swaaye wrote:I'm still waiting for the 3.5GB issue to really surface in a game. Yeah you can test for it with CUDA or whatever, but in games the card has always been a great value for its performance level. Even Tech Report's fancy frame time charts always look good.
Indeed, I bought my GTX970 after the 3.5 GB issue was already known. I just figured that none of the reviews showed anything weird, so the card apparently performs fine, regardless of how it does it.
I bet a lot of people did. I don't think they should be entitled to the $30 really. But it's 'Murica. I guess the point isn't so much that the customers are entitled to the money, but rather that vendors should be careful not to spread misleading information on their products.
swaaye wrote:I think AFR-based multi-GPU is more of a scam. Because of frame latency canceling out the benefit. But apparently the super coolness of dual cards overrules any logic there.
Yea, I've always wondered about that. It shouldn't be THAT difficult to get rid of most of the micro-stuttering, and get a smooth gaming experience.
I mean, you can measure the average framerate, and you can measure how long it takes to emit draw calls for GPU #1 before moving to GPU #2. Since there generally is not that much of a difference in subsequent frames, the values measured from GPU #1 should be very close to what GPU #2 will need to do for its frame.
From that data, you can easily extrapolate a delay-value so that you offset GPU #2 to fall exactly 'halfway' the frames of GPU #1.
The only reason I can think of why they aren't doing this is because although it will give a smoother user experience, the delays would give a minimal drop in average framerate when benchmarked.