I realize there are already a couple of threads running in regards to GPU support. However they are long and
contain a lot of silly argueing and debates of which $300 nvidia cards adobe should be supporting and that is not
We as potential CS5 users desperately need adobe to give us relative performance benchmarks so we understand what
the various cuda cards are going to give us for our money.
Here is a nice comparison chart of the Quadro FX cards with details of their memory and cuda cores. Cores have a
range with 8,16,32,64,112,128,192,240. that is a huge difference. I have not done much research on the cuda
systems, however one major question is are all cores created equal? are the cores on a slower gpu as fast as a
core on the faster gpu? Also a major question is if CS5 supports SLI and scales with 100% efficiency? If i run 2
FX3700 - 112 core cards in SLI mode for $1200 is that the same speed as a single FX5800 - 240 cores for $3000+ ?
What about the new low-end video game cards? the newest one claims it has 480 cuda cores? again, are those cores
the same speed as a quadro core and will they function with CS5?
My company owns well over 100 licensed copies of CS4 master collection, all of which will be upgraded to CS5. I
have 10 systems here in my office all with 8 cores, 20GB ram, and Matrox Axio LE cards that we spent over
$100,000 on. I really need to know what the best upgrade choice is going to be for us to outperform our axio
realtime HD systems. If it turns out that the ideal setup is dual FX5800 SLI cards in every machine then we would
need to spend well over $50,000 in gfx cards upgrades to optimize CS5. But obviously we need to know what we are
getting for our money!
I feel that adobe should be -OBLIGATED- to it's customers to release in-house benchmark results of those cards
they have certified for use, not just simply claim "they work".
Adobe would surely have a good idea of how to do this, but off the top of my head i would simply take a decent 4
or 8 core machine, make a project with a green screen 1080i/p video and put 20 or so copied layers of it. Have it
do an ultrakey and maybe a color correction effect. Start with your slowest card and simply play it and then
enable the next track above, keep enabling tracks and playing it until it chokes on the realtime playback. Then
do a timed render of something that is CUDA accelerated. Then plug in a 2nd card in SLI mode and repeat those
tests. Then do that with each certified card up to the FX5800. You can post the results tied in with that chart i linked
to. I'm sure one smart employee could do this test in 1 or 2 days. (or send me the cards and CS5 and i will
gladly do them for you)
We really need this kind of performance and scaling comparison so all adobe customers can make a well informed
decision on which nvidia card is in their budget and will give them the performance desired. I realize this is a
huge cash cow for Nvidia and they stand to make a fortune from CS5 upgrades, they should be highly motivated to
do this themselves. But if it works as well as you claim then it's well worth it in my books.
Thank you, and any Adobe employees please forward this to anyone who could help make this happen.