Archive 1997–2024

NVIDIA's 1.4 Billion Transistor GPU: GT200 Arrives as the GeForce GTX 280 & 260

Final Words

There's no question that NVIDIA has built a very impressive chip with the GT200. As the largest microprocessor we've ever reviewed, NVIDIA has packed an unreal amount of computational horsepower into the GT200. What's even more impressive is that we can fully expect NVIDIA to double transistor count once again in about 18 months, and once more we'll be in this position of complete awe of what can be done. We're a little over a decade away from being able to render and display images that would be nearly indistinguishable from reality, and it's going to take massive GPUs like the GT200 to get us there.

Interestingly, though, AMD has decided to make public its decision to go in the opposite direction. No more will ATI be pushing as many transistors as possible into giant packages in order to do battle with NVIDIA for the coveted "halo" product that inspires the masses to think an entire company is better because they made the fastest possible thing regardless of value. The new direction ATI will go in will be one that it kind of stumbled inadvertently into: providing midrange cards that offer as high a performance per dollar as possible.

With AMD dropping out of the high end single-GPU space (they will still compete with multiGPU solutions), NVIDIA will be left all alone with top performance for the forseable future. But as we saw from our benchmarks, that doesn't always work out quite like we would expect.

There's another very important aspect of GT200 that's worth considering: a die-shrunk, higher clocked version of GT200 will eventually compete with Intel's Larrabee GPU. The GT200 is big enough that it could easily smuggle a Penryn into your system without you noticing, which despite being hilarious also highlights a very important point: NVIDIA could easily toss a high performance general purpose sequential microprocessor on its GPUs if it wanted to. At the same time, if NVIDIA can build a 1.4 billion transistor chip that's nearly 6x the size of Penryn, so can Intel - the difference being that Intel already has the high performance, general purpose, sequential microprocessor that it could integrate alongside a highly parallel GPU workhorse. While Intel has remained relatively quiet on Larrabee as of late, NVIDIA's increased aggressiveness towards its Santa Clara neighbors is making more sense every day.

We already know that Larrabee will be built on Intel's 45nm process, but given the level of performance it will have to compete with, it wouldn't be too far fetched for Larrabee to be Intel's first 1 - 2 billion transistor microprocessor for use in a desktop machine (Nehalem is only 781M transistors).

Intel had better keep an eye on NVIDIA as the GT200 cements its leadership position in the GPU market. NVIDIA hand designed the logic that went into much of the GT200 and managed to produce it without investing in a single fab, that is a scary combination for Intel to go after. It's not to say that Intel couldn't out engineer NVIDIA here, but it's just going to be a challenging competition.

NVIDIA has entered a new realm with the GT200, producing a world class microprocessor that is powerful enough to appear on even Intel's radar. If NVIDIA had the ability to enable GPU acceleration in more applications, faster, then it would actually be able to give Intel a tough time before Larrabee. Fortunately for Intel, NVIDIA is still just getting started on moving into the compute space.

But then we have the question of whether or not you should buy one of these things. As impressive as the GT200 is, the GeForce GTX 280 is simply overpriced for the performance it delivers. It is NVIDIA's fastest single-card, single-GPU solution, but for $150 less than a GTX 280 you get a faster graphics card with NVIDIA's own GeForce 9800 GX2. The obvious downside to the GX2 over the GTX 280 is that it is a multi-GPU card and there are going to be some situations where it doesn't scale well, but overall it is a far better buy than the GTX 280.

Even looking to the comparison of four and two card SLI, the GTX 280 doesn't deliver $300 more in value today. NVIDIA's position is that in the future games will have higher compute and bandwidth requirements and that the GTX 280 will have more logevity. While that may or may not be true depending on what actually happens in the industry, we can't recommend something based on possible future performance. It just doesn't make sense to buy something today that won't give you better performance on the software that's currently available. Especially when it costs so much more than a faster solution.

The GeForce GTX 260 is a bit more reasonable. At $400 it is generally equal to if not faster than the Radeon HD 3870 X2, and with no other NVIDIA cards occupying the $400 pricepoint it is without a competitor within its own family. Unfortunately, 8800 GT SLI is much cheaper and many people already have an 8800 GT they could augment.

The availability of cheaper faster alternatives to GT200 hardware is quite dangerous for NVIDIA, as value does count for quite a lot even at the high end. And an overpriced high end card is only really attractive if it's actually the fastest thing out there.

But maybe with the lowered high end threat from AMD, NVIDIA has decided to make a gutsy move by positioning its hardware such that multiGPU solutions do have higher value than single GPU solutions. Maybe this is all just a really good way to sell more SLI motherboards.

Originally published at https://www.anandtech.com/show/2549 (Wayback Machine).