The Loser In Our Windows vs. Linux Tests: Intel Graphics

[whizse]

This is disappointing. Why can't Intel driver share most of the code between Windows and Linux versions? AMD-ATI and NVidia seems to be able to do it with the proprietary driver.

They probably could, but at least I would really, really, hate to see yet another proprietary driver.

[whizse]

Several caveats here:

The tests were run using Mesa 7.7.1. Clarkdale graphics are relatively new and the support and optimization for a new architecture were just recently added. Try Mesa 7.8 or 7.9 instead.

Generalizing about Intel graphics is unfair because their older products (X3100, X4500) are generally on par with or better than on Windows. Again, that's because of more time for optimization and general neglect on the Windows side of things.

I guess a lot of distros will still use 7.7, so the benchmark is fair in the way of out-of-the-box performance, but the step up to 7.8 was a great improvement. (Perhaps not in way of performance, but for stability and bugfixes).

[aliendude5300]

Although I wish Linux won this comparison, I am satisfied with the results. I hope Phoronix continues to provide accurate comparisons between operating systems, so that we can know how far ahead of/behind windows we are, how much commitment manufacturers have towards their Linux drivers, and how well different cards perform across platforms. I hope that the realization of our inferior performance will inspire developers to work on performance optimizations, as well as adding new features, such as OpenGL extensions. I'm looking forward to seeing more articles like this.

[karl]

They probably could, but at least I would really, really, hate to see yet another proprietary driver.

I was thinking they should use the same code but keep it open (not closed).
If not the code then at least to use the ideas (as they surely must have the code from windows driver).

[m_gol]

My Intel 4500MHD can handle even relatively new games under Windows. Under Ubuntu, even OpenArena stutters. Come on, this game uses the Quake III Arena engine!

So these results don't surprise me at all.

[some-guy]

- A new Direct Rendering Interface
- 3D acceration with Gallium 3D

DRI2 was made by krh, who works for redhat NOT intel.
Gallium3D support was written by Tungsten Graphics for intel, and intel rejected it, and never wrote a different one.

-X Server can now handle compisiting with AIGLX(Accelerated Indirect Rendering)

I believe you mean WITHOUT AiGLX.

-Run multiple 3D applications at the same time

This is because of DRI2 which was made by krh, not intel.

-Hardware accelerated Video

VAAPI was made by intel, but they never supported it (except in poulsbo, but that's proprietary), nvidia made vdpau which works well under nvidia card, ati has xvba under the catalyst drivers, and broadcom has crystalhd.

The Intel Graphics driver stack has improve vastly since one year ago. Sure you lost some FPS when playing games, but the performance optimizations on the new architecture haven't come yet.

Intel's performance was never that great, and they are avoiding gallium support.

Intel did do a lot, but don't give them credit for things they haven't done.

[deanjo]

With in the past year the intel driver has brought:

- A new Graphical memory manager
- A new 2D hardware acceration architecture
- A new Direct Rendering Interface
- Kernel Modesetting over userspace modesetting
- 3D acceration with Gallium 3D

Pretty much every facet of the driver has changed.
From these changes you these benefits that you didn't have before:

-X Server no longer has to run as root
-X Server can now handle compisiting with AIGLX(Accelerated Indirect Rendering)
-You can change displays and X servers without changing mode
-Two different X server can now have hardware acceleration at the same time.
-Run multiple 3D applications at the same time
-Hardware accelerated Video
-Applications no longer overwrite each others graphic memory when directly rendering to the screen

The Intel Graphics driver stack has improve vastly since one year ago. Sure you lost some FPS when playing games, but the performance optimizations on the new architecture haven't come yet.

In conclusion, they suck less then they did before but they still suck.

[whizse]

In conclusion, they suck less then they did before but they still suck.

They suck, but they suck a little bit less than the alternative?

[Dragoran]

DRI2 was made by krh, who works for redhat NOT intel.

Well he does in fact work for Intel now (But he wrote DRI2 and AIGLX while being at Red Hat)

[deanjo]

They suck, but they suck a little bit less than the alternative?

I'll take a blob to take full advantage of the hardware at hand any day vs using code that sacrifices way to much performance for the sake that a few skilled individuals can actually decipher the code they are looking at.

Putting a piece of hardware right now on FOSS drivers is like taking that tabby ally cat that is full of life and energy, cutting it's nuts off and watching it turn into a slow lethargic lump of fur that sits around and doesn't really move unless it absolutely needs to.