It is primarily the CPU in the SoC; First ARMs were the 32-bit Cortex-A15 and Cortex-A7 (ARMv7). Now all 64-bit Cortex-A have it, see https://developer.arm.com/documentation/102142/0100Just as a side to this and towards the hardware.
What virtualization technologies do the Pi5 and Pi4s SoC have, I know the AMD64/x86 have a lot tagged in if you get the correct CPUs and turn them on in the BIOS, VT-x etc.
?
For 32-bit, there is no standard configure method in the Linux mainline kernel AFAIR. It is long time ago I looked into that, at least the Raspbian is not even ARMv7, so better get generic Debian which has all tools for armhf the same way is has for aarch64. But I don't see formal ARM documentation, so no-one seemed to be interested for 32-bit. Also, you get a Cortex-A53 for only slightly more silicon space as Cortex-A7 AFAIR, so only 64-bit Cortex-A is the way to go.
Besides the in-core stuff, there is IOMMU, options to map external HW devices dedicated to/for a VM. Popular is having an extra PCI-E GPU card in a x86 motherboard and use Intel VT-d (their IOMMU, AMD is different) to map that card exclusively to the Windows VM. So your Linux PC still has native local GUI on on-chip iGPU (or AMD APU) and Windows can then use its own GPU drives (NVIDIA or so) on 'own HDMI cable'. ARM provides SMMU component see https://developer.arm.com/documentation ... -SMMU-does. You can configure this in the virt-manager (adding virtual machine hardware) if the motherboard has it properly wired etc w.r.t. PCI-E.
But RPL has indicated that this is not usable on RPI5 for end-user. It is used in ARM SoCs at least to map HW blocks like HEVC, DMA, etc. I am not sure why x86 needs BIOS config, it might be an legacy/old real-mode issue, back to DOS and A20 address-line gating, or something like that that Intel likes to get rid of, but do they manage that is more the question. We know Apple SoCs are clean 64-bit already since years.
Statistics: Posted by redvli — Mon Mar 03, 2025 9:08 am