Intel copying ARM’s big little is just stupid and should not have been done

Getting major Dunning-Kreuger vibes here. SemiAccurate seems 100% convinced that big:little makes no sense in a laptop. And their evidence? "It's not a phone, milliwatts don't matter". Anyone with a little curiosity will immediately have a few followup questions. Is it true that big:little only saves a tiny amount of energy? Is it true that a tiny amount of energy savings in phones is significant enough to do this architecture? Why does the Apple M1 use different sized cores if big:little is such a "dumb" idea? Why does this article not mention M1 at all?

I'm not convinced.

I’m not at all convinced.

The pictures I’ve seen of the CPUs that implement this kind of technology show that even if you split them into equal numbers of CPU types, the little ones take up way less space, approximately 4x less. So, in the space of 5 “BIG” cores, you could instead get 4 “BIG” cores and 4 “little” cores.

This seems like a huge win to me, since each of those little CPUs could do a lot of background tasks that always need to be done, but don’t need to take up a full “BIG” core to execute.

You get a lot more parallelism this way, for relatively little cost.

That seems to me to be a “BIG” win.

I know puns are not well regarded on HN but it is really appropriate that this article was posted on "semiaccurate".

P and E cores make sense (in laptop and desktop) because power consumption can be reduced for tasks that are simply "beneath" real performance core, it's not stupid.

Now Intel was optimistic with regard to support, time will tell if this does not just introduce weird edge cases but right now various software are breaking because of a lack of awareness. This will get better eventually.

Perhaps little cores on a desktop could be dedicated to realtime / low latency stuff. But the overhead of having them available, even just conceptually, seems high. We've had CPU pinning and NUMA partitioning and shit for a while. They're not widely needed so it not a problem they're hard to use well.

Of course being Intel there's likely to be extra layers of insanity to prevent any interesting potential in the idea from ever being realized meaningfully.