Computers Improving Rapidly

Sun microsystems and IBM have announced petaflop computing systems. IBM's machine has "up to 294,912 cores, or 73,728 nodes." It seems to be tightly integrated, custom hardware. But Sun has achieved similar levels of performance with off-the-shelf technology: "Sun's unique approach to petascale computing combines state-of-the-art technology with system level innovation and off-the-shelf components in an open architecture."

Meanwhile, the NVIDIA Tesla puts a C compiler on a 128-core graphics chip, for half-a-teraflop scientific computing. Per chip. They sell two-chip PCI cards that plug into Windows or Linux PCs, as well as four-chip rack-mount configurations. They claim that their systems are "Scalable from one to thousands of GPUs."

Major hat tip to Sander Olson for all of the above.

And on the slightly more theoretical side, Prof. Uzi Vishkin at the University of Maryland is demoing a multicore computer architecture that's easy to program. In fact, it's so easy, Vishkin is letting high school students play with his system. He says it can scale to 1000 cores.

A couple of days ago, I wrote a column for Nanotech-Now.com where I pointed out that hobbyists have recently gained the ability to do serious simulations of molecular machine components--and this can be expected to accelerate molecular manufacturing. With teraflop processors coming to desktop PCs, and petaflop supercomputers being sold by at least two companies (three, if you count NVIDIA's scaling claims), it looks like computing power is about to take a major step forward.

Chris Phoenix

Tags: nanotechnology nanotech nano science technology ethics weblog blog

Comments

[Hal]

Apparently the Cell processor in the new Playstation 3 video game can be reprogrammed for scientific computing and provides real bang for the buck. With 9 processor cores on-chip and the use of hand-crafted codes it can be much faster than any reasonable home PC. Might be worth getting one just for that, and you get an HDTV disk player in the bargain. Here's an article about an optimized version of the foldingathome protein simulation program:

http://gizmodo.com/gadgets/home-entertainment/ps3s-folding-at-home-is-the-cell-proc-30-times-more-powerful-at-folding-than-a-pc-244528.php

[NanoEnthusiast]

I want an MNT@Home program. I think Nanorex's acquisition of NanoHive would allow for that.

[Tristan Hambling]

According to this article reports Blue Gene/P can be scaled out to 884,736 processors, in a 216-rack cluster to hit the three-petaflop mark. (3×10^15))

http://www.nzherald.co.nz/section/story.cfm?c_id=5&objectid=10448235

and even before its installed the need for even more power .. for...
*Model the Human Brain (2×10^16)=(20 Pflops), or put another way 867,000 PS3's.
* For Mind Uploading. estimated at (2*10^25)
* For Complete Human Systems simulations

Also Brian's article on Zettaflop computing is worth a relook. Ie what will we do with the extra horse power ?
http://advancednano.blogspot.com/2007/03/zettaflop-computing.html