0

I've been reading about different methods of improving processing speed, and speculative processing stands out to me. My question is, could the integration of memristors that are trained by instructions passed through a processor improve speculative execution in a meaningful way, or would the added complexity be too prohibitive to make it useful?

I understand memristors are extremely conceptual, but I feel that talking about them in this context is appropriate.

Edit: I mean utilizing memristors as a physical perceptron, not as just raw memory.

mcchucklezz
  • 101
  • 3

1 Answers1

1

Memristors are basically... memory. More specifically, memory that can sustain itself after poweroff. At a conceptual level, that's what Intel's Optane technology functions as.

Speculative execution happens all day, every day on every modern CPU you use. It's called a branch predictor. It starts executing code that may or may not actually be used going forward. If it is actually to be used, great, you haven't wasted a cycle. If it's the wrong branch, well, back it up and start over.

Why you're conflating these two ideas, I'm not quite sure.

horta
  • 12,738
  • 22
  • 45
  • My understanding of memristors was that they were perceptron gates. – mcchucklezz Dec 01 '17 at 00:13
  • 1
    @script8man You could use a memristor as such. But then you could use a number of logic gates to create a perceptron as well. The only difference would be size potentially. – horta Dec 01 '17 at 00:16
  • I edited the question for more clarity. – mcchucklezz Dec 01 '17 at 00:16
  • @script8man does my previous comment not answer that part? That's exactly what a branch predictor is doing. It's a little perceptron that's trying to figure out the most likely branch you'll take. More often than not, it's correct too. – horta Dec 01 '17 at 00:17