Wednesday, August 12, 2009

The Santa Fe Project, and the University of New Mexico

What happens if Moore’s Law isn’t the only thing that matters? Are Harvard and Princeton architectures the only choices, especially as the world moves increasingly to parallel architectures? Whatever happened to the plethora of alternative architecture that thrived during the 1970’s and 1980’s? If Seymour Cray were alive today, what would he have developed? What is the Linux equivalent for computer architecture?

The “Santa Fe Project” is about challenging everything I learned in undergrad class, and everything that is taught in computer science books. As a matter of fact, there are a fair number of old, rusty, mildew-collecting books at the Santa Fe Institute in New Mexico that do nothing but talk about computer architectures. But they’re all the same, the only thing that ever changes are the number of registers, the bit width of the registers, and the amount of memory available to you.

There are some people that argue that because every computing machine is Turing Complete, it doesn’t make any sense to build anything else, you can “just emulate it all in software”. But that’s an awfully narrow-minded and usually impractically academic mentality. The problem is that software is intangible, it’s electronic. It’s not physical, so it doesn’t occupy meaningful physical space. It doesn’t interact directly with its environment, it doesn’t change its surroundings, and it doesn’t interact with users physically.

What would a computer look like if it was a complex adaptive system that existed in real, 3 dimensional physical space?

Over the next 2 months, Chris, Mike, Omar, Justin, Matt, and Dave Ackley are going to try to challenge everything we know, by intentionally “undoing” computation. The Santa Fe Project will be as much a mental challenge as it will be about physical hardware design and software design.

And while all of this is happening, Dave Ackley is going to host a course at the University of New Mexico, in the computer science lab. Everything we build and develop, and test, and explore, we will launch immediately to the University of New Mexico and at the Santa Fe Institute. In fact, he’s been working on a syllabus, and we’ll talk all about the research and projects here on the blog.

7 comments:

FlorinC said...

I see the skeleton of a new book.

Charles said...

Wow, this is the first time I regret having left New Mexico. Although, I suppose that UNM course would not be open to the general public anyway. I look forward to further posts on this. Thank you in advance for sharing this through your blog.

As far as the study of different computer architectures goes, I found The Connection Machine by W. Daniel Hillis to be an interesting read. I haven't had an opportunity to practice what I learned from that book, but it is definitely a different approach to computing. I found my copy through Alibris. ISBN for the paperback: 0-262-58097-7.

Matt said...

@FlorinC - I think Dave is going to be writing it in real time too, since I'm not much of a "theoretical" writer... want in? :)

@Charles - it's weird... I never knew of New Mexico as the hang out of so many forward thinking guys until I went down there. I serious almost regret living in the northeast... we're way too conventional thinking up here. anyway, i just bought that book, and it should be at my place in a few days. Also, I read a pretty neat writeup about it over here: http://www.kurzweilai.net/articles/art0504.html?printable=1

Matt said...

@jacks ... um... thanks? spam link?

MaysonicWrites said...

Have you taken a look at the ZiiLabs StemCell processors? They look cool as all get out - 2 Arm cores + 24 floating point processors - I just ordered one (embodied in a Zii Egg [q.G.]). $75 (OEM) for a module with processor, 256MB RAM, 16MB Flash (Linux boot kernel ROM), and lots of connections - on a SO-DIMM. One of these, with a pico projector, could be a pocket workstation.

Matt said...

@maysonic - cool, thanks for the reference, i'll check it out. for now, the nxp arm chip is the way it'll be for this round of hardware, but certainly worth checking out others. the nxp is perfect right now for debugging and establishing the basics. plus, the toolchain is open source now, which is mandatory...

Charles said...

@Matt - Thanks for the link (I just got around to reading it). It added a nice human touch to the story of the Connection Machine. Man, Feynman was involved in just about everything it seems.