Ten years after the Armv8 architecture, ARM unveiled the outline of its new Armv9 design at its Vision Day event. The latter does not mark a leap as radical as the passage from v7 to v8, but it will be enough for ARM to give its future cores new capacities and better performance … while allowing mobile games to enter a new era.
Resuming the main bases of the ArmV8 design introduced in 2011 … but by making the formula more complex to adapt it to the three major challenges of the decade according to ARM: security, AI and DSP (digital signal processing) capabilities, it is the raison d’être of the new Armv9 architecture, presented this week by the British group. This new design remains based on the AArch64 instruction, ARM’s first 64-bit architecture, but benefits from enough new extensions to justify a change in nomenclature, AnandTech explains.
Its main architectural bias, however, remains the switch from ARM NEON technology to SVE2 technology (Scalable Vector Extension of the second generation). This change, foreshadowed during fall 2020 after an initial announcement in April 2019, should make it possible to considerably increase the capacities of the next ARM chips in DSP, but also in IA, or even in cryptography, while boosting their performance in computing.
Your future smartphone and supercomputers will have one thing in common …
If this promising transition had so far only been attempted on certain CPUs (such as the Fujitsu A64FX, which powers the Japanese supercomputer Fukagu, world number one), it is no coincidence: the first version of the Although SVE technology, launched in 2016, was also based on SIMD calculation units, it did not allow as much versatility as that offered by NEON technology at the time.
Enough to limit its relevance mainly to the HPC sector (intensive computing and supercomputers). Almost five years later, its successor, the SVE2 technology, allows ARM to reshuffle the cards… by giving the chips of our future smartphones certain supercomputer genes, at least from an architectural point of view.
Gamesdone : ARM via AnandTech
As AnandTech explains, compared to the NEON, the SVE / SVE2 embeds new SIMD capabilities, but also and above all benefits from a vector of variable size ranging from 128 to 2048 bits. An advantage that developers should quickly be able to take advantage of: the SVE2 will allow them, in short, to compile their code only once for more flexibility.
Armv9: the promise of a 30% increase in CPI in the medium term
More down to earth, ARM promises that with Armv9, its next two generations of mobile cores, “Matterhorn” and “Makalu”, will each develop a generational CPI 14% higher … for a total increase of almost 30% of the CPI through these two generations. ARM also specifies that this figure is calculated without taking into account the frequencies that will be applied to chips designed by SoC manufacturers (to which ARM sells its licenses). The end consumer could therefore logically benefit from a more notable increase in performance.
On the GPU side, ARM also discussed the future of its Mali iGPUs, which will also take advantage of Armv9 to gain capacity. The firm explains in particular working on the implementation of Variable Rate Shading (VRS), but also and especially ray tracing. Suffice to say that mobile games should become even more fluid and visually attractive thanks to the next SoCs based on ARM technologies.
It remains to be seen when the next generation SoCs, taking advantage of the Armv9 architecture, will emerge. On this side, no too long delays: ARM should present a replacement for the Cortex-A78 and X1 cores, under Armv9, from this year for an arrival on commercial devices by 2022.