Hacker News Reader

Comment by dijit | Hacker News Reader

They claimed this with the last generation of chips but upon delivery were underwhelming (per core, unless you were comparing apples to oranges with core counts).

Claim: https://www.macrumors.com/2023/12/18/qualcomm-snapdragon-x-e...

Benchmark: https://youtu.be/fl7pYRfRyfw?si=1npM7NuH0CVUGj2T

Now thats a 14 core Quallcom CPU vs an 8 core apple one (counting performance cores)

Compared to Intel the efficiency is amazing, but it is not touching the performance of the M-series, but it was claimed to be so: https://www.theverge.com/24191671/copilot-plus-pcs-laptops-q...

View on HN

Why would you need to only look at single core performance to compare apples to apples? Qualcomm can fit more cores in the same die for a lower cost (or is willing to charge less overhead, at least), providing a faster multicore CPU. Wirh everything from games to web browsers leveraging multiple cores these days, multi core performance is quite relevant.

Their performance may come at a cost of efficiency but they're beating most of the M series chips when it comes to raw CPU performance. Legacy software running only on a single core will be slower, but I haven't had a CPU core run completely idle while another is on high load for years.

Furthermore, this article is on modems, an area where Qualcomm rules and Apple has a lot to gain. I'm sure Apple will catch up eventually, but they're still catching up at this point and if Qualcomm has managed to apply the lessons they've learned from catching up with Apple in the CPU area, that may take longer now that Qualcomm has reason to be competitive again.

Because most CPU bound tasks (especially in consumer space) are single thread limited.

But you have (a) background tasks, (b) core iOS libraries being all multi-threaded courtesy of libdispatch and (c) AI being more commonplace which often is inferred across CPU and GPU.

Not to mention that Apple has been aggressively splitting up their operating systems into lots of independent processes.

Yes, but you hit diminishing returns pretty quickly there so it ends to coming back to Amdahl’s law. As the other work gets distributed and optimized, what users notice tends to be single core performance. This is quite noticeable on phones where people notice Android browsing performance being slower because that stands out the most – most people don’t talk about things which are fast enough, so you don’t get as much credit for all of the optimized multi core work as you get negatives for the slow single-threaded web experience which visibly runs faster elsewhere.

WebKit is multi-process on iOS i.e. GPU tasks happen in a seperate process.

So again users will notice the benefits of multi-core performance.

I’m aware, but that’s only partially true. Browser developers have been adding concurrency for years but the way the web is designed has inherently single-threaded parts and that’s why benchmarks track single core performance. Apple’s had a lead for over a decade because of this.

Sure webkit is multi-process but how many perf critical threads do you think there are for rendering the single page that user is viewing?

This is about modems, not general-purpose CPUs.

The history of overstating technological claims seems relevant.

Yes, especially with the huge gap between the underlying hardware’s maximum capabilities and what people actually get with shipping software. Qualcomm’s historic focus on hardware has left them losing ground on anything which isn’t a simple driver tweak.

Reply on news.ycombinator.com