Hacker News Reader

Comment by neopanz | Hacker News Reader

Love the guy and his passion for its chips, but this 144-computer chip looks like a solution in search of a problem: what is trying to solve? How's the inter-core communication handled? On the other hand, creating weird chips like these, just because you can, is awesome and stimulating a hacker's mind.

View on HN

Some of the industries greatest advances have began as a solution in search of a problem, such as the microprocessor (which its inventor, Intel, didn't think much of compared with memory chips where the real money was). Most fail of course.

If someone can make many-core, in this form, do something useful that can't be done elsewhere (unlike DSP and GPUs, which are already many-core), it will fundamentally upend computing.

> greatest advances have began as a solution in search of a problem

.. and the laser(!)

"Between nodes there are 18 bit, bidirectional parallel buses called Comm Ports. The handshaking on these ports is trivially simple and the latency and jitter are low, on the order of gate delays, tens of picoseconds. When a node is reading or writing a Comm Port, it proceeds with no delay if its partner is ready to transfer data. If the partner is not yet ready, the node waits until it is. While waiting, a node’s power consumption drops, within picoseconds, to nothing but leakage. When the partner becomes ready the idle node resumes its work within picoseconds."

http://www.greenarraychips.com/home/documents/greg/WP002-100...

It appears that CM's intent with the F18A core is to create an extremely low power processor that scales the number of cores to the complexity of the problem being addressed -- look at the reference sheets about what the requirements are to maintain core states.

The design makes more sense when you stop trying to think of it as an ARM or AVR competitor, and more like a replacement for FPGA's.

> what is trying to solve?

Don't picture each core running C programs. Think the whole chip replacing something you would use an FPGA for, but more flexible, capable of adjusting its own behavior according to the environment.

Not sure if it's a legitimate "problem" or not, but how about the availability of single chips, or 10-packs?

Assuming someone wanted to produce a prototype board, but didn't want to purchase 1000's of CPU chips, are there many CPU chip makers who offer low quantities like this?

Reply on news.ycombinator.com