I'd like to get the opinion of other people than myself on this. I've been arguing on another forum that for +100$ the 640MB version of the 8800GTS was not worth it except for high-resolution gaming, which is what most benchmarks have revealed. But I am told Crysis and UT3 and etc. are likely to require a lot more video RAM even at moderate resolutions, and it would be foolish to not plan accordingly, so the 8800GTS 640MB would be the better option.
I've a bit of difficulty with arguments based on future and hypothetical results, but they are difficult to disprove either. So, what do you think about this issue?
The thought behind the argument is based on DX-10 shifting a large amount of the rendering duties off the CPU and onto the GPU and vid card. Wen you go to DX10 you will probably see a slowdown but if you stick with XP you have no worries. Someone just did a test on this. I will try and find it and post a link.
I will most likely be getting the 640MB. Ultimately it comes down to what that $100 means to you. In my case, I don't mind spending it as longs as I get value for my money.
I have a 6800GT now, and since I skip every second generation of card I don't want to be "stuck" with a 320MB card which 1 1/2 years from now isn't doing what a 640MB could do and force me to upgrade sooner than expected.
If you upgrade video cards more often than I do, then yeah get the 320MB becuase by the time you'd notice the difference, you can spend the $100 on your new 9800 card!
Rosu's 320MB won't ever be useless. There just may come a time when 320 vs 640 is a bigger difference than the benches are showing now (which is virtually none).
I skipped the 7 series generation as well but had two 6800GTs I figure, or hope, that this 8800GTS will be good through the next gen and when it gets a little bogged down will get a second used one and SLI it.
And we won't know the effects until the apps ship.
There's efficient ways to do things, and inefficient ways. MegaTextures would be an issue, but efficiency could achieve similar or better with less.
Also there's consideration for things like physics, but IMO that's an optional level of overhead.
We really won't know until we see the first truely exploited DX10 games.
I have a feeling that the sweet spots for the G80GTS series will be just below the situations that are the most demanding anyways. In which case does it make more sense to take the $100, set it in a medium yield term desosit, and then sell the GTS, and buy it's replacement when you know what the effect is. That would be my recommendation for anyone with any bit of hesitation. Save the money, even under your pillow, and then upgrade later if you need to, and likely for less.