ARMv9 Scalable Matrix Extension Support Lands In Linux 5.19

[phoronix]

Phoronix: ARMv9 Scalable Matrix Extension Support Lands In Linux 5.19

The 64-bit Arm (AArch64) architecture changes have been merged into the in-development Linux 5.19 kernel...

https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.19-AArch64

[brucethemoose]

Oh, this extension is fascinating. It looks like SVE2, where the hardware implementation can scale way up/down without the need to recompile anything.


These compute focused ARM designs are going to be absolute monsters. Just look at how happy the NSC already was with the A64FX, straight out of the box, with so little toolchain/library support for SVE1: https://www.nextplatform.com/2021/06...ugh-the-paces/


There's no way x86 can compete with that. Maybe an ARM/RISC-V SoC from a GPU vendor can, but that remains to be seen.

[coder]

There's no way x86 can compete with that. Maybe an ARM/RISC-V SoC from a GPU vendor can, but that remains to be seen.

Perhaps a future version of Intel's AMX will add more generalized matrix operations.

As for GPUs, both AMD and Nvidia now have generalized matrix support in their datacenter compute accelerators.