Nvidia drums up revamped Tesla processors

— 8:00 AM on June 16, 2008

As Nvidia unveils its next-generation graphics processor, the company has pulled the curtain on its next generation of Tesla GPU computing processors. For those not in the know, the Tesla line essentially includes GPUs tweaked for general-purpose computing. These products may look and sound like graphics cards, but they lack video outputs, and Nvidia markets them largely toward heavy-duty parallel computing applications.

The new Tesla 10 series is coming out in two flavors. The Tesla C1060 Computing Processor looks like an oversized graphics card, and it packs one T10P chip with 240 stream processors (or "cores" as Nvidia calls them these days), a 1.33GHz core clock speed, 4GB of 800MHz GDDR3 memory, and a 512-bit memory bus. The card slips into a standard PCI Express 2.0 x16 slot, and it has a power envelope of 160W.

Left: Tesla C1060. Right: Tesla S1070. (Not to scale.) Source: Nvidia.

For folks who need racks full of high-powered GPGPU goodness, Nvidia has also introduced the S1070 1U System. As its name suggests, the S1070 is a 1U-form-factor enclosure that packs four T10P processors each clocked at 1.5GHz, with a total of 960 SPs, 16GB of RAM, and 408GB/s of peak theoretical memory bandwidth. Nvidia rates the whole system for 700W of power use, as well.

For an idea of the kind of computing power these devices deliver, Nvidia says each T10P graphics processor has 1.4 billion transistors and can process around one trillion floating point operations per second (teraFLOPS). That's twice as fast as the previous-generation Tesla processor, the company says, and each T10P has twice the precision (64-bit floating-point) with twice as much RAM riding shotgun.

As before, developers will be able to use Nvidia's Compute Unified Device Architecture (CUDA) language to harness the processing power of these new Tesla products. When talking to us, the company was keen to boast that CUDA now runs on 70 million GPUs—including recent GeForce and Quadro offerings—and that programmers who use the language don't have to manage cores by hand. The compiler "basically manages cores for you," letting a single program work on low- and high-end products, the company elaborated.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.