Comments by "Anders Juel Jensen" (@andersjjensen) on "A Graphene Transistor Breakthrough?" video.

If graphene can reach hundreds of GHz or even THz then it doesn't matter if power consumption goes up by 50x. The reason the transistor count is so absurdly high in today's CPUs and GPUs is because of the lack of switching speed. You roughly speaking need to increase transistor count by 70% to increase IPC (instructions per clock) by 15%. So a Pentium sized chip at 1THz would run rings around a modern core while still using significantly less power. Of cause, if the transistors can also be made as small (or thereabout) as today's leading silicon nodes, then the area of a Pentium would shrink to a pin-prick.
49
@davidgunther8428 Heat density is a myth. We have been "hitting the heat density wall" for the last 30 decades. And you didn't read what I wrote: If each transistor outputs 50x more heat, but switches 500x faster, you still have a net reduction in heat output because the design will have 200x fewer transistors and still yield a performance boost.
5
@E4tHam Power does absolutely not scale linearly with frequency. The reason laptop CPUs can reach 70-80% the performance of desktop CPUs at 15-25% the power draw should clue you in on that.
5
Sorry, but there are patreon responses from 2 months ago :P
4
@wesleyw.terpstra1902 IPC is IPC, parallel or not. Yes, obviously, if your compute problem responds well to parallel processing you don't go for increased IPC, you go for more processing contexts.
2