Hacker News Reader

Comment by GeekyBear | Hacker News Reader

> I did not expect ARM would take so long to catch up with even the Apple M1 or the same core designs in the Apple A14.

As mentioned in the article, you might want to pay attention to the traditional difference between what ARM promises and what it delivers.

> these launch events have a history of making performance claims that don't align with what actually arrives in consumers' hands

View on HN

Part of that though is that Qualcomm etc tend to skimp versus the recommended cache values.

This. Vendors skimp on cache size because of overall cost. But cache size is just as critical as CPU core when it comes to performance. Apple doesn't sell chips to anyone else, so they have no pressure to have small caches.

Which chip that shipped to customers had performance that did align with ARM's projections?

None, because of bad cache sizes and horrendous software

None, from any vendor who licenses stock ARM cores?

They only license the cores, but they pair them with lower cache sizes.

So you can’t actually find any chips that match ARMs own benchmarks because nobody makes ones at the recommended spec for it.

Sure, but it's worth noting that the Cortex X3 in the Snapdragon 8 Gen 2 is _almost_ at the level of the A14 in single core performance, just not quite there. Even if the performance bump is less than promised, they will still equal it.

It's also worth comparing process nodes.

The Snapdragon 8 Gen 2 is on TSMC 4nm compared to the A14 on TSMC 5nm.

Also, the Cortex X4 spec projections are based on the second gen TSMC 3nm process node that isn't in volume production yet. If the M3 is announced (as rumored) next week at WWDC on the first gen TSMC 3nm, we'll have some basis for comparison (at least as far as projected specs go).

Reply on news.ycombinator.com