Ok ok, but ARM is becoming popular, VMware has noticied it and they're developing in ARM branch, I don't want that my favorite system (Proxmox) keep stucked only in x86
.
We already evaluated ARM 64 back in 2016, I worked on a Gigabyte board with an Applied Micro CPU, which was slow, but we expected that to a certain degree (not many cores and not high clock). Further, I could test a big server from packet.net with Cavium/Marvell ThunderX2 CPUs (96 cores) which were way slower than my local Intel based workstation with 8x2 Cores, for example when doing a Linux kernel compilation from memory (almost purely CPU bound task). So that wasn't a promising start.
We even had a POC for porting PVE 5.x on it, but performance and HW availability where abysmal, so it was paused and not picked up again.
ARM64 HW throws out many cores but has awful interconnects between them, so they just stall each other half the time.
Especially intel has not that many cores, but good interconnects and much higher "instruction per clock" (IPC) throughput, AMD has actually both nowadays, good interconnects and high core count.
I've saw a equivalent server but in x86 (same performance, disks...) and it uses kind 2000W - 3000W power supply
this ones use only 1200W (800W-900W in full load) in long term it saves a lot of energy and in the end pay the extra costs.
Did you actually confirm that? As lots of those look good on paper but then just do not work out.
As said above, I worked with such a ThunderX2 system and while I did not measure actual power consumption of my Workstation (which beat that system) it has a ~600W PSU, and really do not think that was maxed out, CPU is rated at 95W TDP.
Well, sorry but I said a simple one and I guess that there aren't Xeons that has consumption of only 5-10W max.
There are Xeon-D CPUs with 35W, and they can deliver up to 4-5x (ballparking) the performance, so actually more efficient if you break it down to Instruction/Watt. There are other, non Xeon models, which use 10W at 4 cores like the Intel Atom C3436L - just to name an example.
So basically one can find just as efficient HW in AMD64 space, plus that HW has good connections (SATA, PCIe) that actually work (lots of ARM boards with SATA cannot use to boot from them, and m.2 PCIE slots often just work with WLAN cards not NVMe), the ecosystem for peripherals is there, they work with mainline Linux kernel (that got a bit better with ARM in the last years, but still far from good) and options for really powerful Server HW are available to buy "off the shelf" - all that really does not help the argument for an ARM port.
As already mentioned somewhere else in this forum, if I'd put in personal energy for a port, it would be for an actual open HW platform, not yet another proprietary platform like ARM. RISC-V seems like it could get there, but powerful HW with virtualization support does not yet exist.
