4 ms·
If I understand correctly, Intel CPUs around that time were super deeply pipelined so you could meet timing closure at such high frequencies. So yeah you can ke
by gaze 10y ago
If I understand correctly, Intel CPUs around that time were super deeply pipelined so you could meet timing closure at such high frequencies. So yeah you can keep pushing the clock-speeds up and up but now a mispredicted branch really shoots you in the foot. Makes sense at the time because people associated clock speed directly with performance. Higher frequency more better. Intel kinda gave up on that at some point and started over with the core architecture design which has a lower clock speed but does more each pipeline stage.
- monocasa 10y agoHaha, yeah P4 had multiple pipeline stages just for propagation delay.
- largote 10y agoThose very deep pipelines are the reason NetBurst was very good at some tasks but not so great for general use. They made branching mispredictions be really expensive as a pipeline flush would result in a large number (~20 IIRC) cycles to be "wasted".