14 DAY TRIAL // You've probably heard a lot of things about Nvidia, Ati and GPGPU. The idea of using the power inside the GPU for other purposes than 3D rendering is not a new one but only Nvidia's 8 series and lately Ati's X1900 and HD 2900XT series are completely programmable. AMD's Fusion is also a known subject but for the moment (at least), nothing new has been released by the giant IC manufacturer.
However that doesn't mean Nvidia stood still while AMD/Ati tried (unsuccessfully) to merge a CPU and a GPU into a single chip. As a matter of fact, Nvidia's CUDA compiler opened a doorway towards General Purpose GPUs and once you've started such a process, there's no going back.
Yesterday NVIDIA revealed that its GPGPU project is actually more advanced than anyone (AMD included) would think. Earlier this year the company announced its CUDA compiler but up until now they still lacked a capable GPGPU platform. Until now.
The GPGPU product line will be known as Tesla. Tesla is a complete product that incorporates a GPGPU card, an external interface and the needed PSU. Currently Tesla has three versions: C870, S870 and D870.
The D870 is an internal PCIe solution and consists of an output-less 8800 card (PCIe x16 interface). The Tesla D870 is currently NVIDIA's only internal GPGPU card that can be used inside a classical PC. The card consumes about 170W, uses two PCIe power connectors and can deliver about 518 Gigaflops of processing power.
The Tesla C870 GPGPU server is based on the original QuadroPlex design and packs two GeForce 8-series GPUs in an external system. The GPGPU subsystem delivers one Teraflop of computing power while consuming up to 550W. Last but not least, the Tesla S870 comes equipped with four GeForce 8-series GPUs and can produce a peak processing power of about 2 Tflops while consuming about 800W. The GPGPU array has a standard 1U form.
Tesla C870 and S870 systems connect to workstation systems through a PCIe Gen2 x16 interface. The machines contain PCIe switches and can be linked to form very efficient multi GPGPU clusters due to the fact that GPUs are far more efficient when working with parallel computing than classical CPUs. The Tesla S870, D870 and C870 have an announced MSRP of $12,000, $1,499 and $7,500 per unit.
We are just a few, but there are many of you, Softpedia users, out there. That's why we thought it would be a good idea to create an email address for you to help us a little in finding gadgets we missed. Interesting links are bound to be posted with recognition going mainly to those who submit. The address is 
.
