If I were designing a CPU and OS from scratch right now, I'd put in at least a hundred cores. An M1 Ultra has over 100B transistors. That's more than 1000x the transistors of an Allendale Core 2 Duo core from 2007 (which had 170M split among two cores). I could literally fit 100 Allendale cores those suckers on less than 10% of the M1 ultra silicon.
I'd probably have 1-4 fast, modern cores for tasks which don't parallelize. I'd then have a massive number of auxiliary cores for tasks which do parallelize.
I'd also make good use of this to isolate processes, so for example, a rowdy browser tab can't slow the rest of my system down.
> If I were designing a CPU and OS from scratch right now, I'd put in at least a hundred cores.
If you are talking about 100 dead slow cores that lack the soul of a modern processor, well, you don't have to wait or build from scratch, try Xeon Phi. They are dirty cheap on ebay.
> I could literally fit 100 Allendale cores those suckers on less than 10% of the M1 ultra silicon.
That is exactly what they did for the GPU and AI engine part of the M1 Ultra chip. They are just more specialized.
> I'd then have a massive number of auxiliary cores for tasks which do parallelize.
Once you have those tasks suitable for GPU and AI engine removed from your list, you'd be seeing your remaining tasks not that friendly to parallelization in general. Xeon Phi failed for a good reason, people figured that out ages ago.
Unless you spend far too much on more cache per core, you’re going to quickly find that the bottleneck becomes the memory system. Hardly any of these low-priority process that would be scheduled in E cores are actually doing much compute. They’re either blasting memory or I/O, so adding more cores for them doesn’t help much.
If I were designing a CPU and OS from scratch right now, I'd put in at least a hundred cores. An M1 Ultra has over 100B transistors. That's more than 1000x the transistors of an Allendale Core 2 Duo core from 2007 (which had 170M split among two cores). I could literally fit 100 Allendale cores those suckers on less than 10% of the M1 ultra silicon.
I'd probably have 1-4 fast, modern cores for tasks which don't parallelize. I'd then have a massive number of auxiliary cores for tasks which do parallelize.
I'd also make good use of this to isolate processes, so for example, a rowdy browser tab can't slow the rest of my system down.