Kernel-Based X11 Server Claims 2x Performance Over X.Org

[Delgarde]

Can this thing actually work with X.Org drivers? Everybody in this thread seems to assume it, but it wasn't stated. So, can you run AMD Catalyst, NVidia's driver, or the OSS drivers with it?

As I said earlier, I assume not. If nothing else, because they're all written as user-space code, and coded to an API defined by the Xorg implementation, not by the X11 spec. I'm guessing this doesn't do drivers at all, and just renders to a kernel-provided framebuffer.

[daniels]

also note that it's just a reimplementation of the libX11/libXext API, not the actual X11 protocol. so no network support.

[Delgarde]

also note that it's just a reimplementation of the libX11/libXext API, not the actual X11 protocol. so no network support.

Oh, ok... hadn't noticed that. I guess that means it needs a different set of userspace libraries too, then - not just the in-kernel module? API/ABI compatible, but not usable with the existing X libraries?

[daniels]

Oh, ok... hadn't noticed that. I guess that means it needs a different set of userspace libraries too, then - not just the in-kernel module? API/ABI compatible, but not usable with the existing X libraries?

exactly. or with xcb.

[ArneBab]

A non-free kernel module. Reminds me of Linus' Finger for nvidia.

[Astronaut In Orbit]

"2x"

"2x" performance boost? Ok, to really benchmark the difference you have to ofcourse understand that X is dependant on configuration.

First of all, ATLEAST renice it to -20, to not have it be a bottleneck. And then with lower interrupts, more time is spent rendering. Try 100 hz kernel. Also latencies, mean rendering is not happening with X being subject to latency. Turn on low-latency desktop.

Those should be minimal requirement for any comparison, since an X running inside the kernel, might be closer to this.