The returns are diminishing, but a single digit bump is still a pretty good when your goal is faster single threads.
Adding nothing to save space is obviously not the answer, because that leads to having more slower cores.
Also the topic of the post, the 2-ahead branch predictor, exists specifically to get more ALUs running in parallel!
> apple silicon, where the focus is running super deep speculation
And to make that speculation profitable, the M1 cores have 6 ALUs among other wideness.
The returns are diminishing, but a single digit bump is still a pretty good when your goal is faster single threads.
Adding nothing to save space is obviously not the answer, because that leads to having more slower cores.
Also the topic of the post, the 2-ahead branch predictor, exists specifically to get more ALUs running in parallel!
> apple silicon, where the focus is running super deep speculation
And to make that speculation profitable, the M1 cores have 6 ALUs among other wideness.