We've already told you about the Tesla family, although we only revealed the C870 in our first article on the subject. Next up we have two products, the D870, which Nvidia refers to as a “Deskside Supercomputer” and the S870 which Nvidia calls a “GPU Computing Server”.

The D870 looks very similar to the Quadro Plex VCS units and it connects to your workstation via a PCI Express Gen 2 card. Internally the D870 has two C870 cards and this means that it has a total of 3GB of memory at its disposal and it can in theory offer twice the performance of a single C870 card.

The S870 is quite a different beast, as this is a 1U rack mount with space for four C870 cards. However, as you can see from the picture below, these cards only have large passive heatsinks on them, as the cooling is done by fans in the chassis.

This monster has an 800W PSU built into it and it would connect straight to your HPC racks and you can attach multiple S870 systems to one server as long as it has enough slots for the Gen 2 cards. Each S870 gives you a total of 6GB of memory and no less than 512 stream processors.

The D870 is expected to retail for US$7,500 and the S870 for US$12,000, hardly pocket change. All the Tesla products will work with Windows XP as well as 32 and 64-bit versions of Linux. A special C compiler is also available for improved software performance with the Tesla products.

You can find out more about the D870 here and the S870 here