Beyond Graphics

General purpose graphics chips were the order of the day at NVIDIA's first ever nVision conference. There was another nVision event a few years ago, but that was a rain-swept Italian misadventure that NVIDIA doesn't really like to talk about. Taking no chances with the weather last year saw the Californian sun beat down on the inaugural show in Santa Clara.

Whether it's due to the slowdown in the advancement of PC gaming visuals with so many big titles developed specifically for the stagnant graphics chips in the consoles, we couldn't say, but NVIDIA is definitely looking to prove that the GPU can be used for much more than just making with the gaming pretties. It was therefore keen to talk CUDA, its programming language designed to run C-based apps on the GPU rather than CPU, at every given opportunity. The massive parallel processing power of the new graphics chips is now being harnessed to provide speed boosts on a host of applications.

This competition between components is fierce, ably demonstrated by NVIDIA's insistence on comparing the processor count on current generation CPU and GPUs, the hardware equivalent of apples and oranges.

Speed boosts are being demonstrated in areas like video encoding, though at Computex in June last year Intel went to great lengths to try and prove that you weren't getting the same quality from video encoded on the GPU as you would on an Intel CPU. That might be a losing battle though, when you want to dump a movie onto your iPod we'd rather it were done quickly than perfectly. On a 3-inch display you're hardly going to notice a qualitative gap.

Microsoft sidled on to the GPGPU scene too at nVision with a deep dive by Kevin Gee introducing DirectX 11. Aside from promised speed improvements and richer animations for less memory bandwidth, the extras intrigued us. Particularly the Compute Shader.

This part of the new API is being designed to run general purpose programs on compatible graphics hardware. The key point is that it will enable any DX 11 compatible GPU to leverage the new shader, be they NVIDIA, AMD or Intel. ‘I wouldn't like to say the word 'compete' at all’ said Gee in response to the question of rivaling CUDA. But then he wouldn't at an NVIDIA conference. ‘They're not going to be identical, but we do target all hardware that's in the market. So it depends what your application is and whether that's important to you’. Sounds like competition to us.