[mikepurvis]

But the machines aren't really "faster" in clock speed— for a long time now the gains have been in better and more local caching + parallelism at both the core and instruction level.

[0xWTF]

> parallelism at both the core and instruction level

Which most programs don't take advantage of.

[Legend2440]

Neural networks do, which is part of why they’re taking off right now.