Major Performance Improvement Discovered For Intel's GPU Linux Driver

An anonymous reader writes: LunarG, on contract with Valve Software, discovered a critical shortcoming with the open-source Intel Linux graphics driver that was handicapping the performance. A special bit wasn't being set by the Linux driver but was by the Windows driver, which when enabled is increasing the Linux performance in many games by now ~20%+, which should allow for a much more competitive showing between Intel OpenGL performance on Windows vs. Linux. However, the patch setting this bit isn't public yet as apparently it's breaking video acceleration in certain cases.

Will it work with my distro?

[NotDrWho]

Because my distro is the best.

Re:

[Anonymous Coward]

Fuck you, mine is better.

Re:Will it work with my distro?

[Chrisq]

Fuck you, mine is better.

I'm waiting for it to be ported to HURD

Re:

[Immerman]

Better get comfortable, maybe have a goblet of wine. Better yet bring a few barrels.

Re:Will it work with my distro?

[Wootery]

Wine? Valve are supporting Linux natively!

Re:

[Wootery]

Huh. 4 Informative? I was going for Funny.

Re: "...waiting for it to be ported to HURD"

[peacefool]

It's hurdly possible. Sorry.

Re:Will it work with my distro?

[Anonymous Coward]

As a longtime user of both "My Distro" Linux and "Mine" Linux, I can honestly say that I have never noticed a performance difference between them. However, I can emphatically state that "Your Distro" Linux sucks.

Re:

[toddestan]

Will it install on My Computer?

Re:

[NotDrWho]

I bet you're not even using the best Desktop, which is the one I use.

Re:

[i kan reed]

It will if you go and set the bit yourself, and correct all the errors it causes.

Re:

[Pope Hagbard]

Xemacs is the best Linux distribution.

Re:Will it work with my distro?

[Anonymous Coward]

I agree. The only thing it is missing is a good text editor.

Re:

[occasional_dabbler]

Oh, for mod points today! Well played, sir, you made my evening :-)

Re:

[Sun]

Sure it does [emacswiki.org]

Whatever

[future assassin]

with this new bit switch my BeOS can now perform multimedia tasks with the power of a uber computer.

Benchmark Bit

[bluefoxlucid]

It's the Benchmark Bit!

Re:Benchmark Bit

[TheRealMindChild]

Probably the "Official driver" bit

Re:Benchmark Bit

[Anonymous Coward]

The Intel drivers for Linux are official and open source. They are actively maintained by Intel themselves. This is not like the Nvidia/Nouveau split, Intel are actually very open source friendly in this area.

Re:Benchmark Bit

[UnknowingFool]

Also most likely the Linux drivers are written by different people than the Windows or OS X drivers so there are probably so noted differences in the coding. Some of the differences are due to APIs. Procedurally there would be practical reasons to keep driver development separate. There may be some NDA stuff on the proprietary side that Intel has to maintain with the OS.

Re:

[arglebargle_xiv]

The Intel drivers for Linux are official and open source. They are actively maintained by Intel themselves. This is not like the Nvidia/Nouveau split, Intel are actually very open source friendly in this area.

So you've got the choice between crappy graphics hardware with OSS drivers and high-end graphics hardware with binary-blob drivers. Damn.

The "paid Microsoft tax" bit, apparently

[pla]

I notice TFA has almost no detail beyond what TFS says. Yeah, so they found this bit that apparently has no side effects to anything else but magically boosts performance by 20%? I'll admit I haven't written a graphics card driver since back in the VESA2 days, but I can't even conceive of what function such a bit could have, without having some down side... Something like (and I don't mean this literally) disabling vsync but accepting tearing.

Re:The "paid Microsoft tax" bit, apparently

[Anonymous Coward]

You read the article, but did you read the summary?

"However, the patch setting this bit isn't public yet as apparently it's breaking video acceleration in certain cases."

Re:

[Impy the Impiuos Imp]

Which is irrelevant if its purpose is to look for and accelerate benchmark-like activities.

Is this perma-set on Windows? Or just on benchmarks? And what would the bit be called?

Re:

[mrzaph0d]

it's "turbo_bttn"

Re:

[jones_supa]

On Haswell architecture there is some special bit that has to be set for the hardware samplers to work at maximum performance. There seems to be no details regarding what this specifically means. The original LunarG post [lunarg.com] seems to contain the best information. What is quite cool is that they claim to have plugged LLVM into the Mesa shader compiler.

Re:

[Anne Thwacks]

Its the "Windows is running" bit.

Re:

[jeffmeden]

I notice TFA has almost no detail beyond what TFS says. Yeah, so they found this bit that apparently has no side effects to anything else but magically boosts performance by 20%? I'll admit I haven't written a graphics card driver since back in the VESA2 days, but I can't even conceive of what function such a bit could have, without having some down side... Something like (and I don't mean this literally) disabling vsync but accepting tearing.

It disables DirectX 7 backward compatibility......

think i am kidding?

I have seen that before

[Anonymous Coward]

Removing flush() from code boosts DBMS performance by 40% but that patch wasn't yet accepted as in rare cases it makes data disappear after power failures.

Re:

[TheRaven64]

I don't know specifically, but for some GPUs there's a flag to disable strict IEEE floating point compliance. This can make things go a lot faster, at the expense of sometimes giving the wrong result. For graphics workloads, a few more floating point rounding errors are normally invisible to the human eye, but for scientific (GPGPU) computing they can be problematic.

Re:

[Compholio]

... conceive of what function such a bit could have, without having some down side..

Power-save -> 0, Performance -> 1

Re:

[pla]

Power-save -> 0, Performance -> 1

Laptop-safe -> 0, crotch-burn -> 1. :)

But yeah, I see your point, that it wouldn't have any side effects in the output-integrity sense.

now you've done it...

[funkymonkjay]

you've flipped the EVIL bit!

Re:

[NotDrWho]

Yeah, there's your problem right there. Someone set this driver to "evil." ;-)

Re:

[funkymonkjay]

Evil Inside (tm)

"apparently it's breaking video acceleration in ce

[grahamtriggs]

That's probably why it wasn't being set then.

Re:

[Anonymous Coward]

Your comment is misleading. Just because a performance boost causes stability issues (whilst still under development!) doesn't mean those issues can't be ironed out. If the bit works on Windows, most likely it will work on Linux too after the devs make the right adjustments. The fact that it wasn't set was a mistake, not a design decision.

Re: "apparently it's breaking video acceleration i

[SpaghettiPattern]

But _is_ set in the Windows driver? No sense in that.

Re:

[m.alessandrini]

Because on windows people don't complain when the system crashes.

All games?

[Anonymous Coward]

Would this help with nethack?

Re:

[Anonymous Coward]

Yes, but the grid bug kills you 20% faster.

The Turbo Button

[CambodiaSam]

Someone found the Turbo Button from my old 386. http://en.wikipedia.org/wiki/T... [wikipedia.org]

Re:

[Carewolf]

Someone found the Turbo Button from my old 386. http://en.wikipedia.org/wiki/T... [wikipedia.org]

Funny thing with the turbo button was that it made your computer slower. Specifically as slow as a 8086.

Re:

[toddestan]

That's exactly how it worked on every computer I ever came across. For normal operation you'd always have the computer in 'turbo'.

Re:

[Nyder]

Actually it depended on the motherboard what the "Turbo" button did. Some it was for speed ups, some it was for speed downs, and some it didn't even work.

Re:

[canadiannomad]

and some it didn't even work.

Probably because people like me wouldn't bother connecting the button on computers we built, and would instead just put a jumper.

This one simple bit bumps your performance by 20%

[Overzeetop]

...and you won't believe how easy it is!

C'mon editors, it's like you missed the entire social media headline writing class.

Re: This one simple bit bumps your performance by

[Anonymous Coward]

One weird bit, benchmarks hate it

Yeah, yeah keep reaching for that low hanging fruit, beating that dead horse.

Re:

[Andrewkov]

The secret bit Intel doesn't want you to know about!

Re:

[Overzeetop]

Oooh, that's a good one.

Man, I hate these headlines. I now reflexively ignore stories which are captioned this way - basically anything which doesn't describe the article content, no matter how many boobs, muscles, butts, or cute/sad animals are in the photo caption. No, I take that back - *especially* if there are boobs, muscles, butts, or cute/sad animals are in the photo caption.

Re:

[mccrew]

I flipped this bit, and you won't believe what happened next!

Re:

[Anonymous Coward]

Windows developers HATE this man!

Developer discovers this one "weird" trick to boost driver performance.

...in certain cases.

[Thanshin]

So, it works 20% better... until it crashes.

And... we all know most of the Linux user base would want to give up random crashes for 20% better performance, right?

Re: Pointless improvement?

[bill_mcgonigle]

People are quick to yell "games" as if the entire desktop hasn't been 3D-accelerated for a decade.

In many cases if a chip can be done 30% faster, not only is the user happy with the visuals, but that silicon can enter low-power mode more quickly. A laptop user might get a few more minutes' battery life with this bit on and the world might burn a few less tons of coal for all of the systems.

Re:

[jedidiah]

> People are quick to yell "games" as if the entire desktop hasn't been 3D-accelerated for a decade.

Except you don't need a good gaming card in order to deal with that crap. The cheapest trailing edge AMD/nvidia stuff that's likely to be not supported in the next driver release is more than fast enough for "desktop 3D acceleration".

Re:

[Guspaz]

Real gamers often do their gaming on a desktop, and have an Ultrabook for portability. Said gamers might not want or need the bulk of a gaming notebook 99% of the time, but might still appreciate the ability of an Intel iGPU to handle basic game rendering on the rare occasion when they want to keep themselves busy while on the go.

I'm a gamer, and I do all my gaming on a relatively high-end desktop. I've got a Macbook Air, because I only have a desire to fire up a game on my notebook a handful of times a yea

A question: would this affect CUDA performance?

[whitroth]

Some of us are using those GPUs for actual work - like serious scientific computation. Anyone know anything about how this bit switch would affect CUDA?

                mark

Re:

[Anonymous Coward]

Some of us are using those GPUs for actual work - like serious scientific computation. Anyone know anything about how this bit switch would affect CUDA?

                mark

CUDA on intel processors? "Science guy", you don't know what you're talking about

Re:

[godrik]

Since when does Intel ship devices that support nVidia CUDA?

I liked the original headline better

[Minwee]

"How A Mom Found A Secret Bit That Improves Her Video Driver Performance By Twenty Percent! Intel Hates Her!"

Re:

[Spamalope]

"How A Mom Found A Secret Bit That Improves Her Video Driver Performance By Twenty Percent! Intel Hates Her!"

Using one weird trick? I'll never believe what happened next!