| ArchangelMadzz said: By controller I meant, the actual game controller, as it seems odd to sell something like Scorpio without the Elite controller. |
Rendering a game is more than just the floating point math.
Let's use the Radeon 5870 as an example here, based on the Terascale 2 architecture, aka. VLIW5 (Very Long Instruction Word-5 way.)
It had 2.720 Teraflops.
It's successor the Radeon 6950 is based on Terascale 3, Aka. VLIW4. (Very Long Instruction Word-4 Way.)
It had 2.253 Teraflops.
Now despite the fact that the Radeon 6950 was almost 450Gflop's slower than than the 5870, it pretty much outperformed it in every game.
But dont take my word for it.
http://www.anandtech.com/bench/product/510?vs=511
Let's take the Radeon 7850 then shall we? Based on Graphics Core Next 1.0, which introduced a massive change compared to AMD's prior architectures.
It has 1.761 Teraflops.
Despite the fact that it has 950Gflop (That is almost an ENTIRE Teraflop!) It actually out-performs the Radeon 5870.
But don't take my word for it.
http://www.anandtech.com/bench/product/549?vs=511
Again I reiterate, Gflops is not a denominator for gauging the complete performance of a graphics processor, you have bandwidth, compression, caching, scheduling, geometry, texturing and more to worry about.
You could have a 9000 Petaflop graphics processor that is slower than a 1 Terfalop Graphics processor if it's design is inefficient.
Even in heavy-floating point compute scenario's (Where the use of Gflop should have some kind of importance) the 7850 will STILL beat the 5870 every single time.
Now it gets even more ironic as this is all a comparison between AMD's hardware, the difference becomes even more pronounced when you throw nVidia into the mix.
So. I ask you this. Still think your Flops is an accurate denominator for gauging performance between different graphics processing architectures? Because it's not. Never has been, just like how "Mhz" isn't a way to determine the performance of a CPU.

www.youtube.com/@Pemalite








