NVIDIA GeForce GTX 580 review
Author: Luka Rakamaric
Date: 09 Nov 2010

After being late with its DX11 lineup last year, NVIDIA finally managed to get their Fermi architecture in the form of GTX 480 out to the market in April of this year. By then, AMD has been out with their 5000 series for over 6 months and although the GTX 480 claimed the performance crown, it was big, hot, and not very price competitive at launch.

Although ATI did launch their 6000 series card with the same last three digits as their previous single GPU flagship, the 6870, it is in fact a midrange part, competing with the GTX 460 and occasionally the GTX 470. Their real successor to the 5870, the 6970, is not out yet, while NVIDIA is today introducing their new flagship, the first card of the 500 series.

First, we must say that the naming scheme is once again turned on its head. With the 200 series, a die shrink of the same architecture just got an additional 5 in the name. This time around, there’s no die shrink, and it’s a whole new series. Unfortunately, we can’t really say anything to NVIDIA on the subject as AMD did almost the same thing but with a twist, making their nomenclature successor card slower in almost all scenarios than the previous generation one.

The GTX 580 is based on a new GPU, called the GF110. It is almost the same design as the GF100. What is new are the power requirements. Even with the higher clocks, the additional operating polymorph engine, it still consumes less power. This has been achieved by looking at every transistor and optimizing it. Lower leakage transistors were used where timings where not as important, while higher speed transistors were chosen for more critical processing paths. What you get is a chip that can operate with higher clocks, uses less power and probably has better yields.

Physically, the GPU has the same 512 shaders organized in 16 polymorph engines as the GF100 had. However, NVIDIA never shipped a graphics card with full 512 shaders using that GPU. This is where GF110 comes in. The GTX 580 uses all of the shaders available, and is which should bring it one fifteenth of the shader performance over the GTX 480.

NVIDIA did make two optimizations in the other areas, especially the FP16 texture filtering. This is most notably seen in 3DMark Vantage, and some other texture heavy applications. The other optimization is in the form of additional tile formats that improve Z-cull efficiency.

NVIDIA claims that architecturally, the 580 should bring you 5 to 15 % improved performance over the GTX 480. That means that this increase is before you increase the clocks and the shader count.

Unlike the other members of the Fermi family, both the GF100, and the today’s GF110 don’t support Dolby True HD and DTS Master Audio througput, which only shows us that the GF110 is just a slight evolution and optimization of the GF100 without significant upgrades. Although ATI also only evolved the 5000 series into the 6000, there are some more additional features. But it’s fair to say that it has been a year for AMD and only 6 months for NVIDIA, which explains why the GF110 is closer to the GF100 than Barts and Cayman are to Cypress.