Sunday, February 14, 2016

Maybe "More Moore" is No Longer the Issue

Perhaps the progress of Moore's Law, based on silicon technology, does end at some point rather near in time.

Perhaps no replacement substrates  (germanium, for example) can be commercially developed. Perhaps no replacement architectures  (optical, biological or quantum) can be commercialized soon, either.

Even in such a dire situation, how much will it matter? It is hard to say. Much computing these days takes place in huge data centers, where heat and energy consumption arguably are bigger problems than processor speed or the cost of memory.

Likewise, end user device preferences arguably are more centered on battery life, weight, device size and aesthetics than raw processing power.

We might be at a point where adapting processing to match the key apps actually will become more important (low power consumption already has become key).

And since today’s smartphones already process as fast as supercomputers used to, it is not clear how much “more value” faster processors provide. That has been true for quite some time in the personal computer space, for example.

Faster connections arguably improve experience and capability more than processor speed or locally-resident memory.

Some of us would still bet on human ingenuity to reignite another round of Moore’s Law advances, though. Still, the fact remains: raw processing speed is no longer the chief constraint on application or device value. Instead, it is human creativity which now is the gate.

No comments:

"Tokens" are the New "FLOPS," "MIPS" or "Gbps"

Modern computing has some virtually-universal reference metrics. For Gemini 1.5 and other large language models, tokens are a basic measure...