OT: Attn: Jeff Fox Re: Rajit Manohar
[email protected] Sat, 10 May 2008 21:20:04 -0600
| Newsgroups | gmane.comp.lang.forth.colorforth |
|---|---|
| Message-ID | <[email protected]> |
Mr. Fox, I was wondering if you were familiar with the processors designed by Rajit Manohar, and if you are, could you briefly give your opinion of them and compare them to the work yourself and Moore et al are working with? I am in a forum dialog that has gone beyond my ability to comment: ----- A: Chuck Moore has, as far as I am able to discern, created such a radically faster and lower power chip that is without peer or competition. G: Why without peer or competition? As far as I can tell, the source of the gains is asynchronous and parallel operation. Caltech computer science has been pursuing these techniques for decades. As I have written for some time, it is becoming mandatory because a single clock pulse can no longer propagate across an entire chip within the switching delay of transistor circuitry. So I am all for asynch--throw away the clock like Easy Rider--but every attempt so far to achieve fully asynchronous computing devices, optimized for speed and low power, suffers rigidities and programming idiosyncrasies that thwart its adoption. Andrew Lines of Fulcrum made a heroic attempt, but Fulcrum is now focusing on networking PHYs, I believe. Moore's SEAForth system seems to entail adopting new architectures, a new language, new testing and debugging methods, and new modes of parallelism. It requires the world to adapt to it rather than it adapting to the world. At the end of the day, the advantages (20 GigaOps with a focus on Fourier transforms that can be performed by other special purpose devices) are not sufficiently overwhelming to enforce acceptance. A better approach comes from Rajit Manohar of Achronix, a Carver Mead protege who built the first asynchronous microprocessor based on the MIPS instruction set and has focused on creating compilers that allow the translation of ordinary ASICs to his asychronous field programmable gate arrays. Rajit and Carver will both be at Telecosm, and if you are intrigued, we might be able to work out a student discount or whatever. GG