Pure Power Performance
NVIDIA's own internal benchmarks reveal that the 8x AGP-compliant GeForce FX has 3x the raw frame rate power of the NV25 (GeForce4 Ti 4xxxx) and 3x the vertex processing. Sporting 125-million transistors and manufactured on a 0.13u process, it's arguably the most technologically advanced GPU available. It's manufactured using the same flip-chip technology as the R300.
We've seen faster and faster RAM specified on modern graphics cards. It seems as if buckets of bandwidth, even with the presence of memory-saving technologies and loss-less data compression, is still paramount. Using DDR-II technology, where bandwidth is further doubled from standard DDR by increasing the data fetch from 2 to 4 bits, we have 1GHz on offer for the first time. Massive bandwidth and the memory-saving techniques that the GeForce FX brings to the table should give us bandwidth to spare until the very highest resolutions.
No other concrete facts are available, but it seems that the FX will debut with a top-end core speed of 500MHz. Using the same 8 pixel pipelines as the R300, and the same 1 texture per pipe, but with 2 TMUs (as opposed to the R300's 1), the FX should outmuscle its competition in the pure power stakes. Reckoned to be providing well over 40GB/s from its comparatively small 128-bit memory interface, it achieves this figure with its effective memory bandwidth-saving techniques.
Add to that a sophisticated vertex and pixel shading ability (used in conjunction with NVIDIA's in-house Cg language), the ability to work in true 128-bit colour for precise rendering, and a proprietary 4:1 loss-less data compression that leads to super-efficient bandwidth use for F.S.A.A, together with some intelligent adaptive texture filtering, and you have an impressive specification sheet. If you want to see what one looks like, this may just be what you're buying in a couple of months time.
Nice cooling, eh ?.
