As its name implies, the GeForce FX 5950 Ultra core is a slight improvement over the GeForce FX 5900 Ultra. Whereas the GeForce FX 5900 Ultra operates at 450MHz, the GeForce FX 5950 Ultra core runs at 475MHz yielding an additional 200Mtexels/second fill-rate, a 5% improvement. Meanwhile, the memory clock frequency has been increased from 425MHz in GeForce FX 5900 Ultra (850MHz effective) to 475MHz in GeForce FX 5950 Ultra (950MHz effective), this bumps memory bandwidth by 3.2GB/sec to 30.4GB/sec total.
The core itself is still a four pixel pipeline architecture, with two texture units per pixel pipeline, and is still based on TSMC’s 0.13-micron manufacturing process with copper interconnects. You’ve still got the same 256-bit memory interface (the key highlight of GeForce FX 5900 Ultra) as well.
Leadtek compared to GeForce4
In the grand scheme of things, this makes the GeForce FX 5950 Ultra a minor progression over the GeForce FX 5900 Ultra’s NV35 core. If you recall NVIDIA’s previous refresh product that was timed for the Holiday shopping season, the GeForce2/3 Titanium series (has it really been that long since NVIDIA released a refresh product in the Fall?), the GeForce FX 5950 Ultra shouldn’t come as too much of a surprise. Fortunately for consumers, its arrival has lowered prices on GeForce FX 5900/5900 Ultra cards.
GeForce FX 5700 core
The GeForce FX 5700 Ultra however is based on an entirely new graphics core that’s intended to serve the mainstream market. Rather than refresh the GeForce FX 5600 Ultra core as they have done on the high-end, NVIDIA has used the NV35 core as the basis for GeForce FX 5700, with modifications made to reduce the chip's production costs.
First, NVIDIA reduced one of the TMUs, bringing the GeForce FX 5700 to a four-pixel pipeline architecture with one texture unit per pixel pipeline. NVIDIA also utilizes a narrower 128-bit memory interface for the GeForce FX 5700, rather than relying on the more expensive 256-bit interface found in the GeForce FX 5900 series. NVIDIA employs dual 64-bit memory controllers for the GeForce FX 5700 family.
These changes reduce peak memory bandwidth and fill rate figures, but at the same time the core is less expensive to produce weighing in at 82 million transistors; 48 million less than GeForce FX 5900.
Pixel shader performance has been improved, thanks to the addition of NVIDIA’s CineFX 2.0 technology, as well as enhanced stencil performance if the application takes advantage of UltraShadow, which is another feature the GeForce FX 5700 boasts. The final addition are the new compression algorithms found in Intellisample HCT, which should improve anti-alasing performance at high screen resolutions. We put NVIDIA’s CineFX 2.0 to the test in our eVGA e-GeForce FX 5700 Ultra review and found that shader performance nearly improved by a factor of two in some cases with ShaderMark, a synthetic DX9 benchmark we’ve been using recently.