The future of supercomputers?

[tomislavp4]

I know that midle of a racing weekend is not suitable for off topic chat but
I´ll forget what I wanted to say untill monday

http://news.nationalgeographic.com/news ... puter.html

Maybe for our kids hah

[Ciro Pabón]

The "DNA computer" concept was created over a decade ago by Adlerman. As the article says, it's good for some calculations, not for all.

The demonstration I read in the 90's was about the "traveling salesman" problem: find the shortest route that passes through a set of cities, knowing the distance between them.

The idea is that you assign a segment of DNA to the trip between two cities, with a length proportional to the travel distance. You "shake" the molecules in a essay tube and make them bind. This way, you will get many DNA strands, each one representing a complete travel. You filter them and sort them: the shortest strand you find is your answer.

Now, DNA computing, even if faster than silicon based one for certain problems, it's restricted to problems that can be solved as I explain.

Quantum computing, on the other hand, could be a real breakthrough... In 2001 IBM factorized 15 (in 5 and 3) for the first time. It doesn't seem impressive until you realize that with some advances (well... perhaps a lot! ) eventually you could crack secure code, like RSA, some day.

[joseff]

You can do much, much more with plain ol' silicon. Check out top500's current speed champ, IBM's Blue Gene/L. It is a massively parallel cluster of 65,536 nodes, each with 2 low-power PowerPC CPUs and 4 FPUs. The current top speed is 360+ TFlops.

In 2008, IBM will complete Blue Gene/P, proudly touted as the computer to finally to pass the PetaFlops barrier

"Blue Gene is a completely oddball, you've-never-seen-anything-like-this-before design," said Illuminata analyst Jonathan Eunice.

What IBM doesn't like to mention is this:
The RIKEN MDGRAPE-3. Built with "just" 4824 custom MDGRAPE-3 CPUs in 201 nodes, they surpassed the PetaFlops mark in 2006. Each MDGRAPE-3 chip is good for 165GFlops, compare this with BMW Sauber's 9604GFlops for the entire Albert-2 system.

Why is the MDGRAPE-3 off the top500 list? Because its custom electronics can't run the LINPACK benchmarking software.

What I'm getting at is: while future quantum/protein computers are blindingly fast, they're never going to do general purpose computing. If you're designing custom-purpose computers, why not build electronic computers? They're based on existing tech, but can be made orders of magnitude faster than existing systems.

[wazojugs]

i thought the same

[DaveKillens]

Ahh, the future, so enticing, yet so far away. We can't get there today, we have to wait until tomorrow......
I'm one of those who believe that past events predict the future. Electricity was discovered, and for hundreds of years it was just a novelty. The same can apply to quantum theory, which really has not reached the stage where engineers are constructing machines using quantum mechanics. But I believe one day, it will come to pass.
The first electronic computer, the ENIAC, was completed in 1946, just 61 years ago. The first transistor computer was constructed in 1955. So in the grand scheme of things, we're just starting. Heck, when the automobile was 60 years old, that was around the '50's, and we all know how far modern cars have come since then.
So please consider the computer and it's relevant technology as very young, and still immature. There are billions of Euros/bucks devoted to increasing our knowledge in this area, and advances will flow for many years.
But if I was a betting man, I'd back quantum computers, and shared resources. These days, with the speed of communications, it's relatively easy for separate devices to share computing loads.