Quote Originally Posted by LightSpeed View Post
Seems to be a 5 VLIW design. 16x5x14 = 1120 shaders total. Might be mistaken though, and I dont know if there are shaders disabled but it is very unlikely. Id say 6870 should be within 10% of a 5870.

6850 should have some shaders disabled and perform slightly slower than the 460 1GB IMO. Not bad

Improved Tesselation, AA, AF, added features, better Perf/mm^2, Perf/Watt, great overclocking headroom (I know that the 6870 is a great overclocker, but lets just say its a rumour) all add up and are most welcome. Heres hoping they are priced right. $199/$249 should translate to £169/£209 over here, thats great if true. Id get better than 5850 performance for the same money. Small leap, but every little bit helps

edit: Tajoh, 5850 @ 5870 speeds is still about 3 - 5% slower. This is faster than 5850 having 320 less shaders at roughly 5870 speeds. Surely there have been changes to the architecture, but maybe not dramatic..

Since they have added a bunch of hardware features, why wouldnt they tweak the arch while they are at it? As I mentioned a month or so back, Barts is a tweaked Cypress with added features.
I think that extra 50mhz is helping out as well as the extra memory bandwidth quite a bit. The extra shaders are not helping that much because a bottle neck is already being reached I think.

I don't think its's really that much better than a 5850. Your getting less hardware for the money. I think a 5850 overclocked to 900 mhz would beat a 6830 quite handily. If I had the choice to buy both at the same price, I would pick the 5850 as it is more potent because of the performance/clock.

Performance per mm is only benefiting the AMD. The power usage seems similar too between the two cards and I have a feeling game performance is going to more similar than different.

Judging from that review I listed of the 5830(notice the vantage scores too), I think a 5830 at 900/1200clk with double the ROPs would perform exactly like barts xt. Bare minimum I think, doubling the Rops would add 10% to the benchmarks results in that review I posted.

I think the better tessellation performance is entirely from clocks. So if you overclock the 5850 to 900/1200Clks, you would get the same tessellation performance.