Apple Silicon for Macs Updated For 2022

Although the event is primarily software-focused, Apple’s WWDC keynotes often set the stage for interesting hardware announcements, and Apple wasn’t disappointed this year. At the company’s largest Mac-related keynote this year, Apple announced the M2, the second-generation Apple silicon SoC for the Mac (and iPad) platform. Advertising a modest performance improvement over the original M1 SoC, which is about 18% for multithreaded CPU workloads and 35% for peak GPU workloads, M2 is Apple’s iterative and updated technology on Mac SoCs. This is your first chance to incorporate and update the underlying technology. Tear laptop in the face of recent updates from competitors.

The M1 SoC champion, the M1 Ultra, isn’t three months late, so Apple isn’t spending time preparing its second-generation Apple Silicon SoC. To that end, the company has prepared the first (and definitely not the last) of the new SoC family with the Apple Silicon M2. Designed to replace the M1 in Apple’s product lineup, the M2 SoC was initially rolled out with a renewal of the 13-inch MacBook Pro and MacBook Air, with significant proprietary redesigns in the process.

With the launch of the M2, Apple will be able to get a real glimpse of how Apple handles updates within the Apple Silicon ecosystem for the first time. In the iPhone family, Apple maintains the annual rhythm of A-series SoC updates. Conversely, the traditional PC ecosystem is closer to the rhythm of the last two years. The M2 seems to split this in the middle, about a year and a half after the original M1, but from an architectural point of view, it looks like an annual A-series SoC update.

Broadly speaking, the M2 has a limited number of changes, at least as many changes as Apple wants to disclose at this time, and how many, in contrast to the jackpot that was the first M1 SoC. The focus is on that important area. .. All of this is preliminary before getting further disclosure from Apple or practical time for the hardware itself, but M2 is very similar to a derivative of the A15 SoC, similar to how the M1 was derived from the A14. increase. As a result, at first glance, an M1 to M2 upgrade is very similar to an A14 to A15 upgrade.

According to Apple, the new SoC consists of about 20 billion transistors, which is 4B (25%) more than the original M1 and 5B more than the A15 SoC. This chip is made from what Apple calls the “2nd Generation 5nm” process. It is likely that this is TSMC’s N5P line, which is the same line used for the A15 SoC. The N5P has improved performance characteristics compared to the N5, but not the density. So while Apple doesn’t disclose the size of the die, the company’s side-by-side die shots are at least accurate in that the M2 will be a larger chip than the M1.

Starting from the top, with respect to the arm architecture CPU core, M2 holds Apple’s four performance and four efficiency core configurations. Apple hasn’t disclosed the generation of CPU cores used here, but based on performance expectations and timing, there’s good reason to believe that these are the Avalanche and Blizzard cores that were first introduced in the A15. It is in.

In terms of performance, Apple says the M2 offers 18% better multithreaded CPU performance compared to the M1. The company doesn’t provide a breakdown of clock speeds and IPC gains, but if our premonition that M2 is Avalanche / Blizzard is correct, we already know what the breakdown is. Compared to the A14 / M1 Firestorm core, Apple has invested most improvements in improving overall energy efficiency, so Avalanche offers only a small performance improvement. As a result, most of the performance gains are due to clock speed improvements, not IPC improvements.

The M2 performance CPU core also comes with a larger pool of L2 caches to help improve performance. In M1, 12MB of L2 cache was shared between cores, but in M2 this is up to 16MB, an increase of 4MB compared to both M1 and A15.

Based on what we’ve already seen on the A15, this bigger update of this generation is on the efficiency core side of the problem. Blizzard CPU cores now behave like smaller cores, offering relatively higher performance and a much wider backend design than those found on other Arm efficiency cores. In particular, Blizzard has added a fourth integer ALU. Combined with other changes, this has significantly (28%) improved A15 performance on these cores. It’s not unreasonable to expect similar benefits to be taken over by the M2, but the wildcard factor is the clock speed that Apple dials.

This also seems to be the reason why Apple seems to have decided to focus on MT performance for Apple’s comparison. Maximum performance gains are achieved thanks to the efficiency core. In performance-limited situations, MT workloads leverage the E-core along with the P-core for maximum performance gains. Overall, Avalanche / Blizzard spent a modest year at the forefront of CPU microarchitecture, which seems to have been carried over to the M2 SoC.

On the other hand, in terms of GPU, Apple is getting bigger. As always, the underlying architecture is hidden, which we simply call the “next generation” GPU, but the M2 has 8 to 10 GPU cores built into the M1. Officially, the GPU is rated at 3.6 TFLOPS, which is 1 TFLOPS more than the 8-core M1. Similarly, the new GPU comes with a larger shared L2 cache, but Apple doesn’t disclose the cache size there.

With a higher number of cores and a 10% increase in GPU clock speed (based on TFLOPS), Apple is touting two performance numbers for M2 GPUs. At Isopower (~ 12W), the M2 should provide GPU performance that is 25% faster than the M1. However, the M2 GPU, for better or for worse, can consume more power than the M1 GPU. According to Apple, at full power of 15 watts, it can provide 35% higher performance.

Overall, this shows that while Apple has been able to improve energy efficiency (GPUs love to run widely and slowly), Apple’s peak GPU power consumption is rising. This should minimize the impact on light workloads, but it’s interesting to see what that means for a relatively heavy and constant workload, especially on a fanless MacBook Air. prize. On the other hand, the GPU’s display controller doesn’t seem to have changed at first glance, and 6K is the best for an external monitor.

Adjacent to the GPU update, M2 also comes with an updated video encode / decode block. At first glance, this is very similar to the simplified version of the block used by the M1 Pro / Max. These SoCs have added support for Apple’s ProRes and ProRes RAW codecs, and that support is now filtered to base M2 SoCs. Similarly, Apple now officially supports 8K video decoding on the M2, but the M1 has no official resolution specification, but was basically a 4K part.

Finally, on the processing side of the problem, M2 inherits from A15’s updated neural engine. According to Apple, this is still a 16-core design and has the same rating of 15.8 trillion times / second (TOPS) as the A15’s neural engine. It’s 40% faster than the M1’s neural engine, which topped the 11 TOPS, even though it’s comparable to the A15.

Overall, Apple predicts that it has great confidence in the performance of its second-generation Apple Silicon chips, as well as its competitiveness with Intel. You have to wait to get the hardware to see the performance, but the M1 certainly responded to that claim. Therefore, expectations for M2 are high as well.

Memory: LPDDR5-6400, up to 24GB

While Apple’s latest SoC core logic looks primarily like an enhanced version of the A15, it has one very notable feature advantage of LPDDR5 support.

Vanilla M1 (and A15) only supported LPDDR4x memory, while M2 supports the new LPDDR5 memory standard. Its biggest change is support for much higher memory clock speeds. Based on Apple’s numbers, the M2 is operating at 6400Mbps / pin (LPDDR5-6400), a significant improvement over the original M1’s 4266Mbps / pin (LPDDR4x-4266) memory clock speed. The end result is that on the SoC’s 128-bit memory bus, the memory bandwidth of the M2 is 100GB / sec, which is 50% higher than the M1 (about 68GB / sec).

Significant increases in memory bandwidth help Apple maintain its position, as the unconventional use of Apple’s memory technology remains one of the key advantages over competitors in the laptop space. increase. Improving memory bandwidth further improves every aspect of the SoC. This is especially true for GPU performance, where memory bandwidth is often the bottleneck factor, and the addition of LPDDR5 will be a major enabler for larger 10-core GPUs. In this case, it’s catching up with M2 playback in a way, but all M1 Pro / Max / Ultra ship with LPDDR5 support first, but M2 is actually the last to get new memory. M series chip / tier.

Beyond that, Apple is once again placing the LPDDR5 memory package on-chip along with the processor die itself. Therefore, each M2 chip must be pre-loaded with memory, and device supply can vary slightly based on memory capacity, especially depending on the most common early configurations.

M2 devices can be used with either 8 GB, 16 GB, or 24 GB of memory. Given that Apple still uses only two stacks of memory, it seems that Apple is finally taking advantage of support for non-squared LPDDR die sizes (such as 12Gb dies). This allows 12GB of memory in a single package with no more Shenanigan. And assuming Apple replicates this for mandatory Pro / Max / Ultra SoCs, the top memory capacity of all Apple SoCs should increase by 50% over the previous generation.

And the rest: updated ISP, same USB

To conclude today’s M2 announcement, there are a few more items that guarantee a quick call.

First, M2 has an updated ISP and an updated Secure Enclave. Like the other aspects of the M2, these are likely to be inherited from the A15, and the A15 received similar updates.

On the other hand, if you look at the new MBA and MBP specifications, you can see that there are no noticeable changes in USB or other I / O support for the new SoC. Nothing has changed here, as the M1 was already at its peak in 2020, when it was launched with USB4 support. However, this means that the SoC appears to be limited to Thunderbolt 3 support, even though Thunderbolt 4 hasn’t been used for over a year. Both the MBA and MBP come with two USB ports, which still seems to be the true limit of the SoC.

Apple hasn’t even talked about PCIe features at all. You’ll know more if you have the hardware at hand, but at least for now, there’s no reason to believe that Apple has added support for PCIe 5 or changed the number of lanes available. I / O is still a limiting factor for the entire Apple Silicon family, so I’m wondering what this means for the final Apple Silicon Mac Pro.

Available in July

To conclude today’s announcement, the M2 will ship with the new 2022 MacBook Air and the refreshed 2022 12-inch MacBook Pro. According to Apple, these devices will be available in July and pre-orders will begin today.

Meanwhile, M1 doesn’t go anywhere. In addition to being at the heart of the Mac Mini that didn’t receive updates today, Apple maintains a 2020 M1-based MacBook Air. Therefore, both versions of the entry-level M-series SoC will continue to be used for some time to come.