Hacker News Reader

Comment by TheCoreh | Hacker News Reader

Hopefully this partnership will allow Qualcomm to bring the Snapdragon performance on par with the Apple CPUs.

This is unlikely. Apple's CPU budget for it's devices is likely much higher than anything Qualcomm's customers are willing to pay (vs the current snapdragon pricing). Apple's CPUs are very very large and consequently relatively expensive. As an example, a latest gen snapdragon 835 has 3 billion transistors, while Apple's A11 has 4.3 billion.

And those numbers severely understate the difference since, IIRC, a much larger fraction of Apple's SoC transistor budget is devoted to its application processors.

EDIT: Found an image. More a simple understatement than a severe one.

https://wccftech.com/apple-a10-fusion-cores-bigger-than-comp...

Of course competitors want more powerful options, and they certainly have the budget for it given the pricing of premium Android devices.

Qualcomm simply fell behind. While Apple went for powerful, enormous cores in small numbers, Qualcomm and friends bet on simpler, smaller cores. Apple has started copy/pasting their fantastic cores, yielding the enormous transistor counts.

I admit to being a skeptic when Apple went off to do their own thing, sure that a whole industry (and options like the Tegra) would make them eat their hat. Apple executed and are in a remarkable position now.

Why would that be the case? Do you foresee this putting a great deal of performance-related pressure on Qualcomm that wasn't there before, despite them being a top-shipping smartphone CPU vendor?

Given Qualcomm's just-announced partnership with AMD [1], Qualcomm is clearly pursuing multiple strategies to compete with Intel, but they can leverage AMD to cover their weaknesses, instead of trying to outpace Apple's A-series in places where the market seems fine with the current Snapdragons.

[1] https://news.ycombinator.com/item?id=15855188

Well, Apple and Qualcomm have taken different approaches. Having a (successful) desktop/laptop product will cause them to make architectural decisions which don't make sense on a smartphone in general.

> Do you foresee this putting a great deal of performance-related pressure on Qualcomm that wasn't there before

Possibly. Although I don't doubt Apple may have superior expertise in designing ARM chips (which sounds strange when comparing to the #1 ARM chip maker), I think one of the main reasons Qualcomm hasn't been able to match Apple is that Qualcomm has to optimize its chip for a variety of OEM needs.

Apple on the other hand, could even push the TDP by 20% on its chip, if it know that it can make up for that loss of efficiency with more optimized software that's developed specifically for that chip.

The AMD partnership is very interesting. I hope they double down on that.

Fun fact: Qualcomm wanted to buy AMD in 2015, but because of some terrible management and engineering choices with the Snapdragon 810, which led to a sales slump that year, the board was so furious that i wanted to sell Qualcomm's chip business! (and sell just modems).

Of course, that would have been a terrible idea, business-wise, which is why if Broadcom buys Qualcomm I hope it does get rid of that board and Qualcomm's CEO from day one.

Another fun fact: Qualcomm's Adreno GPU is AMD's old Mobile Radeon GPU division, and Adreno is an anagram of Radeon.

I've been hoping someone like Qualcomm or Samsung buys AMD. Perhaps it may actually do that eventually, with or without Broadcom's help. Then we can see some real competition against Intel, and AMD alone simply lacks the funds to compete head-on with Intel in many areas.

Also, now I'm starting to wonder if Broadcom knew about this deal and it's why it wanted to buy Qualcomm now.

> I think one of the main reasons Qualcomm hasn't been able to match Apple is that Qualcomm has to optimize its chip for a variety of OEM needs.

This just makes no sense at all. So you are saying some OEM's want slow chips? No, of course not. Or are you saying that Samsung S8 has different needs to LG V30 or to the Google Pixel? No, of course not. There needs are all the same.

Smartphones and tablets are not Qualcomm’a only market by far. They have plenty of other customers, too.

Why would this partnership have any significance on the performance of Qualcomm SoC's? Qualcomm doesn't really drive the performance of the architecture they license - that would be ARM. The next generation Snapdragon, Exynos and Kirin SoC's will all use ARM's new A75 and A55 core dynamIQ reference designs and their performance will be the same.

The A11 chip is insane for a tablet. Would love to see a Surface Pro with this performance.

The i5 and i7 of Surface Pros absolutely melt A11's performance. The A11 would have run laps for battery life though.

Not really according to geekbench.

a11: https://browser.geekbench.com/v4/cpu/5286336

i7 (rmbp): https://browser.geekbench.com/v4/cpu/5285899

4.1k vs. 4.8k

Interesting that the multi-core score is a lot lower. Probably you're seeing thermal management in that multi-core score, meaning a macbook with active cooling should be able to reach much higher multi-core numbers.

The A11 is way too overpowered for a phone or tablet. Apple is clearly in the awkward phase where they've almost gotten the right CPU to put in their laptops. The A12 will be a no-brainer. Same performance as i7, way better battery life, way cheaper to produce.

I think hyperthreading with the i7 also gives it an edge with multi-core

Note that the A11 has two "big" cores, and four much slower but power efficient "little" cores. If they were making an A11 variant for desktop use they would have the power profile to use more big cores.

I think we are headed into the world of multi-cpu world. I see the ARM being helped by a CUDA power efficient co-processor with I/O similar to Power9. The issue is our tools as developers wouldn't be able to take advantage of this for a while. 95% of people's applications don't take advantage of multi-threaded processors.

I would love to see a rendering benchmark. This is only single thread comparison and multi-thread isn't close.

Are you guys serious? Not saying A11 isn't fast, but come on - this is beyond ridiculous.

Raw performance is nowhere near intel chips.

https://www.pcworld.com/article/3006268/tablets/tested-why-t...

Although that article is from Nov 2015, using the A9X. Would like to see an updated test, same benchmarks.

It was ridiculous but I was still surprised how strong the A11 is getting.

Apple's A11 (single thread) 4217 (multi) 10164 https://browser.geekbench.com/ios-benchmarks

Intel i7 8700k (single thread) 6089 (multi) 26654 https://browser.geekbench.com/processor-benchmarks

4.1k vs. 4.8k

The A11 is running at 2.4Ghz. The i7 in the single-thread test at 3.1Ghz (actually up to 4.1Ghz). Cycle for cycle the A11 is faster than the fattest Intel chip.

Which is insane. I am no Apple booster, but this is quite an incredible switch up. With a fatter TDP budget (the A11 has a TDP of something like 5W, the i7-7920HQ a TDP of 45W) they could not only increase the clock ceiling, they could start copy-pasting the big core.

A few years back people were talking about Apple starting to move their chip designs to their laptops and desktops. It sounded insane. Now it's actually entirely possible if they can get their software stacks and targeting all working.

Reply on news.ycombinator.com