Slashdot Mirror
Princeton Researchers Announce Open Source 25-Core Processor (pcworld.com)
An anonymous reader writes: Researchers at Princeton announced at Hot Chips this week their 25-core Piton Processor. The processor was designed specifically to increase data center efficiency with novel architecture features enabling over 8,000 of these processors to be connected together to build a system with over 200,000 cores. Fabricated on IBM's 32nm process and with over 460 million transistors, Piton is one of the largest and most complex academic processors every built. The Princeton team has opened their design up and released all of the chip source code, tests, and infrastructure as open source in the OpenPiton project, enabling others to build scalable, manycore processors with potentially thousands of cores.
Relax. In between architectural basis and the relatively low performance, it's insignificant. A few hundred million transistors for a 25 core chip in a day where your stock chip is multibillion in terms of transistor count.
I've been hearing about massive number of cores for years ... the problem however is they are great for demonstrating that you can put a bunch of 'cores' on a chip ... not that they are actually useful for anything.
Connecting 8k of these things together? You've just proven you actually don't understand how the real world does things.
If you have 8 million cores that can add 20 super floating point numbers a second ... thats WORTHLESS because I need to do things other than add two numbers.
If you have 8k cores that can be interconnected ... that must be one awesome bus if those interconnects are useful because the congestion on that bus is going to be insane, oh ... you've got a solution to that problem? funny how that solution kills the theoretical performance
Sorry, but I've heard this stuff so many times over the years that I just get annoyed when some professor tells us about this super awesome CPU he has that is utterly fucking worthless outside of theoretical land.
And by the way, 25 cores is on the tiny side for these silly academic projects.
Blah blah blah I made this awesome processor but it only works for one tiny problem domain that can't even be used for that problem domain because of the constraints on it that allow you to make so many cores.
Not once has one of these things actually been useful in the real world, and I know thats not the point of research but the only reason you list something about so many cores is pure clickbait. No one with a clue believes you've built something useful when you make such ridiculous statements.
No, I didn't read the article. I don't have to. These papers are only about getting grant money by making ridiculous statements, not about producing anything useful and 9 times out of 8, its done using methods that the real world (read people who actually get shit done) has already deemed don't actually work outside of academia and theory.
Yes, I'm bitter. I hate useless people wasting money that could be spent doing real things, not reiterating something intel and amd knew in the 80s.
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
Chances are you're not content to watch video in 240p anymore.
That is not really true. Most workloads can be executed in parallel. Pretty much all the field of scientific computing (would that be physics, chemistry, or biology) are typically quite parallel. If you are looking at database and data analytics, they are very parallel as well, if you are building topic models of the web, or trying to find correlation in twitter post, these things are highly parallel.
Even on your machine, you are certainly using a fair amount of parallel computing, most likely video decompression is done in parallel (or it should be). It is the old argument that by decreasing frequency you can increase core count in the same power envelop while increasing performance.
For sure, some applications are not sequential. Most likely, they are not the one we really care about. Otherwise, hire me, and I'll write them in parallel :)