AMD Cayman info (or rumor)

[KoHaN69]

SO is the 6970 2GB?

[Pantsu]

According to the article there's only 2GB 6970, and two versions of 6950, 1GB and 2GB.

[kuroikenshi]

Yep... almost always 2x faster than Cypress and up to 3x faster

edit: looks like slide 72 still needs to be filled in.

What a tease

oh, no... who said it was not going to perform to the level of the gtx580???? i know of many people who did

[zerazax]

Cayman is REALLY different from Barts indeed. 4D is confirmed as well + lots of OpenCL / DirectCompute features

From the site:




Oh and for all the people crying about 32 ROPs:


Last time AMD beefed up the ROPs.... RV770

[Chrono Detector]

Specs look good though I'm disappointed they are still using a 256 bit memory bus, but I guess having 2GB with 1920 SP's is pretty good.

[zerazax]

Specs look good though I'm disappointed they are still using a 256 bit memory bus, but I guess having 2GB with 1920 SP's is pretty good.

I bet bigger width is unnecessary... AMD's been really efficient in terms of memory bandwidth efficiency

It would seem silly for them to have doubled RBE power, doubled tesselation power, increased shaders etc. w/o having considering memory bandwidth

[kuroikenshi]

i smell antilles will be a knockout

[donitsi]

The specs are nice indeed!

There's a update in the end which says 6970 is 3,5 TFlops and 6950 2,5TFlops

Folders rejoice!

[SimBy]

If utilization is improved a lot we can't compare flops to Cypress anymore.

[Lanek]

Yep... almost always 2x faster than Cypress and up to 3x faster

edit: looks like slide 72 still needs to be filled in.

What a tease

32 pixel/polygon vs 16 for the 5870 ( and surely more but with a cost on perf ) ...

[duploxxx]

it seems like AMD(ATI) had lots of time to redesign the arch (Fermi delay) ... Barts which is optimized cypress and full new cayman with VL/W4.

[spursindonesia]

So, with 2x tesselation & geometry capability at high tesselation factor, is this leaked graph making any sense ??



HD 6970 is slightly more than twice faster over HF 5870, but that might be because better optimization in Cayman mArch, clock advantage (900 vs 850 MHz), or AMD being slightly modest in stating Cayman advantage over its older brother.

[railer]

So this new generation is still slower then Nvidia's last in tessellation
If i correctly understand the slides only 2x performance increase from 5870? I hope it does better in general gaming. Specs looks nice on paper.

[SimBy]

So this new generation is still slower then Nvidia's last in tessellation
If i correctly understand the slides only 2x performance increase from 5870? I hope it does better in general gaming. Specs looks nice on paper.

If I'm not mistaken nV has 1.8x peak performance lead over Cypress in Heaven benchmark with extreme tessellation.

So how exactly is it slower?

[flopper]

So this new generation is still slower then Nvidia's last in tessellation
If i correctly understand the slides only 2x performance increase from 5870? I hope it does better in general gaming. Specs looks nice on paper.

yes Nvidia have done tesselation right and proper.
amd is trying to catch up and likely might do it with 7000 or 8000 series.

[railer]

Well is that the case with 580? Not sure where you got 1.8 i thought Nvidia has about 2x the performance. Lets see when the real thing comes out how its going to pan out. Right now its just speculation anyway. If Cayman is that good in tessellation and catches up to the green its the best scenario for us. Me like competition

[SimBy]

Well is that the case with 580? Not sure where you got 1.8 i thought Nvidia has about 2x the performance. Lets see when the real thing comes out how its going to pan out. Right now its just speculation anyway. If Cayman is that good in tessellation and catches up to the green its the best scenario for us. Me like competition

Add 20% for GTX580.

This is official nV slide btw.

[zerazax]

looks like the Polish site got taken down lol..

If its 1.6x on average with another 20% for the 580... according to AMD, Cayman should be right there

[railer]

Well thanks for the slides. Lets hope these speculations come true. If Cayman will beat Nvidia offerings in tessellation it's sure to be sweet for the red team. Then we will see lower prices and more innovation.

[Mechanical Man]

Here: ALL DIAS

[zerazax]

This looks awesome btw:


So Cayman has:
Better Tesselation performance (2-3X over 5870)
Better shader design (4-VLIW)
Improved GPU compute
New AA mode - EQAA
Beefed up ROPs in RBE
Power management via adjusting power per clock cycle

[Dainas]

twice the polygon per clock is intriguing.

[Oliverda]

This looks awesome btw:

So Cayman has:
Better Tesselation performance (2-3X over 5870)
Better shader design (4-VLIW)
Improved GPU compute
New AA mode - EQAA
Beefed up ROPs in RBE
Power management via adjusting power per clock cycle

AMD's Turbo Core reached the GPU line.

[kuroikenshi]

AMD's Turbo Core reached the GPU line.

cha-ching

[Vipeax]