ARMv9.2: new more efficient mobile Cortex-X4 cores and much more

Company ARM represents a new generation platform for mobile processors. TCS23 (Total Compute Solution) brings many useful improvements. It is built on architecture ARMv9.2 and again brings improved efficiency, i.e. a better ratio of performance and consumption, it also becomes possible to combine a larger number of cores. ARMv9.2 brings, for example, higher security thanks to MTE (Memory Tagging Extension) memory protection, SVE2 vector instructions were also introduced. Virtualization capability has also been improved, floating-point calculations have been speeded up, and other details. E.g. when using SVE2, performance increased by 23% in FP16, and by 10% in FP32. AArch32 support is disappearing and kernels are now exclusively 64-bit. Efficiency has been increased, for example, by enabling half or all of the L3 cache to be turned off if it is not needed.

The most powerful cores Cortex-X4 it brings 15% higher performance in single-threaded applications and 40% lower consumption for the same performance as its predecessor. L2 cache can reach a capacity of 2 MB per core. For example, branching prediction has also improved, the number of ALUs has increased from 6 to 8. We also have powerful cores Cortex-A720, which increase efficiency by 20%. Against the Cortex-A715, they achieve higher performance at the same frequency due to the increase in IPC. AT Cortex-A520 worked mainly on efficiency. They have 512kB L2 cache, support for SIMD (SVE2/Neon), however, the number of ALUs has been reduced from 3 to 2. Although this should lead to lower performance, this simplification has made it possible to achieve performance improvements in other areas and lower power consumption. As a result, performance is 8% higher and efficiency was even increased by 22%.

DSU-120 now allows up to 14 cores in one cluster and supports max 32 MB L3 cache. This should make it possible to create much more powerful processors for mobile computers (Chromebooks and the like). As a result, there is a wide range of possible configurations from the simplest 0+0+1 through 0+0+4, e.g. in virtual reality glasses, for smartphones there can be e.g. configurations 0+2+6 and 0+4+4 in low-end, in higher classes it might be 1+3+4, and laptops could also have 14-core processors in a 10+4+0 configuration.

A new GPU has also appeared, namely Mali-G620 with 5 cores and less, Mali-G720 with 6-9 cores and the most powerful is the Immortalis-G720, where it will be a GPU with 10 cores and more. The new GPU brings a 15% higher maximum performance, the efficiency also improved by 15% (ie the consumption remained the same and the performance increased), the use of the memory system was reduced by 40%.