NVIDIA Tesla

The NVIDIA Tesla is a fairly interesting computer from NVIDIA, whom are famous for their graphics chips. The Tesla is a supercomputing workstation that combines a number of graphics chips and a regular processor. The result is a remarkable desktop machine with the total computational power of 933 GFlops at under US$ 10,000.

While the computer is being touted towards the scientific and engineering community who have to deal with complex simulations, the more exciting application for this computer is – games! As a comparison, the PS3’s Cell processor has a computational power of 204 GFlops while the Xbox360’s Xenon processor has a computational power of 115 GFlops. All these Flops are good for graphics, audio and physics processing.

The way that the Tesla is configured, it comes with a quad-core processor from either AMD or Intel, which serves as the main processor, just like the PPE on the Cell. This main processor will be in charge of running the operating system and allocating tasks to each graphical processing unit (GPU), just like the SPE on the Cell. There are up to 4 GPUs in each Tesla with 4Gb of memory each. You can think of it has a more powerful Cell computer.

Considering that it is based on a regular PC design, it should be fairly trivial to create games for the Tesla. If only it did not have the price tag that it has, it could fairly well be the most powerful gaming platform on the planet, but now it will just need to be satisfied with being the most expensive gaming platform on the planet.

I reckon that is why it is being touted as a scientific computing platform. Although there are efforts at bringing scientific computing to the PS3, for example, it is kind of a hack. The gaming consoles were not designed for scientific computing and the people who wish to use them for it will have to jump through a few hoops first. With the Tesla, the platform is ready for them.

Ultimately, this will be a niche product as Joe Blow will not be shedding ten grand for such a rig. Also, the power consumption of such a system is fairly high (at 1.2kW). However, it would still be interesting to see someone hook a bunch of these machines together into a compute farm. It would be interesting to see the Tesla enter the supercomputing Top 500 list.