The following is copied and pasted from the comments on
http://www.informationweek.com/windows/showArticle.jhtml?articleID=201202009
Peter Kogge was forced out of IBM for demonstrating that memory chips can make use of the thousands of bits that are actually read in every memory cycle if processors are built into the memory chips. Even if only 8088 type processors were built into a chip that also had the full adressable 1 megabyte of memory and operated at current processor cycle speads, the computing, and gaming capacity of the billions (Carl Sagan) of transistors in modern PCs would be beyond the imagination of the most avid gamers. The hundreds of processors that could be built on a single chip could be connected by data paths that resemble the operations of data paths of nerve cells in the brain. Every nerve cell in the brain has hundreds if not thousands of connections to other cells and a few giga-bytes of data storage. USB might not be a bad starting design, after all, with user programmable processors and RAM with the 4 gigs of eprom. There could even be an 8088 section of the thumb processor that could run DOS 3.1 and WordPerfect 4.7(After all, WordPerfect was used with most of the DOS 3.1 units.) right on the "ThumbDrive". Vista could connect the screen and keyboard in a few seconds to the virtual screen and keyboard in the RAM of the THUMB computer. A much simpler program running under Caldera Dos could do the same in a few milliseconds. Such a system would do almost all of the word processing that is now done on super-pentiums with 4 gigs of ram.
Interesting. I've been saying for years that instead of just making ever-faster processors that can access ever-larger memory spaces, we ought to try making some processors with onboard memory (think 'cache' if you must), which can then be connected into clusters. Setting up DMA channels between the processors. as well as pipeline architecture to let one chip stream its output to the input of the next, would make for some really powerful configurations.
On 8/6/07, David Nicol [email protected] wrote:
The following is copied and pasted from the comments on
http://www.informationweek.com/windows/showArticle.jhtml?articleID=201202009
Peter Kogge was forced out of IBM for demonstrating that memory chips can make use of the thousands of bits that are actually read in every memory cycle if processors are built into the memory chips. Even if only 8088 type processors were built into a chip that also had the full adressable 1 megabyte of memory and operated at current processor cycle speads, the computing, and gaming capacity of the billions (Carl Sagan) of transistors in modern PCs would be beyond the imagination of the most avid gamers. The hundreds of processors that could be built on a single chip could be connected by data paths that resemble the operations of data paths of nerve cells in the brain. Every nerve cell in the brain has hundreds if not thousands of connections to other cells and a few giga-bytes of data storage. USB might not be a bad starting design, after all, with user programmable processors and RAM with the 4 gigs of eprom. There could even be an 8088 section of the thumb processor that could run DOS 3.1 and WordPerfect 4.7(After all, WordPerfect was used with most of the DOS 3.1 units.) right on the "ThumbDrive". Vista could connect the screen and keyboard in a few seconds to the virtual screen and keyboard in the RAM of the THUMB computer. A much simpler program running under Caldera Dos could do the same in a few milliseconds. Such a system would do almost all of the word processing that is now done on super-pentiums with 4 gigs of ram.
-- Prioritize based on common sense? Is that some kind of joke? _______________________________________________ Kclug mailing list [email protected] http://kclug.org/mailman/listinfo/kclug
On 8/6/07, Monty J. Harder [email protected] wrote:
some processors with onboard memory (think 'cache' if you must), which can then be connected into clusters.
I got the impression (from what I quoted) that the heresy was to suggest "memory with an onboard processor" instead of "processor with onboard memory."
On 8/6/07, David Nicol [email protected] wrote:
On 8/6/07, Monty J. Harder [email protected] wrote:
some processors with onboard memory (think 'cache' if you must), which can then be connected into clusters.
I got the impression (from what I quoted) that the heresy was to suggest "memory with an onboard processor" instead of "processor with onboard memory."
What's the difference?
On 8/6/07, Monty J. Harder [email protected] wrote:
On 8/6/07, David Nicol [email protected] wrote: "memory with an onboard processor" instead of
"processor with onboard memory."
What's the difference?
marketing emphasis; is the new chip marketed as a cpu or a memory chip? The separation between cpu and memory chips is so ingrained in the industry that MWOP consititutes a disruptive innovation, while PWOM is what we have now, with processor cache serving to optimize memory access (and complicate SMP.)
MWOP architecture would be like a NUMA node, whereas PWOM is a CPU with a cache.
-
David Nicol -
Monty J. Harder