what does price point have anything to do with the cards potential.Nope. They're at totally different price points.
Hard to say both a relatively new architectures. I would say both improve at about the same rate.
Agreed. And the gtx 580 can benefit from better drivers, a good example is BFBC2 where Fermi slightly underperformed , the gtx 580 is now returning more performance.lava,
I'm saying if the 6970 were likely to perform at 580 levels or above AMD would have released it at a closer price relative to the 580. AMD knows the limits of what their card is capable of.
how is the 5xx series a ripoff?The 5xx series isn't even close to a new arch. Its nvidia's typical ripoff method. (The 68xx is a negative as well but at least there were pretty decent changes). Is it still possible for nvidia to improve drivers? Absolutely.
But no, drivers won't allow AMD to overtake the crown. There WILL be decent improvements down the road though...
Not a chance.
But the 580 is not really a "new" arch like the 6970 is. The 580 is just a corrected 480.
The 6970 will benefit far more from maturing drivers (read: AMD/ATI going from VLIW 5 arch to now VLIW 4). Hell, I don't think the current drivers were even ready for release as it is. The 6950/6970 are desperate for improved drivers.
Will it catch the 580? Maybe, maybe not. But the 580 doesn't have much more to gain. It's drivers have already been maturing for a year.
The GTX 580 is a 520mm² die card, while the 69x0 cards are a 'mere' 389mm². That the two cards are as close as they are is a testiment to how well the small-die strategy is working.
The exact memory amount is also dictated by the memory bit-interface on the gpu. It comes down to the physical connections with what is standard video memory.
Its why the 384 bit 580 has 1.5 and the 320 bit card has 1.25 .
256 bit interface can have multiples of 512-1024-2048
Nvidia can drop a memory controller possibly, but that would be working backwards from design.
As far as I understand things.
As I understand things, yes.So they're stuck with 1.5gb or could they increase to 3GB?![]()
There are 12 GDDR5 memory chips from Samsung on the PCB. These are the same K4G10325FE-HC04 chips with a capacity of 1 Gb (32 Mb x 32) as used in the Radeon HD 5800 series but organized into a 1536MB bank the GPU has access to via a 384-bit memory bus. It was not easy for the manufacturer to wire such a complex bus. And this must be the reason why the memory chips cannot be clocked even at 1000 (4000) MHz, let alone at their rated frequency of 1250 (2500) MHz. The card’s memory frequency is only 924 (3696) MHz, exactly as specified by Nvidia. The wide bus coupled with the low (for GDDR5 memory) frequency produce a peak bandwidth of 177.4 GBps as compared with 153.6 GBps of the Radeon HD 5870 that uses a 256-bit bus and has a memory frequency of 1200 (4800) MHz. The Radeon HD 5970 has a much higher memory bandwidth – up to a theoretical peak of 256 GBps. The GeForce GTX 480 drops its memory frequency to 67 (270) MHz in power-saving mode.
The 5870 was the same architecture as the 4870 yet they squeezed performance out of it. I think both are going to see gains over time.But the 580 is not really a "new" arch like the 6970 is. The 580 is just a corrected 480.
The 6970 will benefit far more from maturing drivers (read: AMD/ATI going from VLIW 5 arch to now VLIW 4). Hell, I don't think the current drivers were even ready for release as it is. The 6950/6970 are desperate for improved drivers.
Will it catch the 580? Maybe, maybe not. But the 580 doesn't have much more to gain. It's drivers have already been maturing for a year.
Actually the 6970 being such a small gain from 5870 is showing the exact opposite.
