[B3D Analysis] R600 has been unleashed upon an unsuspecting enthusiast community

Are you talking 3870 in relation to 8800 GT? How do you calculate a 48% FLOPS differential?

 

Yep, and that's exactly what I said in the same post you replied to.

Used 2 flops per shader for the GT

775 * 2 * 320 ~= 1.48 * 1500 * 2 * 112

 

Except its 3 FLOPS.

 

Heh, depends on who you ask. Doesn't change the picture much though.

 

When you're discussing shader efficiency, its kinda critical. You're factoring only 66% shader efficiency for G92, which by that token mean that all the cases citing in the examples here are actually more efficient on R6xx than you're accounting for with G9x/G8x.

 

Wavey: Come on, you also have a fixed-function MUL for interpolation on R6xx, just that it's 100% unusable for anything else (not that G8x/G9x is really much better in practice - it was still a pretty dumb idea to market that MUL imo) - so if you use it in efficiency calculations, you need to consider it on R6xx too...

On the other hand, there are some other things that should probably be considered and that tend to be in ATI's favour here. One example is the fact many SFU operations on G8x/G9x require a one-cycle operation on the main ALU to 'setup' the operand (i.e. range clamp etc.) - AFAIK, R6xx doesn't need to do that. So these things really can be substantially harder to compare quantitatively than it might appear on first glance...

 

Isn't the comparison really simple though?
Looking at specs like transistorcount, clockspeed, memory speed, bus width, processing units, GFLOPS rating, power consumption etc, the R6xx has the upper hand on virtually everything...

However, doing some practical benchmarks shows that the G8x/G9x are faster. Obviously they're the more efficient ones. They're better at the things that matter most.

 

Interpolation on R6xx is not part of the general shader array and never purported to be, thats not the case of the MUL on G8x/G9x.

 

I don't want to sound rude or demanding, but I think it would be a very interesting for the staff at B3D to write up a follow up analysis on R600 [compared to G80].

I know I would really enjoy reading the staff's thoughts on what they think of R600; 12 months after the launch.

 

I second this. I think there need to be more articles that compare how things turned out in the long run. If things changed with newer drivers, etc.

 

...

 

I started it today.

 

Better finish soon because next month looks to be busy.

 

Excellent. Looking forward to it.

 

Ditto, looking forward to it!!

 

Also looking forward to it.

 

Awesome, thanks!

I'm greatly looking forward to it.

 

Wow, talk about pressure! Rys, you probably should've kept it a suprise

 

yummy can't wait.

 

Neither can I, I've been waiting for an article like this for months.