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

[dee.]

Congrats are in order... I didn't know someone could make a display server even more horribly stupid and misguided than Mir.

[mydoghasworms]

Anyone want to build a distro on this for the lolz?

I would actually love to see it in action.

[Alejandro Nova]

Watch carefully the video. Performance with standard X is worse, check. CPU utilization is higher, check. But have you noticed the absolute lack of tearing in X versus the tearfest with microXwin?

[Auzy]

Have any of you naysayers considered that this might be a reasonable solution in certain scenarios? It uses less memory and is faster, so it may be good on low-end kiosk-type hardware. Security could be better, but with specific use cases, may not be that big of an issue.

Low end Kiosk hardware have plenty of memory these days. Even a raspberry pi Type-B has 512MB and an Ouya ($99), has 1GB. In fact, most media players these days have at least 512MB of ram..

And, there are already plenty of RAM-efficient X implementations (such as TinyX by Keith Packard), http://www.pps.univ-paris-diderot.fr...re/kdrive.html

And, any application where speed would be beneficial, stability would be a factor (in a kiosk, kernel panics would be embarrassing). Sorry, but, I honestly cant think of many scenarios where performance is important, and RAM is important. It would also likely introduce significant security issues long term.


Its a cool project & idea (I'm particularly interested simply because it tells us what a near perfect/efficient Xserver implementation could achieve), but, in practice, the disadvantages outweigh the benefits, especially since GOOD hardware is cheap.

Andrew

[cb88]

Its no supprise they can do this. X.org performance is absolutely horrendous in many cases.... and this is markedly worse the older the hardware you try to use it on. I would love to use something like this on my Transmetta Crusoe based latptop since it just doesn't have the oomph for X.org.

That said this is completely worthless to me... I run custom kernels on my old machines so the kernels can be a tad slimmer/optimal. I like the idea of what they are doing but the execution is pretty poor imo. They've been sitting around with this tech for years... with little to show for it becauses they think that they just have to remain proprietary... well look at all the other failed proprietary X11 solutions there are still a few around but only because they are backed by hardware companies.

[mark45]

..leading to a security and compatibility nightmare...all for...what exactly?

Once in a while there pops up a hacker or a group of hackers somewhere with really stupid utopic ideas, like "let's create a better MS-DOS and kill Win7 with it", "let's work on Hurd", "let's optimize the telegraph and try to outcompete the e-mail", "let's supercharge the AVI container and kill MKV".

In this case another hacker (read idiot) decided to supercharge ancient technology (read crap) instead of being realistic - it's nothing new, there's lots of stupid people with lots of time, I should know I did stupid shit myself until I started to value my time.

[Kayden]

(disclaimer: this post is my personal opinion)

I'm sure we'll hear about all the downsides of such a thing in this thread. But for a hard core gamer, an X server that provides this much of a performance win would be worth it.

This does nothing for gaming. If you look at their Performance page, you'll see that they measured using gtkperf and x11perf, which only cover 2D. Not only that, but the x11perf tests they called out are an extremely limited subset of x11perf. It basically looks like they've optimized server round trips a bit, which is nice but not terribly interesting. Common practice is to avoid costly round trips wherever possible.

I'm also skeptical that their system allows for hardware accelerated 2D drivers, such as xf86-video-intel/ati/nouveau...which would hurt performance of 2D significantly. I don't see any information to indicate that it works, at least.

So, basically, we have a closed source system...which puts a complicated system that barely even has a regression test suite directly inside the kernel...leading to a security and compatibility nightmare...all for...what exactly?

[siride]

Have any of you naysayers considered that this might be a reasonable solution in certain scenarios? It uses less memory and is faster, so it may be good on low-end kiosk-type hardware. Security could be better, but with specific use cases, may not be that big of an issue.

[Delgarde]

I'd use this if the performance is really that much better. I've got an i7-3770 and a GeForce GTX 660Ti.. I use nVidia drivers as it is and I have no objection to it.

Yeah, good luck on that. A proprietary driver designed to work with the X.org driver model? And you think it'll be useful on an obscure proprietary non-X.org server?

You may also notice that there's no mention of OpenGL support in the announcement. It looks like this is just basic X11 from the late 80's - no GLX, and few if any of the other modern extensions. For example, their Compatibility page indicates it works with Gnome 2 using OpenBox WM - but not with the default Metacity, since it doesn't support the xkb extensions required. Seems to me you'd barely be able to run a desktop, never mind GPU-intensive games.

[halfmanhalfamazing]

I'm sure we'll hear about all the downsides of such a thing in this thread. But for a hard core gamer, an X server that provides this much of a performance win would be worth it.

Perhaps we should consider merging Xonotic into the Kernel............... I mean, for a 2x performance boost? Sure! That Jedi Knight game was recently open sourced, they should merge that into the kernel too. And how about Unigine!!!