Princeton University Researchers Unveil 25-Core Piton Processor

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
This guy was built specifically for data centers. What is most commendable about this chip is how scalable it is—designs can be built that go from a dozen to several thousand cores, and the architecture enables thousands of chips to be connected into a single system containing millions of cores.

It's called "Piton," named after the metal spikes rock climbers hammer into cracks or seams of mountainsides to anchor their position, and it's built after a scalable architecture that could boost processing speed while cutting back energy use. Piton represents several years of research and development by David Wentzlaff, a Princeton assistant professor of electrical engineering and associated faculty in the Department of Computer Science, and his students. It's also a rare thing—Wentzlaff says it's not often that a physical piece of hardware is created in an academic setting.
 
Cool. I love to see innovation. Now, the people who will hate on anything will chime in:

3

2

1
 
so it's scaleable. But I would need more information on how good or bad it is with specifics.
 
GPUs are scalable too and already have thousands of cores. The article doesn't mention ISA so who knows what code will scale with it. I'm glad they're researching stuff, but why not just sell the info to Intel or AMD or ARM and be done with it. Those companies will and can produce actual products in mass.
 
lol @ another university research processor. I predict it will be as successful as the several dozen previous ones offered for commercial use that no one has heard of.
 
lol @ another university research processor. I predict it will be as successful as the several dozen previous ones offered for commercial use that no one has heard of.
And eventually we might get one that competes with current off the shelf hardware, and can actually be put into production. Doesn't mean I expect something to be comparable to an intel desktop CPU, but there are a lot of devices out there that use CPUs and competition in the marketplace is a good thing.
 
GPUs are scalable too and already have thousands of cores. The article doesn't mention ISA so who knows what code will scale with it. I'm glad they're researching stuff, but why not just sell the info to Intel or AMD or ARM and be done with it. Those companies will and can produce actual products in mass.

"featuring 25 modified OpenSPARC T1 cores that operate at a 1GHz per-core clock frequency."

That's why.....can't sell an exclusive x86 license on open source stuff.
http://www.tweaktown.com/news/53619/open-source-piton-cpu-scale-million-core-system/index.html
You think Intel makes processor to advance human kind? Puleeeeeeeze!

Of course Intel could do this, but they ain't doing this for charity.
 
I thought Pitan was a new gpu! [H] nomenclature overrules some university...
 
This is really fascinating. I want to see more regarding the interface between physical CPU's to see how they're doing the offloading and delegation of tasks. Anyone have a link on its writeup?
 
Is it really cheap and really scalable? And by cheap I mean cheap to make and cheap to power and cool and cpaable of a pretty dense machine room footprint.

What is it really good at that other things aren't?

Is it readily available with at least a functioning OS?

Because we have a blue gene sitting in our machine room that went from cutting edge to also ran in a surprisingly short amount of time. So did the last two supercomputing predecessors we had. Unless you have something really big for the things to chew on at once, it turns into a time sharing cluster fuck that a more traditional cluster of servers can handle just as well if not better. The pressures form the commercial sector pretty much ensure that the gap any of these things has over mass market hardware gets eaten up pretty rapidly.
 
Back
Top