Final thoughts on GeForce GTX 480GF100
NVIDIA GF100 GPU represents the first grounds-up design since the excellent G80 was released over three years ago. Plagued by delay after delay in which time the competition has released a top-to-bottom range of DX11 GPUs, NVIDIA needs to get its latest graphics architecture right.
Viewed as a design, the also-DX11-compliant GF100 (Fermi) is much more than a renderer of pretty pixels. At its highest level, the 3bn-transistor chip uses four generally self-contained mini-GPUs that combine to provide decent theoretical performance for a wide variety of applications and code.
Fermi indicates NVIDIA's GPU direction by paying particular focus on increased geometry and tessellation via numerous GPU-distributed engines, to GPGPU usage through significant on-die cache, along with pretty-looking pixels via efficient antialiasing performance. The design is well-balanced for the most part, providing robust features for both upcoming Tesla and incumbent GeForce lines.
NVIDIA's mindset regarding Fermi's all-new architecture has been one obvious cause of delays, but physically implementing a new thinking on a GPU whilst also being competitive against the AMD's arsenal of quality 5-series cards has been another stumbling block.
GeForce GTX 480
The GeForce brand needs to take the forward-looking Fermi design and bring it to market with economic and manufacturing realities fully taken into account. GeForce GTX 480 represents the desktop manifestation of the inevitable compromise between idealism and pragmatism.
The GeForce GTX 480 chip can't be manufactured in any real volume with the full 512 cores intact, meaning that it loses a stream-processing unit composed of 32 cores. Ramping up clocks on a 3bn-transistor GPU manufactured on a 40nm process is also difficult without running into fundamental thermal concerns, hence why GTX 480 isn't clocked especially high, and the card chews through watts with gay abandon.
Conjecturing somewhat, GeForce GTX 480 is probably 75 per cent of the high-end GPU that was imagined by NVIDIA early last year. Our numbers show that NVIDIA's finest single-GPU card is, on average, 10-20 per cent faster than AMD's Radeon HD 5870 1,024MB at a 2,560x1,600 resolution. GeForce GTX 480 is due to cost some 40 per cent more, so whilst the trade-off between extra expense and performance isn't ideal, it's not completely disastrous for NVIDIA.
How does it play out for the gamer right now? Radeon HD 5970 remains the world's fastest graphics card, GeForce GTX 480 becomes the world's fastest single-GPU card, and Radeon HD 5870 is still a very good bet at sub-£300. Given a choice with value factored in, we'd take the HD 5870.
Bottom line: NVIDIA's GeForce GTX 480 could have easily been better, perhaps should have been considering the time of arrival, but we feel that, underscored and handicapped by a paradigm shift in GPU thinking, it retains enough features and visceral ooh la la to be worthy of a £350-£400 price-tag, if not £450.
Fastest single-GPU card around
Very good scaling in SLI
The not so good
Comparatively expensive at £440
Hot-running GPU and high temperatures
Very noisy at full load
Not the full 512-core model we had hoped for
NVIDIA GF100 architecture
NVIDIA GeForce GTX 480 1,536MB
At HEXUS, we invite the companies whose products we test to comment on our articles. If any company representatives for the products reviewed choose to respond, we'll publish their commentary here verbatim.