nVIDIA lands CUDA, "100x faster"

Today nVIDIA casts out its eighth generation school of GeForce graphics cards, and first to natively support Direct3D 10, in the form of the 8800GTX and 8800GTS.

Landing with this otherwise evolutionary launch is a much juicier catch.

Following ATI, in nVIDIA’s own voracious bid to rid the world of the CPU, it has also boosted its own support for performing more generalised calculations on GPUs.

Tagged as ‘CUDA,’ it’s not to be mistaken for the common slang of the edible Barracuda fish, as it stands for the slightly more fishy ‘Compute Unified Device Architecture.’

This unified GPU architecture is only available on the eighth generation of GeForce graphics cards, and, presumably, future generations. Future workstation Quadro graphics cards will also benefit from CUDA support.

The new GPU sports a feisty amount of unified shader units, which are specialised processors that calculate floating point numbers. Going beyond stream computing, CUDA allows all of these unified cores to communicate and share data.

“GPU computing with CUDA is a new approach to computing where hundreds of on-chip processor cores simultaneously communicate and cooperate to solve complex computing problems up to 100 times faster than traditional approaches,” declares nVIDIA.

These cores can either share load through threaded applications or via stream processing “… in specific applications such as imaging,” where the threads don’t communicate with each other.

For the 8800GTX, there are an incredible 128 unified shaders swimming at 1.35GHz, and on the 8800GTS, 96 shader units glide around on a slightly slower 1.2GHz clock rate. For comparison, ATI’s first Direct3D 10 chip, the R600, is rumoured to contain 64 such shaders, which in itself is exciting.

Even though these unified shader processors are highly specific in their function, they will allow for substantial performance boosts in many areas, such as gaming physics, data analysis, and scientific purposes.

Perhaps, now, nVIDIA’s cards are powerful enough to hook support for Folding@Home, like ATI.

To boost development support for more generalised computing on GPUs, nVIDIA also announced what it flaunts as “… the industry’s first C-compiler development environment for the GPU.”

Gone is the mermaid mascot, the new, highly-detailed model, ‘Adrianne’, is a digital representation of fashion/Playboy model, Adrianne Curry, the likes of which one might categorise under ‘uncanny’ or ‘utterly creepy’.

Warm feet worldwide will be miffed at the new line of GTX cards, nVIDIA’s Pat Gelsinger warning “The raw performance and visual effects of the new GeForce 8800 GTX is just going to knock your socks off.”

Foot for a fin, we say. Fishing rights for socks.

Leave a comment

Name:

eMail:

Comment: