Moore's LAW shattered with 500 gigaflop Deskside Super Computer

Moore's LAW shattered with 500 gigaflop Deskside Super Computer

www.nvidia.com

IBM kicked to the curb as NVIDIA unveils a desktop supercomputer that trounces IBM's. NVIDIA® Tesla™ D870 deskside supercomputer is the first to bring a massively multi-threaded architecture to high performance computing (HPC) applications for scientists, analysts and other technical professionals.

[edit on 9-7-2007 by UM_Gazz]

Forget 400 petaflops...bah..doesn't matter if you can utilize the power in standard computing operations....Nvidia has a super desktop ready right now...

www.nvidia.com
(visit the link for the full news article)

[edit on 9-7-2007 by DisabledVet]

Uh...I don't know about "kicking to the curb".

500 gigaflops is pretty nice, but it's 0.5 teraflops.

A petaflop is 1000 teraflops.

So IBM's machine isn't in any danger here.

Also, the Nvidia machine is a vector processor sort of thing, it's not a general purpose computation machine. It's for crunching floating point math problems. That's great if that's what you have to do.

Graphics chips are perfect for this sort of thing. The Navy churns out technical papers on it from time to time. At one point, there was a 3DLabs part (the P10) that was used for sonar processing and got something like 160Gflops...in 2002.

IBM's only does 400 Gigaflops....read the story

[removed quote of entire previous post]
Mod Edit: Quoting – Please Review This Link.

[edit on 9-7-2007 by sanctum]

Originally posted by DisabledVet
IBM's only does 400 Gigaflops....read the story

Sorry - going by what you posted in your second post where you said petaflops.

At any rate, the Nvidia box isn't for general computation either.

If you're just going for a floating point vector or array processor like the Nvidia, you can put one together off the shelf either using ClearSpeed boards or Mercury cell boards.

A lot of sonar and specialty radar analysis requires teraflops of processor power, also NSA buys vector processors for voice analysis type stuff, and DOE for simulations. There are and have been companies that build this sort of stuff for them way before Nvidia.

We did a co-design with another company on a holographic computational engine for a major aerospace company back about 10 years ago, it could do right at 10TF of computation, but it's not quite the same with an electro-optical computational rig as with a general purpose array processor.