Conclusion

First off, you'll notice that we didn't check out ROP throughput explicitly just yet. That's pending update to this piece which will appear shortly, after we finish organising the data. If you have a G80-based board, it's not hard to see 60Gpixels/sec+ Z-only (with 8800 GTX), which is easily 4x what's possible with something like 7900 GTX or Radeon X1950 XTX, should you want to see one of the big numbers associated with G80 ROP throughput before we add them in. Until we get there, the conclusion is that the ROP rates for pixel and Z fill, and blending, are class leading (often by some margin).

Therefore, with all that data in hand, be it real-world in games at high resolutions or using synthetic tools for measurement, it's not hard to conclude that NVIDIA G80 is the highest performing consumer graphics (and now professionally too, powering a new generation of Quadros) chip ever produced. It achieves that via a balanced architecture that heavily threads execution over a set of resources that seek to always compliment each other.

From the sampler hardware to the shading core to the 24 ROPs, all 681 million 90nm transistors conspire to outrun anything that's come before it, at least in GeForce 8800 GTX form. We'll go into more detail about shader core efficiency, video performance and other minutiae in a piece that'll touch heavily on CUDA and other aspects of the core, as described in the architecture piece.

GeForce 8800 GTS generally outperforms GeForce 7950 GX2 in games at common resolutions and the cost of applying the higher levels of G80's AA ability are sufferable in many titles. Further, it appears NVIDIA are working with Microsoft to allow Direct3D to correctly expose coverage sample AA (CSAA) via the API, negating one of the criticisms we had during the initial look where we mentioned CSAA is less effective if the user has to enable it outside of a game. If that happens we'll be the first to let you know.

Therefore GeForce 8800 GTX sits atop the consumer performance pile, G80 in Quadro form looks healthy and with support for D3D10 (in-house benchmarks coming!), we can see why many are looking for NVIDIA to expand the base architecture's traits and performance characteristics down into the mid-range and low-end performance spaces. Consumers are looking for the base performance and features elsewhere, so this summer's fight between NVIDIA G86 and G84, and AMD RV6xx looks to be a sizzler.

As for R600 vs G80, which looks set to be the biggest heavyweight graphics fight since consumer real-time 3D graphics became a reality, we wait with very baited breath. With so much to cover in a performance analysis, we hope we touched on the big stuff in this piece. Let us know if there's anything specific you'd like us to cover which we didn't here, that isn't mentioned as coming soon.

Comment

Want to comment on this article?  We have a comments thread setup for that very purpose.