Comments on: It’s all about the connections http://neurodudes.com/2005/06/02/its-all-about-the-connections/ at the intersection of neuroscience and AI. Tue, 19 Aug 2008 22:19:29 +0000 http://wordpress.org/?v=2.6 By: David http://neurodudes.com/2005/06/02/its-all-about-the-connections/#comment-662 David Fri, 03 Jun 2005 16:17:36 +0000 http://neurodudes.com/?p=120#comment-662 This measurement ignores that every switching operation performed on a playstation is stimulated by one of two possible values. Equivelent "switches" in the brain may respond differently to an infinite number of values. I think we may need to add a few more variables than just transister count and clock speed before a worthy comparison is made. This measurement ignores that every switching operation performed on a playstation is stimulated by one of two possible values. Equivelent “switches” in the brain may respond differently to an infinite number of values. I think we may need to add a few more variables than just transister count and clock speed before a worthy comparison is made.

]]> By: Bayle http://neurodudes.com/2005/06/02/its-all-about-the-connections/#comment-661 Bayle Fri, 03 Jun 2005 03:43:43 +0000 http://neurodudes.com/?p=120#comment-661 I should add that, as implied by the title of his book, Hillis agreed that “It’s all about the connections”. His solution to the von Neumann bottleneck was to creat computing architectures with lots of CPUs, each with a little memory, and with extreme connectivity. I should add that, as implied by the title of his book, Hillis agreed that “It’s all about the connections”. His solution to the von Neumann bottleneck was to creat computing architectures with lots of CPUs, each with a little memory, and with extreme connectivity.

]]> By: Bayle http://neurodudes.com/2005/06/02/its-all-about-the-connections/#comment-660 Bayle Fri, 03 Jun 2005 03:41:46 +0000 http://neurodudes.com/?p=120#comment-660 neat... so why is this off from Kurzeil's estimate of 2020? I assume it's that you are counting transistors in the machine, and he is counting how many operations the software actually gets to use? Btw, Daniel Hillis made an interesting point in "The connection machine" called "the von Neumann bottleneck", which he credits to J Backus. In the old days, "processors were made out of relatively fast and expensive switching components, such as vaccum tubes, whereas the memories were made of relatively slow and inexpensive components, such as delay lines or storage tubes. The result was a two-part design that kept the expensive vacuum tubes as busy as possible." But today, this is inefficient: in a modern components, most of the transistors (/most of the silicon/most of the wire length) are in memory, which is mostly idle at any given moment (~97%): the "von Neumann bottleneck". This gets worse the bigger we build machines. neat… so why is this off from Kurzeil’s estimate of 2020? I assume it’s that you are counting transistors in the machine, and he is counting how many operations the software actually gets to use?

Btw, Daniel Hillis made an interesting point in “The connection machine” called “the von Neumann bottleneck”, which he credits to J Backus.

In the old days, “processors were made out of relatively fast and expensive switching components, such as vaccum tubes, whereas the memories were made of relatively slow and inexpensive components, such as delay lines or storage tubes. The result was a two-part design that kept the expensive vacuum tubes as busy as possible.”

But today, this is inefficient: in a modern components, most of the transistors (/most of the silicon/most of the wire length) are in memory, which is mostly idle at any given moment (~97%): the “von Neumann bottleneck”. This gets worse the bigger we build machines.

]]>