The design of this is in one sense fascinating. He has essentially thrown out all conventional wisdom of how processors should work, and designed his own system

The design of this is in one sense fascinating. He has essentially thrown out all conventional wisdom of how processors should work, and designed his own system completely from scratch, including the design tools used to design it. If you consider the possible amount of computation this system is capable of in theory, it is actually a lot, even if it uses very little power. Fully utilizing this power is another story. I tried to figure out how to use this chip theoretically for some computational problems. It has 144 nodes (called computers in GreenArray's material), and fast communication in a 2D grid. Each node has essentially only 64 words of RAM + similar amount of ROM. Also, the chip is missing any floating point capabilities. The 64 word RAM basically means that any nontrivial algorithm must be implemented by many nodes communicating. There are simply not enough memory on any core to do much nontrivial stuff. This makes it very much a challenge to do anything nontrivial on it. It only has access to 1 Megawords (2 megabytes) of external memory though one 18 bit data bus. This limits not only the amount of memory that can be used, but it very soon becomes a bottleneck for many algorithms. Now I realize that the low memory per core is one of the keys to the low power usage, but it does have a cost. These limitations ensure that many workloads can't fully utilize the chip because they are either restricted by the amount of RAM or memory bandwidth. It is basically a very unbalanced system for many problems. For example, it could in theory maybe do well on some deep learning problems, but the restrictions to 1 megawords makes it only suitable for rather small models. You also get into similar bottlenecks for many other types of problems. I have not been able to find interesting problems where the capabilities of this chip can be fully utilized. This may be my lack of imagination. The chip does require you to think in a completely different way when writing programs, so it can be a worthwhile exercise, even if it ends up showing that it is not a good fit for you problem.