1. 2
  1. 1

    This seems pretty wild, but the kickstarter page isn’t all that clear on a few things.
    1. What makes this different from a multicore chip? They said in the video that multicore chips won’t work. 2. What’s wrong with GPUs?

    Why hasn’t Intel built something like this? One of the quotes says that they aren’t afraid to let go of x86. Will that mean that it isn’t interoperable with most software? Have they hacked Ubuntu specially for this? What makes this better for exploring parallel computing than a machine with two or four cores? Does it really only use 5 watts of energy?

    This is cool, but the kickstarter page is not that clear.

    1. 2

      I was curious about the same and posted a comment on HN to the same effect. Not sure if it is kosher by Lobster rules, but here is the link: http://news.ycombinator.com/item?id=4702456

      From what I’ve gathered on the subject, it’s better than a multi-core processor because it is explicitly designed to scale. So while the first prototypes have 16 or 64 cores, they are designed so that the chip can continue to scale to 100, 1000, etc cores without serious detriment (unlike current x86 multicore CPUs). Each core is tiny and relatively simple, with a small amount of attached per-core memory. Each core can also access the memory of all other cores (but optimized to access the memory of it’s immediate neighbors). They are general purpose, low power cores.

      It differs from GPU processing in that each core can process it’s own branches, whereas GPU cores must all be processing the same branch or sleeping. I’ve toyed with GPU computing some in the past and it’s awesome if your task is super-parallel. But as soon as you have to start branching just a little bit, many of your cores just go to sleep because they simply cannot handle any type of branching. Either the entire pipeline is processing your set of commands, or none of them are.

      So GPU will still be the king for stuff that is highly parallel, but the Epiphany may be better for more general purpose computing where everything doesn’t fit into a simple parallel pipeline.

      From what I can tell, it really does use 5 watts of energy. And the chips are RISC architecture (as opposed to x86). I’m not sure if they are using stock Ubuntu, but there is a RISC port of ubuntu available. Regarding Intel, I have no idea. They have invested a lot of time and money into x86. The RISC vs x86 debate is a lot like vim vs emacs. Neither is inherently right or wrong, it’s more of a philosophy difference.

      It would also be suicide for Intel to back out of x86 compatible architecture now…it would break all PCs/laptops and they would no longer have a market.