Comparison of Apple M1, A14 Shows Differences in SoC Design

This website might get paid affiliate commissions from the links on this web site. Conditions of use.

Ever due to the fact Apple introduced the M1, there have been inquiries about how the new SoC would differ from the A14. When the two SoCs are dependent on a common CPU microarchitecture, the M1 incorporates supplemental on-die performance that the A14 does not, alongside with additional CPU cores in-total and a bigger overall die dimensions.

TechInsights has printed a comparative shot of the two SoC dies:

Apple A14 SoC. Image by TechInsights. CPU 1 = FireStorm, CPU 2 = IceStorm.

Apple M1 SoC. Image by TechInsights. CPU 1 = FireStorm, CPU 2 = IceStorm.

In comparison with the A14, the M1 has 2x as quite a few DDR interfaces, double the amount of superior-functionality CPU cores, double the amount of GPU cores, and the L2 cache backing up Apple’s Firestorm cores is 1.5x bigger than the A14 equal. The smaller sized IceStorm cores use the exact-sized L2 in the two A14 and M1. The NPU is also equivalent among the two chips.

The M1 has 25 % less overall technique cache than the A14, according to TechInsights, and its overall die dimensions is 1.37x bigger. The greater die dimensions is typically pushed by two aspects. First, the M1 boosts the amount of CPU and GPU cores, the L2 cache, and the amount of DDR interfaces. 2nd, the M1 integrates silicon that the A14 does not, like the Apple T2 stability processor, as effectively as aid for benchmarks like PCIe.

According to TechInsights, Apple used about 2.1x additional silicon to implement 2x the CPU and GPU cores observed in the M1. The slight disparity could be evidence that Apple used transistor libraries optimized for functionality fairly than electric power usage with the M1, but there is no evidence as of but to aid this argument, and the M1’s clocks are only modestly bigger than the A14’s.

There has been a terrific offer of speak about the deployment of specialized accelerator blocks inside of the M1, but somewhat minor dialogue of what they are. We know there is an graphic sign processor and Apple’s own Thunderbolt 4 controller, and there will be the common storage and I/O controllers any chip of this kind necessitates. It is possible that Apple even implemented sure software program functions straight into components to make them a lot quicker and additional electric power-economical. This process of working with so-identified as “dark” silicon is something we talked over back again in 2013 as just one way SoC designers could enhance transistor utilization without having blowing up electric power budgets.

It is not obvious if Apple took the principle as far as mapping distinct programs into circuitry, or what functions the organization has mapped to the continue to-unlabeled components blocks. Rosetta 2 is not believed to map to any distinct components functions on the chip, and there is no indication that the M1 is somehow cheating at different benchmarks by using some kind of top secret components acceleration further than that which Computer CPUs possess. The fact that the M1 carries on to execute strongly underneath Windows 10 when as opposed against Surface Pro X also pushes back again against the idea that the M1’s functionality is staying pushed by specialized components blocks.

Added examination perform will likely shine a light-weight on these methods lengthy-expression, but it is exciting to see the curtain pulled back again on Apple’s first homegrown notebook SoC, just one little bit at a time.

Now Study:

Leave a Reply

Your email address will not be published. Required fields are marked *