GT200: Nvidia GeForce GTX 280 analysis

Final Thoughts...

Nvidia's GeForce GTX 280 is an interesting products in many respects and there are some great optimisations that have been made inside the GT200 GPU. The increased performance in the shader core—and in particular geometry shader throughput—is a boon, and there's not a great deal wrong with the architecture itself.

Many will class it as an evolution of the GeForce 8-series, but then the GeForce 8-series has been a huge success for Nvidia so it didn't surprise me to see the architects fix the biggest shortcomings and keep the good bits from G80 and its derivatives. But the question is whether those improvements are enough – in some respects, I don't think they are.

There's no doubt that this is the fastest single GPU solution on the market today and you're going to have to pay a high price to jump on that ship. The GTX 280's perceived value will largely depend on what your feelings are for multi-GPU solutions and, as you know, I'm not a big fan of them personally – they're too limiting for my own usage model. Scaling is often flaky and SLI multi-monitor support is non-existent at the moment.

The problem is that the GeForce 9800 GX2 is available for around £300 (including VAT), while the GeForce GTX 280 is about another £110 more. For a card that's not really faster, that seems like a raw deal to me and it makes me think twice about the 9800 GX2's limitations.


That said, making a compute heavy ASIC with 1.4 billion transistors is an amazing feat, but to be honest I had hoped to see more performance from such a big chip – the GT200's performance is often higher than the 9800 GX2 in the targeted tests we've run, but in many real-world cases it isn't, as we'll show you very soon in our GeForce GTX 280 and 260 gaming performance article.

I can't help but feel this is a strange position to be in with the release of a completely new architecture because, generally speaking, the new generation of hardware completely outclasses everything that's gone before. That isn't the case here and I get the feeling that Nvidia has been a little conservative on the graphics front. Instead, it has focused a lot of resources on improving the GPU's general compute capabilities.

What concerns me though is that Nvidia seems intent on pushing CUDA on its own at the moment, which is fine in many respects, but it's not the only player in the GPU computing market and applications developed using CUDA don't work on other hardware. That limits its appeal to me, unfortunately, even though I do think what Nvidia has achieved with the platform in such a short time is fantastic, as we really need a standard that all vendors support in hardware.

Stay tuned for more on Nvidia's new products over the next week or so, as we dive into gaming performance with single and multiple GPUs.