Looks like an accurate slide for 6970/Cayman specs

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Skurge

Diamond Member
Aug 17, 2009
5,195
1
71
Apparently the memory bus is the hardest thing to scale down. That is, if we pretend that shaders can be scaled down linearly with the process size, the memory bus wouldn't scale linearly. Thus it costs a lot more to do 384-bit than 256-bit (not to mention you'd have to add more memory chips, and PCB complexity increases)



Actually, AMD is 1600/5 = 320 SIMDs



Shader groups matter in real world benchmarks because apparently, it was extremely hard to feed AMD's 5-way configuration. In fact, it was rumored that internal benchmarks showed that the 5-way was only being used 80% of the time - meaning the 1600SP Cypress acted more like it was a 1280 SP card. Synthetic benchmarks, coded specifically to use all 5, however, were able to utilize the 1600

Going to 4D apparently puts it VERY close to the equivalent in 5D - meaning that a 320 SIMD 4D configuration is very close to the 320 SIMD 5-way config currently existing.

Thus SIMD count matters a lot.. going from 320 -> 480 is a 50% increase

And apparently a more real slide:

amdradeonhd6900slide6.jpg

Nice, it says November 18 2010 at the bottom.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,812
1,550
136
So, Antilles matches GTX 580 for geometry throughput? Tha'ts not completely impressive, but at the same time it's good that they are moving Forward.

The much more impressive number here is shader performance. Antilles has as much dual precision muscle as GTX 580 has single precision.
 

formulav8

Diamond Member
Sep 18, 2000
7,004
523
126
One thing you can count on is the new GPU giving nvidia a run for its money in the consumption compartment :hmm:

It will of course perform very well but I bet it will be a huge die.

I like hardware as much as the next guy but fannyism is just stupid. I won't lose any sleep whether AMD kills the GTX580 or not. Its just stupid hardware in the end. ;)():)
 

dookulooku

Member
Aug 29, 2008
93
0
0
It's been speculated that the slide with the 5970 specs has been altered. The SIMD engine and stream processor numbers seem to use a different font/font size than the rest of the slide. It could just be a bad editing by whoever made the slide though.
 

Gloomy

Golden Member
Oct 12, 2010
1,469
21
81
And whats your point? I'm not a fanboi like 99% of the people on this forum. I play both sides :biggrin:

My point is that after spending 499 dollars on the latest and greatest, I wouldn't be looking at anything else for a while, green or red. ;)
 
Feb 19, 2009
10,457
10
76
So we have a 50% increase in SP groups. 480 vs 320 in cypress and 224 in barts.
There's beefed up TMUs no longer linked to each SIMD thus should increase efficiency hugely.
Scalable independent tessellator.
Improved bandwidth usage buffering.
Doubled poly/clock, tweaked rasterizers.

And the funny part, two of these fits 300W TDP. Therefore, its not going to be >225W each and not a huge die like NV fud suggests.

*IF* they get their drivers right for this architecture, there's room for much more than 50% increase over cypress (There were leaks with old cat10.10, had cayman 35% better than cypress. There was another set of leaks on internal ATI drivers optimized for cayman had it ~50% better than cypress.). Thats my only concern. Totally new architecture is going to be a nightmare on their driver team.

Why do i think more than 50%? 480 vs 224 barts. Crossfire barts hammer the gtx580. Thats barts without most of the new architecture improvements.

Edit: Something to ponder about, these specs look legit, performance should be very good. Even if there's drMOS constraints, they have enough for a soft launch earlier, get reviews out there to put pressure on NV. Why didn't they? I think its their messy drivers, probably buggy or crash prone in games. :(
 
Last edited:

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
64 pixels / clock.

Mad AA powweeeeeeeeeeeeeeer incoming!

slide32renderbackend.jpg


Mad double-precision powweeeeeeeeeeeeeeeer incoming!!!

HD4870
1.2 TeraFLOPS Single-precision
240 GigaFLOPS Double-precision

HD5870
2.7 TeraFLOPS single-precision
544 GigaFLOPS double-precision

HD6990
6.0 TeraFLOPS Single-precision
1.5 TeraFLOPS Double-precision :biggrin:
 
Last edited:
Feb 19, 2009
10,457
10
76
64 pixels / clock.

Mad AA powweeeeeeeeeeeeeeer!

slide32renderbackend.jpg


Mad GPGPU powweeeeeeeeeeeeeeeer!!!

HD4870
1.2 TeraFLOPS Single-precision
240 GigaFLOPS Double-precision

HD5870
2.7 TeraFLOPS single-precision
544 GigaFLOPS double-precision

HD6990
6.0 TeraFLOPS Single-precision
1.5 TeraFLOPS Double-precision :biggrin:

Antilles most likely uses downclock XT cores. So cayman xt would be what, 0.85Tflops dp? Holy cow, are they going for the HPC crowd or what.?