A new performance champion - NVIDIA GeForce GTX 580
NVIDIA releases the GeForce GTX 580 graphics card on November 9, 2010. The purpose of the release is two-fold: to launch the fastest GPU on the planet and to pre-empt arch-rival AMD's release of the competing Radeon HD 6970 graphics card.
The GeForce GTX 580 slots in right at the top of NVIDIA's consumer line-up, supplanting the erstwhile champion GeForce GTX 480. The naming scheme employed suggests that GTX 580 is a brand-new architecture designed from the grounds-up, but this is not the case.
Available to purchase for £399 from a range of partners, the new GPU takes incumbent GTX 480's architecture as a base and corrects much of what was wrong with the first-generation Fermi card, released in March 2010.
Providing some context, the following table highlights the differences between the new GTX 580 and presently-available GTX 480 and GTX 470 cards from NVIDIA.
|GPU||NVIDIA GeForce GTX 580 1,536MB||NVIDIA GeForce GTX 480 1,536MB||NVIDIA GeForce GTX 470 1,280MB|
|Memory interface||384-bit, 1,536MB GDDR5||384-bit, 1,280MB GDDR5||320-bit, 1,280MB GDDR5|
|Board power (TDP)||244W||250W||215W|
|Outputs||2x dual-link DVI; mini-HDMI||2x dual-link DVI; mini-HDMI||2x dual-link DVI; mini-HDMI|
GTX 580 uses a newer revision of the Fermi architecture, known as GF110. It improves upon GF100 (GTX 480) in four main ways. Firstly, the new performance GPU is clocked in higher; shipping with a 772MHz core, 1,544MHz shaders, and 4,008MHz memory - representing a near-10 per cent boost over GTX 480.
Secondly, it uses the full 512 cores available from the Fermi architecture, up from 480 cores, and this provides an extra few per cent of gaming power. The modular design of Fermi means that this improvement is also applicable to the geometry setup, known as PolyMorph engines, as well the shading cores.
NVIDIA also makes an improvement with respect to GTX 580's filtering of what are known as FP16 textures. These are used for high-quality lighting in games - known as HDR - and the new GPU can process them at double the speed of GeForce GTX 480. Not shown on the table, NVIDIA also improves what is termed the z-cull algorithm. In plain English, this means the GPU intelligently discards parts of the image that won't be seen by the viewer - a rock behind a tree, for example.
Improving the speed of GTX 580 is moot if the amount of power pulled by the card increases. Present GTX 480 is a hot-running beastie, so increasing the speeds further brings thermal issues very much into play.
NVIDIA says that its engineers have re-architected large portions of the GPU with the express aim of fitting in more speed with a lower power-draw. The table shows that the GTX 580 pulls six fewer watts than GTX 480 when under load, yet it is faster in every respect. What's more, the new card, GTX 580, has a revised cooler that uses an upgraded heatsink-and-fan unit for quieter running at both idle and peak speeds.
Carrying on the improvements, GeForce GTX 580 also has hardware-based monitoring on the PCB which interfaces with the software driver and evaluates the power being consumed by the GPU. The main reason for this is to ensure that the card doesn't overheat when particular stress-testing applications - such as FurMark and OCCT - are used. Should a flagged program be loaded, the monitoring chips reduce the GPU's speed to keep the card below the 244W TDP. The hardware monitoring is limited to stress-testing apps, though, and users are free to overclock the card and play all regular games, even if this pushes the TDP past the rated 244W.
Through the improvements discussed above, GeForce GTX 580 is around 15 per cent faster than a GeForce GTX 480 across a number of high-profile games. Improvements vary on a per-title basis, and those games which use lots of FP16 textures and significant geometry setup see gains of up to 25 per cent at high resolutions.
NVIDIA's GeForce GTX 580, then, is the fastest single-GPU graphics card around, albeit not the fastest graphics card. The statement is laced with the proviso that, while thunderously quick, GTX 580 isn't quite as rapid as the twin-GPU AMD Radeon HD 5970 graphics card launched close to a year ago. AMD's current single-GPU champion graphics card is the Radeon HD 5870, which is some 30 per cent slower than the GTX 580, but is going to be replaced by the all-new Radeon HD 6970 on November 22, 2010.
The question on enthusiasts' lips is which one will be best: NVIDIA GeForce GTX 580 or AMD Radeon HD 6970? The answer to this tantalising question will be known within two weeks, so while NVIDIA has done a good job with its best Fermi card yet, wait a short while for the definitive verdict.