I’m a pretty happy user of tailscale and not someone who’d just switch because some other product claims to be so much faster/better but I’m wondering how there can be a 5x speed difference when the underlying technology (wireguard) is the same?
There can’t be a 5x speed difference
Are you talking about the chart on their homepage? This one?
Yeah, that’s really not possible. TS should be on about the same level as WireGuard (slight differences would be attributable to TS’s use of DERP relays for connection setup). There’s some serious number-fudging going on here.
Yes, that is what I was referring to.
That graph doesn’t specify the version they used to test or the testing environment so it’s hard to say exactly how they arrived at that particular number. In my experience, Tailscale’s throughput is much higher than shown in that graph.
Depending on the scenario, raw WireGuard may be meaningfully faster. For example, kernel WireGuard avoids certain types of overhead in packet processing. The Tailscale NAT traversal layer is fairly well optimized but it does add some cost vs plain wireguard-go. I expect further optimizations to become available over time but currently it appears to be fast enough that it works well for most people in most situations.
There are also platform-specific differences, such as that 32-bit Raspberry Pi is much slower than 64-bit Raspberry Pi due to optimized Go crypto implementations for 64-bit ARM. The graph shared above is about cross-cloud communication (and thus probably x86-64) so that is unlikely to be related in this case.
and here is their full blog post about these speed tests with more info: Which VPN is fastest? A speed test | netmaker