Today NVIDIA launches their new flagship graphics product, the NVIDIA GeForce GTX 580. This is NVIDIA's highest performing part to date, and the current fastest single GPU graphics card in the world. Designed to provide the ultimate performance in current and next generation of DirectX 11 titles, NVIDIA's GeForce GTX 580 also addresses some of the issues seen with their first Fermi-architecture consumer graphics card, the GeForce GTX 480.

NVIDIA's FERMI architecture was introduced in late 2009, as NVIDIA began to generate excitement and interest in their new DirectX 11 product line. NVIDIA missed the important Microsoft Windows 7 launch and Christmas sales bubble targets but, beginning in April 2010, ultimately delivered four GPUs in seven months. The Fermi architecture is the basis for new products in the GeForce, Quadro and Tesla product lines, with GeForce the most interesting to us as it's the consumer, desktop line. Currently Fermi GPUs power seven products, covering the entry level market with the GeForce GT 430, the mainstream gamer and performance enthusiast market with the GTS 450 and GTX 460 series, and finally the high end enthusiast and ultra enthusiast market with the GTX 465, GTX 470 and GTX 480.  Now, seven months after Fermi graphics cards hit the stores, we get our first look at Nvidia's GTX 580.

NVIDIA GeForce GTX 580

Specifications

$499 SEP

3Bn Transistor 40nm GF110 core

512 CUDA Cores

16 Streaming Multi-Processors (SM) in 4 Graphics Processing Clusters (GPC)

772MHz Core clock / 1544MHz Shader Clock

1536MB GDDR5 RAM at 1002MHz / 4Gbps QDR

384-bit Memory Interface with 192.4GB/sec Bandwidth

2 Dual-Link DVI outputs + 1 mini-HDMI 1.4a output (max two active at a time)

Dual-Height with PCI-Express 2.0 x16 interface, 10.5" Long

Single PCI-E 6pin plus Single PCI-E 8pin power connectors required, 600W PSU recommended

244W Max Board Power

What we have here is a tweaked from the transistors up Fermi, revamped for better performance, better power usage, and with an improved cooling solution. Can AMD's current line up compete with Fermi 'done right'?