Late last month, Arm announced a pair of new CPU cores that the company licenses to chip partners who then use these designs to build chips of all kinds. Arm itself is owned by the Japanese conglomerate Softbank but was acquired because of the dominant position the company has in smartphones and is poised for IoT. The smartphone market is declining in sales due to COVID-19, but was already declining prior to this due to a lack of new killer apps, increased reliability and decline in the US of carrier subsidies. With all of that said, 5G phones are picking up momentum and could be driving smartphone sales if operators can build the networks that deliver on consumers’ expectations and sell the use cases. Arm has also been enabling its partners to compete in the PC space with increasingly bigger and more powerful designs that allow them to compete with Intel and AMD and they are gaining momentum. Heterogenous computing is of absolute importance but it seemed as if the smartphone industry concluded we already had enough CPU performance. Apple kept running up the score on performance benchmarks and it became an issue for the whole industry.
Arm announced two new CPU designs, one an updated design to the previous generation and another an entirely new design. The first, the Arm Cortex-A78 is a follow-on to last year’s Cortex-A77 which is part of the same family of processors as the A76. The A77 brought some incremental improvements to the A76 and the A78 is a lot more of the same, again. However, that said, the Cortex-A78 is less about peak performance and much more about improving efficiency. The Cortex-A78 was designed with area and power efficiency as the primary goal with sustained performance as a key metric of performance. Even aiming for maximum sustained performance and efficiency, Arm still claims 20% more sustained performance over the previous generation, which is still nothing to scoff at. However, it is unclear what smartphone workload that can be done now that could not be done on its predecessor. More performance is always better in the same thermal envelope, but its nice to know what the user can do differently with the Cortex-A78.
The Cortex-A78 is designed to be the replacement for the A77 in next year’s phones, if Arm’s partners choose to adopt it, which many are likely to. The Cortex-A78 is still going to be in the highest performance chips from Arm’s partners, with 5nm being the targeted process node. That said, not all of Arm’s partners will be using the 5nm process node, like Huawei, and those customers may not benefit from of the areal and power improvements that 5nm FF can bring. Arm’s Cortex-A series will no longer be the highest performance Arm chips for mobile which leads us into Arm’s second chip announcement.
The Arm Cortex-X1 is Arm’s newest addition to the Cortex family of processors and introduced an entirely new tier of chip with a new structure of partnership. The new Cortex-X1 is designed to be Arm’s processor targeted squarely at performance first and as a result is Arm’s most powerful Cortex CPU ever. Arm claims that the Cortex-X1 is 30% faster peak performance than the last generation Cortex-A77 announced last year and in this year’s devices. The Cortex-X1 is unique in that Arm is allowing customers to customize the design to fit their individual needs based on application, which they kind of started to do with some select customers. The Cortex-X1 does away with many of the power constraints of the Cortex-A78 and allows for customizable designs to meet customers’ design specifications. The Cortex-X1 also doubles ML performance over the Cortex-A77 and A78 which is also considerable when you think about how much ML phones are doing today and the CPU is always the fallback processor for AI when other cores are unavailable. The Cortex-X1 isn’t designed to operate on its own, but rather as a part of a multi-processor system including 3 Cortex-A78 cores and four smaller Cortex-A55 cores. This design increases peak performance by 30% while increasing cluster area by 15% compared to the Cortex-A77 designs, assuming the same node (not 5FF). The Cortex-X1 will offer up to 1MB of L2 Cache and up to 8MB of L3 cache which is double what the Cortex-A78 offers to Arm partners. I really like the Cortex-X1 as it enables some customizations without the need for a full architectural license, but I believe that Arm was already doing this in some capacity but was not a formal program.
Arm’s CPU designs are starting to become more predictable in the performance improvements that they deliver, and that’s a good thing. Because of Arm’s hard work in improving CPU performance, the company has helped to be the tide that rises all boats when it comes to mobile computing. However, that said, Arm’s new Cortex-X1 processor is a clear indication from Arm that the company and its partners intend to come after the likes of Intel and AMD in the low power PC category. While I could absolutely see some smartphone designs embracing the Cortex-X1, I believe that we are much more likely to see it in tablets and laptops where thermal constraints are more relaxed and the need for bursty peak performance is very important. We are slowly seeing a convergence between the design philosophies of Arm and Intel as they start to compete with one another in the PC space. While Arm has the big powerful Cortex-X1 paired with the efficient Cortex-A78 and Cortex-A55 cores, Intel has the high-performance big Sunny Cove core paired with the efficient Tremont efficiency cores (which we wrote a paper about). The strategies are converging and I believe we may finally see which of the two has won round one of this battle probably next year.
Moor Insights & Strategy Analyst Anshel Sag Contributed to this Article