I’m always going to be excited by better competition.
In 2019, the mobile landscape was dominated by Qualcomm and Apple SOCs at the premium tier of phones and tablets. HiSilicon’s Kirin was facing issues due to political pressures placed on Huawei. MediaTek maintained a respectable reputation as a mid-range and entry-level chip provider.
We didn’t have much choice in what powered more expensive gadgets.
Flash forward two years, we’re starting to get some insight on what will power devices in 2022, and this market is a LOT more exciting.
MediaTek took the wraps off a premium tier competitor. We’re about to hear more from Qualcomm’s Snapdragon team. Apple’s A15 is solidly refined over the A14. Google launched their first custom chip with the Pixel 6.
A stale landscape has exploded, and more companies are rumored to be designing their own solutions over the next several years.
When it comes to pocketable portable performance, we just have one small issue…
Power.
I’m using the MediaTek announcement as a guide to engage in some not-too-wild speculation, but pointing to some trends that might be concerning for next gen powerhouse phones. I think it’s also fair to highlight some of the work Apple accomplished moving from the A14 SOC in the iPhone 12 to the A15 in the iPhone 13.
There are some lessons we all could be learning, and hopefully these manufacturers won’t have to learn every lesson the hard way.
MediaTek revealed their newest System on Chip, the Dimensity 9000.
An SOC is literally that, an entire compute system bundled on a chunk of silicon. It’s nerdier to talk about, but these collections of components have gotten a LOT more complicated than just a “CPU”. We’re increasingly improving designs. For several years now, these mobile parts have delivered laptop-grade power for portable electronics.
Apple’s newest laptops are powered by larger versions of this style SOC now, further reinforcing the power to performance advantages, and also maintaining tighter control over software and hardware optimization.
As these parts get more powerful however, we still need to contend with traditional computing issues like heat. A phone with a screen, battery, and camera components is a densely packed little gadget. It will always be a challenge delivering more power and getting heat OUT of a phone.
A more powerful SOC can accomplish tasks faster and play games more fluidly, but more powerful components can generate more heat and put more strain on a battery.
There are manufacturing considerations that help make an SOC more efficient, and over time we HOPE to see more powerful chips that ALSO draw less power and generate less heat. It’s a delicate balancing act. Chip manufacturers invest billions in manufacturing equipment, and these mobile chips are planned out years in advance.
Consumer trends can change over that course of time, and the delivery of these compute products also need to line up with other tech industries. A new SOC needs to include new radios for Bluetooth, WiFi, and cell phone data. A new SOC needs to include hardware to control increasingly complicated camera hardware. A new SOC needs to improve on “auto” compute features like AI and Machine Learning services. Faster screens. Faster charging standards. Better security. Better GPS. There are a LOT of moving parts.
Year over year, there’s no guarantee that a new SOC will truly represent an improvement over an older SOC for every feature included in a phone. Often, we need to balance an improvement in one area against a compromise somewhere else.
2021’s Snapdragon 888 is a perfect example of those concerns.
A brief history.
In 2015, numerous phones launched using the Snapdragon 808 and 810. These chips were touted as being significantly more powerful than the previous generation of Snapdragon SOCs from 2014.
The 810 featured a BIG/little octa-core CPU.
Eight cores, but four cores were “BIG” and four cores were “little”. The “BIG” cores were more powerful and more power hungry. The “little” cores helped manage less intense tasks and were supposed to reign in power draw.
It kinda didn’t work great.
Many phones from 2015 suffered inconsistent performance, and ran noticeably warmer to the touch than phones from 2014. Manufacturing issues arose for several phones from that higher thermal profile, resulting in component failure.
Qualcomm significantly altered the manufacturing and design for the following Snapdragon 820, and greatly improved on those thermal issues. From 2016 on, Qualcomm enjoyed a string of successes. From the Snapdragon 820, we saw roughly five generations of SOC where performance improved within a reasonable envelope of power draw and heat.
In 2021 Qualcomm delivered the Snapdragon 888.
This was a notable release, as this SOC included a new kind of CPU core and a more powerful GPU. Intended to be a more significant step up in performance, the 888 benchmarked well. We saw “BIGGER SCORES” out of testing apps. Unfortunately, we also got reports of phones that ran hot to the touch, and inconsistent sustained performance.
As we close 2021, the devices that have performed more consistently over real world apps and games have all included SOME kind of throttle or management to “turn down” the power of this SOC.
These mobile chips are exciting sprinters in a burst of activity, but can run into issues for longer tasks. Gaming performance looks fantastic in short interactions, but the full draw of the GPU in this SOC can drain batteries quickly. The GPU can also run so warm, that a phone likely won’t be able to sustain faster frame rates after short periods of time.
Apple’s SOC Challenges…
So far we’ve been focusing on Qualcomm, as they’ve been the primary supplier of premium tier SOCs in the Android space, but Apple also recently had to deal with similar issues in their A series of phone SOCs.
Moving from the iPhone 11 to the iPhone 12, we saw similar boasting of improved CPU power and better gaming performance.
In synthetic tests and in short gaming sessions, that proved true. The iPhone 12 was more powerful than the iPhone 11.
However, in more sustained tasks and longer game-play sessions, the iPhone 12 could often heat up and throttle down performance to a lower sustained rate than the iPhone 11.
Play for a couple minutes, the iPhone 12 was a champ. Play for a little longer, the older iPhone could sometimes outperform the newer iPhone.
Moving to the iPhone 13, Apple’s claims were far more modest. The newest Apple SOC was claimed to be faster, but improvements weren’t pitched as dramatically. The initial reaction was cynical, and many figured Apple was “slipping” or having an off year.
The A15 in 2021 is an interesting SOC though.
While not dramatically more powerful, it seems to deliver a better balance of performance and power draw. We still see some significant slipping in longer game-play sessions, but the iPhone 13 doesn’t slip below previous generations of iPhones like the iPhone 12 did.
From an SOC standpoint, the iPhone 13 isn’t as exciting an improvement, but it’s better refined.
This brings me back to the Dimensity 9000 announcement.
MediaTek is showing off an SOC design influenced by the bleeding edge of ARM hardware. It’s a similar arrangement of components found in the Snapdragon 888 and the upcoming Snapdragon 898 SOCs from Qualcomm.
The “CPU” is comprised of three different kinds of CPU cores. We’ll be getting a new “BIGGEST” CPU core called the Cortex-X2. We’ll be getting refined “BIG” CPU cores, and we’ll be getting more power efficient “little” CPU cores for less intense tasks.
MediaTek published numerous marketing slides comparing performance, battery life, and power.
The comparison most concerning to me doesn’t actually come from MediaTek though. I’m most worried about this marketing slide from ARM showing the difference in power from the Cortex-X1 core compared against the Cortex-X2.
This is the “BIGGEST” CPU core. It’s the fastest and most powerful, but also the most power hungry.
We can’t use this graph to perfectly detail an oranges to oranges comparison between the Dimensity 9000 and the Snapdragon 888. Mobile computers are WAY too complicated to be compared this simplistically.
However, it illustrates one performance claim I find concerning.
From the company that designs these CPU cores, ARM is claiming a 16% improvement in performance, but also a higher peak power draw. I’m always excited about better performance, but I’m concerned if that performance drains batteries faster and generates even more heat. I’m putting a LOT of concern on this one line graph, but it vibes with a trend we’ve seen play out in this market over the last two years.
When we look at the above graph, I hope phone manufacturers see what I see.
ARM’s chart details modestly improved performance at the same power draw, AND it shows the same performance at a lower power draw. That’s REALLY good.
The Cortex-X2 peak performance line should be concerning to us after a year with phones that already can not sustain their peak level of performance for long before getting too hot.
I’d be worried that the Cortex-X2 will get hotter faster than the Cortex-X1 at peak performance. In which case, the phone will need to throttle down faster, and it’s even less likely we’ll see tangible improvements to real-world applications and games.
Techies are in a LAME benchmarking arms race…
A lot of noise gets made about these simplistic number scores which help us grade performance. However, these scores rarely predict which phones will perform better in real-world tasks. Most synthetic benchmarks are a series of TINY performance tests that spike peak performance, then have a brief recovery, before moving on to another quick spike of activity.
No one uses their phone like that with millisecond transitions between apps.
While this is some NERDY data to chew through, it’s not difficult conceptualizing this idea for your family and friends. In a simple conversation, ask the people around you if they would want their current phone to be %15 more powerful in gaming OR if they would want 15% better battery life for how they currently use their phone today. I think we already know how MOST people are going to react to that question.
Unfortunately, the last two years of SOC design have somewhat catered to this techie notion of “grading” performance. The Snapdragon 888 is a perfect example of this design philosophy. Phone manufacturers have been accused of rigging benchmarks in years past, but now the companies that design and manufacturer the SOC are delivering chips that are better for very brief interactions.
The SOC is already built to accomplish the biggest benchmark score, but the phone manufacturer often has to find ways of turning that performance down for a better consumer experience.
When I see the above “power to performance” chart from ARM embedded above, I personally get excited.
I’m not excited because the absolute upper limit of performance is going to get better in limited scenarios that will nuke my phone battery. I get excited because I hope phone manufacturers using the Dimensity 9000 or the Snapdragon 898 might follow Apple’s example here.
I hope we get an “off” year for performance gains in 2022. I hope we get phones that perform similarly to phones from 2021, but run cooler and get better battery life.