From the article:
In ensuring a fair and accurate comparison of "Windows vs. Linux" performance for Intel OpenGL graphics, Windows 7 Professional SP1 was installed on the same ASUS Ultrabook using the latest Intel Windows 7 x64 driver, which was version 126.96.36.199.3071 from late March.From what I read on this site the 10% improvement on the windows drivers was suppose to be in driver 15.31.xAs was shared recently at the GDC2013 conference, Intel Windows driver developers are also reportedly working on a more-optimized Ivy Bridge driver for release soon. It will be interesting to see how much faster the forthcoming Intel Windows driver release is for our benchmarks, but Intel has reported that it will be a measurable gain.
So if that's the case, the linux drivers are in pretty good shape.
Hi everyone. I just registered here after years of lurking because there is a question I find important to ask, which is not clear to me from the article:
Is S3TC activated in any of the linux configurations? Is any sort of libtxc_dxtn activated (either for S3TC or S2TC)?
What about floating-point textures?
From my experience this makes a world of difference in some games. This is very noticeable for example on Warcraft 3 under wine using the opengl renderer, on both r600g (Mobility Radeon 3650) and i915 (the one in ASUS EeePC 901), and also noticeable, but to a lesser extent on Neverwinter Nights (native port, never tried it with wine).
PS - Michael, thank you for all the regular benchmarks.
It's clear Intel is spooked by the leverage of ARM to Linux, so their initial reaction is to protect the Window scene therefore holding down sales in their camp. They don't realise that ARM will supersede them if they don't get properly onboard with Linux.
There's little reason why the same algorithms can't be used it Linux with modification to suit the display server and other kernel integration, so it's clear there are team secrets, or benefits being withheld from Intel Linux Drivers.
If they keep up with these shenanigans then I will go complete AMD mobo/cpu and ARM devices.
That was why I asked the question. I know it can be a potential risk to use those technologies where software patents are valid (I.E. USA), but at the same time, they will likely give you a nice performance boost.
As for the alternative routine - ST2C is an alternative to S3TC. It was an implementation of the technology specifically thought to work around the patents, but AFAIK it's also not 100% risk free. AFAIK floating point textures have no "patent safe" implementation.
I think what you mean is that no SINGLE kernel is good at everything?
The RT guys are working on that, actually. One of the main devs, Gleixner, claims that RT will eventually be as fast as linus' tree, and faster at some things (network packet handling).
And then Fedora promptly disabled it on their distro. So i'm not sure what Michael's got.
Edit: you are supposed to be able to go to the results on openbenchmarking.org somehow, but i can't figure it out. I thought there was supposed to be a link in all the charts, but i can't find one.
If you can get to it, i think that records a list of the glxinfo results, which would have the extensions available, so you could check if S3TC is showing up there.
Last edited by smitty3268; 04-05-2013 at 09:33 PM.