Zerotier performance on modern hardware?

This is sort of a BPI-R4 question but really applies to anything fairly modern.

Anyone tested zerotier performance between devices? I've seen a recent ~>1Gbps wireguard test on BPI-R4 which is promissing.

I heard that Zerotier is custom protocol which is inferior when compared to Tailscale in terms of performance, while Tailscale is based on Wireguard so I think you will see slower speed with Zerotier.

I actually get higher speeds across zerotier on fast desktop hardware vs wireguard. Zerotier is very very fast with encryption disabled (trustedpath) but it's also not licensed for commerical use at the scale/price I'm interested in. Basically I think this comes down to encryption method between wireguard/tailscale and zerotier's performance.

However, I have a lot of use for an administrative tunnel, where I don't need any encryption.

Then you are comparing apple and orange, if encryption is not needed, then even low end hardware can do the job because CPU is not really involved, any device with NAT (or not even need to count NAT since traffic is between router <> another peer end) speed >= 1G would give you same feeling.

that response wasn't really on topic, just pointing out that 'zerotier' is quite fast, encryption choice is what slows it down.

Most of the time a no-encryption zerotier isn't an option, and they have that 'trusted path' scheduled for removal for v2 so it's all irrelevant anyway.

Software encryption on most arm chips seems to be the slow point. I can push a LOT of data through wireguard or zerotier or whatever on x86-64 hardware, even handle transparent fragmentation well, but on the limited number of arm/arm64 platforms I've tested, encryption kinda sucks. Fragmentation also seems to cripple everything I've tested. So pushing 1500MTU through vxlan or GRE on a 1500MTU WAN with the tunnel transparently handling fragmentation.

For example, I can push 5Gbps over a mikrotik rb5009 on vxlan, that drops to under 1G introducing fragmentation into the vxlan or gre tunnel. That's far more than doubling the PPS rate would cost. zerotier (which is an older client on that platform) drops down to a few hundred Mbps from well over a gig. I know this isn't openwrt but I don't have anything else on the higher end of arm/arm64 to compare.

Other than Apple Silicon, I don't really see any generally fast ARM platform, there might be no initiative for manufacturer to produce those ARM chip (it's unlike x86, even same ARM chip on different boards requiring different bootstrap), x86 was being used widely for that many years of course will have more processors available on market and support.