got around 2x the performance.
My eyes lit up when I saw the first specs of 1500 cuda cores but then I knew something was up. They only dropped one node, how could they triple the cores?
Anandtech has a great article explaining it. Keplar's (680) cuda cores aren't apple to apple comparisons to Fermi's (580). No shader clock, smaller memory bus but 3x the amount...cant wait for the real kepler big card, not the dual-gpu one either..