ATi 4870X2(R700) PCB Debut

[Helmore]

I am still wondering, why many see this as a 2 x 800 SP solution, while viewing the 8800 GTX as a 128 SP and GTX 280 as a 240 SP solution. That is just flat out wrong and even GPU-Z gets these numbers wrong. If 8800 GTX is a 128 SP solution, then HD 4870 is a 160 SP solution. If HD 4870 is a 800 SP solution, then 8800 GTX is a 256 SP (and perhaps even a 384 SP) solution.

I don't know where your numbers from or your math is seriously skewed, but that is just not right what you are saying there. It has also been know for a long long time that you just can't compare shader/pipeline (and whatever you want to call it) count between ATI and NVIDIA, if you could then the X850 cards from ATI would have been way faster than NVIDIA's 6800 cards for example and this was clearly not the case.

A GTX280 has 240 stream processors while a 4870 has 800, these SPs on the GTX280 can do 1 MADD and 1 MUL per cycle (1 MADD+1 MUL/clock cycle) while those SPs on the 4870 can do 1 MADD per cycle. So that means the 4870 can do 1600 FLOPs/clock cycle and the GTX280 can do 360 FLOPs/clock cycle, that's not to hard to see now is it? It's just that the SPs on the GTX280 run at a way higher clock and are a little more flexible because there is less dependency between the SPs, but if you would write code that is optimized specifically for each architecture then you can truly get 1200 GigaFLOPs out of the 4870 and only 933 GigaFLOPs out of the GTX280. But we by now know that these numbers mean nothing for in game results though.

[Shintai]

I don't know where your numbers from or your math is seriously skewed, but that is just not right what you are saying there. It has also been know for a long long time that you just can't compare shader/pipeline (and whatever you want to call it) count between ATI and NVIDIA, if you could then the X850 cards from ATI would have been way faster than NVIDIA's 6800 cards for example and this was clearly not the case.

A GTX280 has 240 stream processors while a 4870 has 800, these SPs on the GTX280 can do 1 MADD and 1 MUL per cycle (1 MADD+1 MUL/clock cycle) while those SPs on the 4870 can do 1 MADD per cycle. So that means the 4870 can do 1600 FLOPs/clock cycle and the GTX280 can do 360 FLOPs/clock cycle, that's not to hard to see now is it? It's just that the SPs on the GTX280 run at a way higher clock and are a little more flexible because there is less dependency between the SPs, but if you would write code that is optimized specifically for each architecture then you can truly get 1200 GigaFLOPs out of the 4870 and only 933 GigaFLOPs out of the GTX280. But we by now know that these numbers mean nothing for in game results though.

His math is actually right. Its yours thats wrong, unless you believe marketing.
Even your math is wrong, unless you count a MUL for half or something.

Its simply a simple shader vs complex shader discussion. And marketing couldnt stand being "less".

It would be like claiming a Kentsfield/Yorkfield is a 16 "something" CPU and Bloomsfield 32. And a Phenom X4 for 12. You would instantly debunk that, but for some reason you eat the same logic if applied to GPUs.