The "high-level CPU" challenge

I think he's got a point that Von Neumann machines get trashed a lot by people who have no better alternative. After designing a processor in a senior level computer architecture class, I think the Von Neumann machine really is the best organization given your building blocks are binary logic gates. Also, a lot of work has been done on instruction level parallelism, caches, and branch prediction that keep the whole processor humming. Hardware is very expensive to design and build, so it's economical to build to a general purpose and widely used instruction set architecture and then optimize that as much as possible. The next best thing is probably massively parallel architectures, but the compilers and programming languages to support those are just not there yet and there is so much code written sequentially. Perhaps when we have some crazy new medium to build processors on (think analog neural nets or quantum computers), a better architecture for those mediums will supplant Von Neumann's.

I'm a bit shocked. Here's an industry-insider article upset about "trashing Von Neumann machines when there's no alternative".

But pensioners who read the New York Times (http://www.nytimes.com/2007/12/17/technology/17chip.html) know the problem is that Von Neumann architecture is already dead. The CBE (the PS3 chip) has 8 cores, Tilera has 64, Intel is working on a 128 core chip, etc.

The question isn't, "Will vector processors prevail?", but, "What are we going to do in the software industry to handle the fact that this is the future."

he could search the literature for existing material on the topic. you can, for instance, read quite a bit about the architecture of the lisp machines. searching google scholar for "tag coprocessor" comes immediately to mind as producing some literature on the topic. i think work has also been done on hardware support for generational garbage collection.

so, random challenge, promising big prizes, from someone who hasn't searched the literature... crank.