Semiconductor industry observation: Since the first Apple-designed smartphone chip (A4 in 2010), the company has been a force to be reckoned with in the field of mobile chip design.
Editor’s note: This article is from WeChat public account “Semiconductor Industry Watch” (ID: icbank) , translated from “macworld” author: Jason Cross.
Since the first smartphone chip designed by Apple (A4 in 2010) came out, the company has been a force to be reckoned with in the field of mobile chip design. A few years later, when Apple used its own CPU design instead of an authorized architecture, the A6’s performance leadership really began to improve.
In the past few years, Apple’s CPU has really been hard to find another opponent. A11 Bionic not only has a custom Apple-designed CPU, but also gives up its PowerVR-based graphics processor for its own custom GPU. In addition, it introduces a neural engine, a custom chip that separates from the CPU and GPU, and focuses on accelerating machine learning calculations.
Since then, Apple’s mobile chips have dominated the smartphone chip space. Maybe the iPhone doesn’t have the fastest mobile phone speed in the industry, but Apple can make sure that no one can have faster CPU, GPU or machine learning acceleration.
The A12 Bionic, which has been around for a year, is still faster than any chip in any Android phone, and the A13 chip that debuted on the iPhone 11 and iPhone 11 Pro is much faster. Here’s why A13 bionic wins again in the smartphone competition.
The process of high-tech 7nm+
Apple always uses the best manufacturing technology available from its foundry partners (almost always TSMC). Today, this means that it is one of the first companies to use TSMC’s next-generation 7-nanometer process. It is similar to the 7nm process of last year’s A12 bionic processor and AMD’s Ryzen 3000 series processor.
The second generation process, sometimes referred to as 7nm+ or 7NP, etches some of the chip layers with extreme ultraviolet lithography (EUV). This means that smaller, tightly packed transistors leak less power.
TSMC’s second-generation 7nm process uses part of EUV technology to create faster, more power-efficient transistors.
In practice, this means having more transistors in the same space, but with higher clock speeds at the same power consumption, or less power at the same clock speed. This is exactly what we saw on the A13 Bionic.
Apple stuffed 8.5 billion transistors in the A13, a 23% increase over the A12. It is estimated that the chip is about 20% larger than the A12’s 83mm2, about 98mm2. As a result, Apple achieved only a small increase in density, but achieved significant improvements in maximum performance and power efficiency.
However, this is not the biggest chip in Apple’s history. The A12X in the new iPad Pro has more transistors (10 billion), an estimated 135mm2, and both A5 and A10 Fusion are over 120mm2.
The fastest CPU will get faster
Apple’s custom CPU design typically delivers faster single-core performance than any competitor’s smartphone chip, although it’s hard to rely on benchmarks when talking about completely different platforms. Multicore performance is also difficult to surpass.
A13 Bionic continues this tradition. Apple said its two large high-performance CPU cores and four energy-efficient CPU cores are 20% faster than the A12.
CPU performance is the best. No Android phones are close to these numbers.
In the real world, you almost never actually see the theoretical maximum performance improvement, but our benchmarks show a CPU performance improvement of about 20% and even more multicore performance.
This is an amazing improvement compared with the same period of last year. For Qualcomm, Samsung and other companies, it is a big challenge to match it.
catch up on GPU performance
If there is a field where a competitor’s smartphone processor can beat Apple, it’s graphics processing performance. Apple’s GPU performance in the real world is the best in the industry, but some cross-platform benchmarks will benefit the latest Android phones.
AThe 13 GPU is still Apple’s custom quad-core design, which Apple calls “metal optimized”, which is 20% faster. It was much faster in our tests. Geekbench’s computing performance using the Metal API on the GPU is 40% faster in both Geekbench 4 and Geekbench 5.
The high-end 3D Mark 5 impact test didn’t improve much last year, but it has made great progress on the A13.
Our 3DMark Fire Strike test speed has increased by 50% to 60% (this is a good measure of the performance of modern high-end 3D games). Earlier 3DMark Ice Storm Unlimited was running nearly 30% faster. In this test, Apple’s chips have far surpassed the competition, and this is only an increase in the score.
The old “Ice Storm” test did not improve much with the “Fire Strike” test, but it is already much faster than other smartphone chips.
If you combine GPU computing performance with game-level 3D benchmarks, Apple can be said to be the fastest overall GPU in smartphones. However, this is a closer prediction than CPU performance, and it is easier to see competitors grab the crown before the A14 launch in the fall of 2020.
Interestingly, we see that the performance of the real world far exceeds the 20% improvement requested by Apple. It’s hard to know what caused it, but I think the memory bandwidth needs to be done more than the theoretical peak performance of memory.
Apple claims that the A12’s GPU performance is 50% faster than the A11, but we haven’t seen any similar improvements, and some tests haven’t shown any real improvements at all. At the time, we suspected that the memory bandwidth limitation prevented the chip from achieving the best theoretical performance in some 3D graphics tests. Perhaps this year, although the speed of the chip is not up to 50%, it can make better use of the available memory bandwidth.
Whatever the reason, Apple’s improvements in GPU performance far exceed the previous promise of 20%.
Higher energy efficiency
Apple said that the power consumption of the A13’s GPU and high-efficiency CPU core is reduced by 40%, while the power consumption of the high-performance CPU core is reduced by 30%, and the power consumption of the neural engine is reduced by 15%.
However, these numbers have a big asterisk. This is an energy saving effect when providing the same performance as the A12Bionic. In other words, when the components of A13 run at the same speed as A12, they consume much less energy. When they run faster, the power saved is greatly reduced.
In the benchmark, the battery life of the iPhone 11 is approximately equal to the iPhone XR (although much faster). In actual daily use, we found that it lasted longer.
Apple said that this part is achieved by setting hundreds of voltage domains to reduce the power consumption when the main components of the chip are not in use, and setting hundreds of thousands of clock gating fields to disable unused logic gates. This sounds impressive, but to be honest, this is a fairly standard part of modern top chip design. Maybe Apple is farther away from competing products in these design elements. It is impossible to know without precise numbers, but this should not be regarded as a special new innovation.
The voltage and clock domains ensure that unused portions of the chip are not wasted any power during this precise microsecond.
Aside from these warnings, the performance of the A13 per watt has an amazing leap than the A12. Despite the similar size of the battery, the same display provides higher performance, but in our Geekbench 4 constant reference battery test, the iPhone 11 has the same lifetime as the iPhone XR. In daily use, the battery lasts longer, which is perhaps the best indicator of all batteries. iPhone 11 Pro has a much longer battery life than the iPhone XS, but this is not only