Author: kjj 2014-07-04 16:50:00
Published on: 2014-07-04T16:50:00+00:00
The author discusses the possibility of algorithms being better performed in application-specific devices than in general purpose computers. The author believes that if there is such an algorithm, it must perform best on a specific platform, making that platform the de facto application-specific device. The author acknowledges that proving or disproving this theory may be difficult but sees it as very likely to be true.The author also addresses the idea of IO-bound versus memory bound for devices with enough memory. They argue that 20 GB is now trivial and that it isn't until crossing 128 GB that one would require more permission. However, if the IO was to an oracle outside of the device's control and limited in throughput, that centralized oracle would go against anti-ASIC design goals.Furthermore, the author highlights the importance of keeping algorithms simple and ASIC-easy to promote diversity in the marketplace. While a harder algorithm can still become an ASD, it requires more skilled designers and expensive fabrication, which concentrates the ASIC advantage into the hands of fewer people, contrary to the stated goals.
Updated on: 2023-06-09T00:41:45.418443+00:00