NVIDIA’s 216 shader GeForce GTX 260
Going into the summer, NVIDIA must have felt pretty confident about their graphics lineup. Their G92 GPU had been adapted to run in a variety of cards ranging from the GeForce 8800 GS/9600 GSO all the way up to the dual GPU GeForce 9800 GX2. In each of these segments G92 was the unquestioned performance champ, and NVIDIA was still riding the wave of GeForce 8800 GT sales -- the 8800 GT will probably go down as one of the most popular graphics upgrades of its era. Meanwhile NVIDIA was preparing to unleash a pair of new graphics cards to take on the high-end segment: the GeForce GTX 260 and the GTX 280.
With 240 stream processors clocked at nearly 1.3GHz, a 512-bit memory interface providing over 141GB/sec of memory bandwidth and 1.4 billion transistors, NVIDIA’s GeForce GTX 280 GPU boasted impressive specs on paper. Its cheaper sibling, the GeForce GTX 260 is also a pretty formidable graphics processor featuring 192 shaders and a slightly narrower 448-bit memory interface capable of delivering 111.9GB/sec of peak memory bandwidth to the GPU. In practice with actual games, the GeForce GTX 260 ran 19-27% faster than the GeForce 8800 GTX in our testing, although there were a few cases where it was nearly 40% faster than the 8800 GTX. The GeForce GTX 280 generally ran around 15% faster than the GTX 260.
All this sounds good in theory right? It turns out ATI had a game changer with their RV770 architecture.
Going into 2008, everyone underestimated the capabilities of ATI’s RV770 GPU. We all knew ATI had two SKUs planned for launch, a 4850 and a 4870, and that the architecture would boast more shaders than RV670, but performance expectations were rather mild. It was expected that the 4850 SKU would run a little bit faster than NVIDIA’s GeForce 8800 GT, while the 4870 was gunning for the GeForce 9800 GTX. Estimates pegged it to run about 10-20% faster than the 9800 GTX.
Ultimately those estimates from January 2008 were way off: the Radeon 4850 was priced like an 8800 GT, but actually outperforms the bone stock GeForce 9800 GTX in most cases, while the Radeon 4870 is more of a competitor for the GeForce GTX 260 than anything else NVIDIA offers while boasting a $300 price tag.
As this became more apparent NVIDIA was forced to respond. The MSRP of the GeForce GTX 260 went from $450 ahead of the GTX launch to $400 on launch day. Then over the July 4th holiday NVIDIA cut GTX 260 prices again, with the GPU falling to $300-$330. Today GeForce GTX 260 cards can now be found selling online for less than $250 after rebate. Meanwhile the GeForce GTX 280 went from $650 on launch day to less than $400 after rebate today.
Obviously with a 1.4 billion transistor part sporting a 404-bit or 512-bit memory interface with 896MB or 1GB of memory and a massive die size that’s nearly 600 square millimeters, these cards aren’t cheap for NVIDIA to produce. NVIDIA is now scrambling to bring the GPU’s die size down by going from TSMC’s 65-nm manufacturing process to their smaller 55-nm process, and it’s believed that this 55-nm GT200 GPU will also use a narrower 256-bit memory interface with GDDR5 memory, similar to the arrangement ATI employs with their Radeon 4870. These two steps should reduce GT200’s manufacturing costs for NVIDIA. But in the meantime until these 55-nm GPUs are ready, NVIDIA’s strategy is to improve GT200 performance. More specifically the GT200 GPU that’s hurting their margins the worst: the GeForce GTX 260.
By improving GTX 260’s performance they should be able to charge a higher price for this enhanced GTX 260 GPU variant. As a result of the higher price, profit margins should improve, as NVIDIA’s production cost remains fixed.
That’s the theory at least. But is NVIDIA’s enhanced GTX 260 GPU capable of delivering enough performance to justify a higher price tag? That’s the question we’re here today to answer.