At the Computex 2023 event in Taiwan, Arm announced its latest computing platform, TCS23. (TCS stands for Total Compute Solution.) The new platform boasts a slew of updates across the board—indeed, it’s the first time in a long time that Arm has updated all its core IP technologies at once.
Over the years, Arm’s TCS program has helped the company work with its partners to deliver a more platform-centric approach that considers an entire SoC’s performance and how both design and implementation ultimately affect the user experience. As such, Arm is always looking for ways to improve performance while also improving power consumption and total efficiency. This year is no different, especially as Arm is motivated by competition from Apple’s M-series chips and the anticipated arrival of Qualcomm’s custom Arm cores developed from its 2021 acquisition of Nuvia.
CPUs lead the way
Since Arm is best-known for its CPUs, it’s no surprise that the company’s TCS23 lineup is dominated by new CPUs. TCS23 includes the high-performance X-series, A700-series and A500-series CPU cores. However, Arm is recommending a new configuration for its three different tiers of cores; previously, it recommended that OEMs use a single X-series high performance core, three A700-series performance cores and four A500-series efficiency cores in what is known as a 1+3+4 configuration for flagship. With the introduction of a new generation of performance cores, however, the company now recommends a 1+5+2 configuration. This is because Arm believes that its new Cortex A720 cores deliver enough performance and power efficiency that more workloads should run on them rather than the A500-series cores. Arm is also confident that partners who continue to deploy a 1+3+4 configuration will still be very happy with the new CPU cores, as it is claiming a 33% improvement in Speedometer browser benchmark performance over the TCS22 CPU core configuration from last year.
The Cortex-X4 is Arm’s new high-performance core, which the company says it has already taped out on TSMC’s N3E process node. This is not the case with any of the other Cortex A-series CPU cores, but the X4 is also the largest and most complex core of the three and likely the most challenging to tape out. Arm claims that the Cortex-X4 is the fastest Arm CPU ever built, with a 15% increase in performance over last year’s Cortex-X3 which can be found in most of this year’s Android flagship phones.
Arm also happily claims that the introduction of the X4 marks the fourth consecutive year of double-digit performance gains for the Cortex X series. The X4 also brings Arm v9.2 compatibility for maximum performance and security as well as 64-bit instructions, which the entire TCS23 family of CPUs has. Arm also says that the IPC (iterations per clock) gains achieved with the Cortex-X4 enable up to a 40% power reduction on the X -Series, assuming the same performance as the previous generation.
The Cortex A720 is the new performance core for sustained workloads, with 20% more power efficiency than last generation’s A715. The A720 is also 64-bit and Arm v9.2 compatible like the Cortex-X4; the company claims it has better performance at the same power consumption as the A715. Speaking of 64-bit, Arm has also talked about the growth of 64-bit apps and its plans to phase out 32-bit applications. Arm has fully shifted to building only 64-bit processors and chipsets and devices such as the MediaTek Dimensity 9200 and Google Pixel 7 are already taking advantage of that.
Forbes Daily: Get our best stories, exclusive reporting and essential analysis of the day’s news in your inbox every weekday.
In addition to the Cortex A720, Arm announced the A520, its new “LITTLE” core focused on efficiency and lowest power-per-area. Arm claims that the Cortex-A520 delivers a 22% improvement on power efficiency, even assuming the same process node, which means that all the aforementioned IP cores could deliver even better power or performance once they move to new process nodes in production.
Along with the new CPU cores, Arm also introduced a new DynamIQ Shared Unit 120 (DSU-120) that helps connect all the cores together and creates new cluster configurations that were previously not possible. This enables up to 14 configurations with as many as 14 Cortex-X4 CPU cores and up to 32MB of L3 cache, if that’s what you need.
Arm also talked about its CPUs’ improved performance for machine learning (ML), with the X4, A720 and A520 seeing 12%, 9% and 13% improvements to ML performance. When you take into consideration all the new cores, combined with the new DSU as a complete solution, Arm says that it expects an uplift of 27%, assuming the same process node, on the GeekBench 6 multi-thread CPU benchmark. That is a huge generational improvement.
Arm is introducing its fifth generation of GPU technology with its new Immortalis-G720, Mali-G720 and Mali-G620 designs. In TCS21 and TCS22, Arm introduced the Mali-G710 and Immortalis-G715 GPUs based on the fourth-generation Valhall architecture. The Immortalis name denotes the highest specification of Arm’s GPU core technology and the one that comes with ray tracing. The new Immortalis-G720 delivers both a 15% higher peak performance and 15% higher performance-per-watt, partially thanks to a reduction of as much as 40% in memory bandwidth required. Arm derives a lot of the memory bandwidth reductions it achieves from deferred vertex shading (DVS), which also enables higher quality graphics without taking up too many resources, thus helping reduce DRAM power usage and increase performance.
Arm also talked about ML performance on the GPU side, claiming that the Immortalis-G720 showed a 25% improvement in peak performance compared to the Immortalis-G715, along with 22% less memory bandwidth usage. Alongside the new Immortalis model are the Mali-G720 and Mali-620, which are intended for use in more mainstream SoCs for use in devices such as Chromebooks and TVs.
Arm’s multi-decade investment in software is quietly the company’s secret weapon for fending off competitors. Arm says that more than 45% of its engineering staff are software engineers, which tracks when you look at how much software investment is necessary across so many semiconductor companies. When you think about how Arm’s processors, whether CPUs or GPUs, enable so many of the world’s leading SoCs inside most of the world’s Android devices, it makes total sense why Arm would have so much involvement in software that enables its partners to make the most of Arm’s technologies.
On a related topic, Arm says that it is also working closely with Google to support Google’s many different APIs so as to maximize the capabilities of Arm’s partners.
Arm is yet again pushing the envelope of what’s possible for the Android ecosystem and for the vendors that supply that ecosystem with chips. This is especially true when you consider that the new 1+5+2 CPU cluster is expected to be 27% faster than last year’s 1+3+4 configuration on the GeekBench multi-thread benchmark.
Arm is facing challenges from many different directions, with competition from Apple and Qualcomm on CPU core IP as well as pressure from RISC-V from an architectural standpoint. Arm is wisely using its deeply entrenched software ecosystem to enable its partners to be successful within the Android ecosystem. However, I do believe that Arm must continue to keep its foot on the pedal for performance and efficiency to continue to stave off competition.
That said, I do think there’s likely to be less competition soon due to the constrained macroeconomic climate; I suspect that many companies will abandon fully custom silicon and instead look to adopt solutions that use Arm’s IP built with its silicon partners like MediaTek. Whatever the case, the rest of 2023 promises to be very interesting, especially considering that Apple and Qualcomm are both expected to announce competitive solutions to Arm’s partners using the latest cores.