This card will NOT use 192bit VRAM. The type of VRAM and the bus used are dependent partially on the GPU. Let's also get with the program on modern memory: 192 is old and probably harder to obtain that higher bit VRAM, so this makes it even less likely.I have a feeling they will use either 384bit (as found on the 780Ti) or finally move to 512bit.The memory speed is probably more of a consideration, though bus width certainly matters under some circumstances.
With the same production node they can not make this much quicker without making this chip bigger. 5-10% top more speed than 650. (if the size remains the same.) As someone said, this is a test run, and "real" 800 series comes later with smaller production node. What this allso means, is that 20nm production node is not quite in there just yeat. Interesting to see how long it will take untill it will be out and running in highend GPU production.
28nm refers to the size of the transistor, not the size of the chip. A 28nm GPU or CPU could feasibly be the size of a football field, or it could be a centimeter.There's nothing about the 28nm process that inherently prevents a new chip with a new design from being more powerful or in any way restricts the size of the GPU.
its a RUMOR and the big "news" of this rumor is shared GPU and CPU memory. which with nvidia must mean system memory. um.... AMD already did HSA and actually makes CPUs/GPUs...and their first card would be a midrange card no one with current gen would go for? sounds really week to me
Nvidia already did it, too, and AMD hasn't even launched their version. CUDA supports UVA which allows the drivers to be configured to have system memory assigned to them, however it hasn't been optimized w/in the current Nvidia lineup and we haven't really seen what it can do yet. AMD's HSA is also targeted at being used in conjunction with one of their "APU" CPUs but not limited to this.I think the first implementation we'll see of HSA will probably be in scenarios where performance can hurt due to physical limitations on the system, like a laptop with integrated GPU (like Radeon HD graphics) where moving the GPU off-die would allow for better performance due to heat and equivalent performance or better could be achieved by the shared memory architecture.
I would have thought that their lack of keeping up with the direction in which the industry is heading would turn most of its fans off. I mean proprietary closed ended solutions are just so 2004.Oh well... I guess they'll just be as surprised as the 3Dfx fans were when that company went belly up.
Facebook
Twitter
Instagram