NVIDIA releases GeForce 8800GTX, 8800GTS, 680i, 680a

The first DirectX 10 graphics cards have hit us. Everything's changed. Here's the lowdown.

We were invited recently to attend NVIDIA's press tour in Taipei, Taiwan where it revealed to us its new products to close out the year – the GeForce 8800GTX/GTS, and nForce 680a/i.

NVIDIA have hit the world with the first DirectX 10 parts – the GeForce 8800GTX and GTS.

Four years in development, the card formerly known as G80 is a complete redesign around a 90nm process, supporting DX10, SM4.0 and as a consequence a unified shader pipeline.

681 million transistors do the hard slog of bringing an image to your screen, and contribute to the biggest GPU chip dimensions-wise that we've seen. As in, seriously freaking huge. We can bet that NVIDIA is hankering for the next process shrink to get the die down to a decent size.

The first thing you notice about the GTX is it's long, a whole 27cm and hangs easily off the end of a standard motherboard. In fact the first impression with its large heatsink and of course length was that the GeForce FX was reborn – fortunately a complete misconception as it makes very little noise and 3D-wise trucks along at quite a pace – at least as far as the demonstrations showed on an overclocked Core 2 Duo X6800 @ 3.8GHz, on an FSB of 472 (multiplier 8x). Yeah, you heard right, that's what the new nForce 680i was running, although we hadn't the opportunity to check the stability with Stress Prime. But we'll come back to this.

GeForce 8800

The GeForce 8800GTX requires two power plugs and may be too long for some cases.

The GTS is a little stockier, requires only one power plug and is missing the second SLI-type connector.

Weird Bits
The second thing you notice is the second SLI connector, which isn't actually for SLI. The first certainly is, and a longer bridge will be required than in the past as the 680 PCI-E x16 slots are further apart, however the purpose of the second connector is for, direct quoted 'future stuff'. We're putting money on a discrete physics part, or at a very long shot a triple SLI solution but we'll see.

SLI bridges are now considerably longer due to increased distance between the cards.

An extra chip sits closer to the back plate, having been separated from the main chip – this contains the TDMS logic. That is, it handles such things as Dual Link DVI out, TV out, and HDCP. It was split out to simplify the board routing and minimise cost.

Two PCI-E power connectors are required, with the GTX typically consuming 115-120W of power with a max TDP of 177W, while the specs so far for the GTS indicate a max TDP of 147W, leaving NVIDIA to recommend a 450W and 400W PSUs respectively.

The 8800GTX features a 384 bit memory interface and 768MB, the GTS 320bit with 640MB. Why the weird numbers? Well in cost versus performance, NVIDIA claims that if it widened the bus or increased the memory capacity there was not a significant enough performance increase to justify the cost, at least in terms of this generation. NVIDIA is also still using GDDR3 – presumably due to cost of integration.

We have yet to see the rumoured water cooler and NVIDIA, like all clever companies, will not comment on unreleased product, whether it exists in the pipeline or not.

The shader processors (which NVIDIA call 'Stream Processors', or 'SPs'), since unified have been given a clock speed of 1350MHz on the GTX and 1200MHz on its little sibling – at the moment as far as overclocking is concerned the SPs will scale with the core clock – future versions of nTune should allow independent clocking.

	8800GTX	8800GTS
Core Frequency	575 MHz	500MHz
Mem Frequency	1800MHz	1600MHz
Mem Capacity	768MB GDDR3	640MB GDDR3
Stream Processors (SP)	128	96
SP Frequency	1350MHz	1200MHz
ROP	24 pixels per clock	20 pixels per clock
PCI-E Power Connectors	2	1
SLI Connectors	2	1