We don’t often say that a new graphics card really does bring something startlingly new to the market. With DirectX 10, and the first card to support it, we’re witnessing a complete reworking of how graphics cards operate and where they sit in the processing chain.

DirectX 10 has been completely rewritten from the ground up, and the GeForce 8800 series architecture has been designed specifically for the new API. Asus’ EN8800GTX is the first card to be based on this architecture and, with its GeForce 8800GTX GPU, features the largest graphics processor to date. Architecturally, the core of the chip boasts 128 ‘streaming processors’ operating at 1.35GHz, and is capable of rendering outputs at 16x antialiasing with 128-bit HDR, something Nvidia’s previous generation of GPUs were physically incapable of doing.

DirectX 10 introduces two significantly different concepts – the much publicised unified shader architecture and a change in the role of the actual GPU. The GPU now takes some of the burden of rendering and graphics programming back from the CPU, so there’s less of a bottleneck on processing. This is done through a number of techniques in DirectX 10 that call the CPU less than before, enabling more to be done with code interfacing the graphics card directly – such as more advanced instancing (models built up using variations on one instance of an object).

Shader developments

The Shader Model 4’s unified shader architecture represents the biggest leap forward though – instead of a graphics card boasting often widely different vertex and pixel shaders, unified shaders have access to the full gamut of graphics cards resources. Programming for vertex and pixel shaders requires different techniques, and need to be considered as separate entities. DirectX 10 also introduces the concept of geometry shaders, which support the creation and destruction of data on the GPU.

The full version of the DirectX 10 API will only be available with the full Vista release at the end of January 2007. Add to this a small installed user-base for some time to come, and supporting the new API isn’t going to be enough. DirectX 9 performance is important, at least in the short term. Thankfully, this card offers outstanding performance on today’s applications as well as future ones. 3DMark06 returned the highest score we’ve seen to date at 12,320, which is even faster than a pair of Radeon 1950XTX’s in CrossFire mode. Real-world tests saw a similar story – F.E.A.R, Half Life 2 and Oblivion all returned results that were close to or faster than a pair of the fastest current graphics cards – 87fps, 120fps and 116fps respectively.

Nvidia is releasing two versions of this new GPU – the GTX reviewed here and 8800GTS. The 8800GTX is clocked at 575MHz, has 128 streaming pipelines and has access to 768MB of DDR3 memory through a 384-bit bus. The GTS version has 96 processing units and a slower clock speed of 500MHz and has a smaller 320-bit memory bus. It also has lower power requirements to the GTX, although 400W as a minimum is recommended. Nvidia is introducing a new motherboard chipset, nForce 600i SLI, to accompany the new cards, which supports quad-core and a pair of 8800s.

The GeForce 8800GTX is a stunning piece of engineering – it’s not outrageous to suggest that this is the most important leap forward for graphics cards since the introduction of TnL lighting. This is a vision of the future that doesn’t require a ridiculous leap of faith.