Something very interesting about these new N3 Qualcomm chips - assuming that the leaked scores are accurate, is that they allow us to exactly quantify the micro-architectural difference between Qualcomm and Apple. It’s the same process and almost identical performance (A18 only gets ahead in SME-enabled tests), but the A18 is clicked 5-7% lower. That’s the IPC difference.
Well, we can do that today, though. What will be nice is less about the performance in absolute but that the frequencies are similar enough (since IPC isn’t constant, though Intel/AMD guys exaggerate how much they gain at lower clocks and functionally I don’t think we should adjust too much for them if they need high clocks where they have more misses to get good performance), and the node is the same helps but it’ll probably be similar to the Oryon arch on laptops. We’re a ways off from a a sizable (in QC’s position, 8%+) upgrade.
I think power at similar performances be the most interesting, now that it’s the same node, N3E, and we’ll get good power testing in these phones — and there are E Cores.
And spoiler: I think Apple will probably slaughter them. I would love to be wrong but even in the best realistic case Apple has such an incredible lead. You see this where the X4 and 8 Gen 3 on N4P are really doing about somewhere A15 or A14-level ST depending on the test, but at about 25% more power (in of SpecInt and SpecFP, where it’s maybe 5-10% better than the A14 at best).
That’s a significant node refinement worth -22% power or - 5-10% power (N5, N5P) and still plenty of cache along with LPDDR5 being used on the motherboard vs LPDDR4x in the A14/15 cases, and 2-3 years of time.
But then what’s funny is that X4 is still in a league of its own vs AMD/Intel much like Oryon is, and I suspect the X5 on N3E is going to put up Lunar Lake ST at a fraction of the power, same with Qualcomm’s next phone chip (even at a hypothetical 10W, the 3000-3200 of the 8 Gen 4 would easily blow out Lunar Lake which does MTL ST of 2200-2400 at 8-12W AKA half of MTL’s power, per Intel)
So it’s pretty interesting we’ve got Apple, drop, Qualcomm/Arm, bigger drop, Intel/AMD.
Again I think people too often separate it into Apple vs the rest when it’s really three-tier system when you look at anything from IPC to power on similar nodes and area use.
