LWN.net Logo

Free software drivers for the Intel 965GM Express Chipset

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 7:16 UTC (Fri) by pheldens (guest, #19366)
Parent article: Free software drivers for the Intel 965GM Express Chipset

I'd like to see some real world quake3/doom3 performance please, no glxgears.
Does it outperform Ati X850 on r300_dri.so? or even 9250 on r200_dri.so?

Thanks.


(Log in to post comments)

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 8:18 UTC (Fri) by drag (subscriber, #31333) [Link]

hear hear

Anybody here with a 965G care to talk about 3d performance?

glxgears -info ?

Tremulous, torcs, Nexiuz, and such should be aviable via apt-get if your using Debian. You should be able to get quake3 performance or enemy-territory performance results pretty easily.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 8:21 UTC (Fri) by drag (subscriber, #31333) [Link]

Vdrift is aviable at least through Debian unstable apt-get..

It looks very simple to do benchmarks with it...
http://www.michaellarabel.com/index.php?k=blog&i=133

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 8:40 UTC (Fri) by bobort (subscriber, #5019) [Link]

We use both r200 and G965 for molecular visualization type stuff. Comparing a 2x2.4GHz opteron w/ r200 vs. 2x2.9GHz core2 w/ G965, the latter is noticeably faster/smoother, but not tremendously so. It's hard to say whether the cpu or gpu is more responsible for this, and since r200 is AGP-only and the Intel boards are PCIE-only a proper comparison is impossible.

I don't have any performance numbers handy, and we've never used them for games.

Both of them crash the machine far too often, but G965 is worse at present.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 9:04 UTC (Fri) by nix (subscriber, #2304) [Link]

Interesting. I wonder what's different between your system and mine? My r200 (9250) has not once caused a crash in X, no matter what I did with it.

Maybe you have some other hardware problem, or you're doing heftier things than I am...

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 12:15 UTC (Fri) by tialaramex (subscriber, #21167) [Link]

I've owned a variety of hardware with R100 and R200 family chipsets. On some of the hardware 3D it crashes pretty easily (e.g. one machine wouldn't last more than a few minutes running Quake 3), on other hardware it's quite stable, but it's still definitely the least reliable part of the system.

So far as I can tell the big two vendors (I haven't tried Intel) design hardware for performance first and foremost, then cost, and reliability is definitely an afterthought. It would be possible to design these chipsets so that you had to do something fairly stupid to lock them up, but instead they're incredibly sensitive to everything from timing to data layout.

The vendors own drivers aren't able to make this stuff actually stable, as gamers on Windows soon learn. You can expect that any new game will cause a few crashes or lockups until the drivers get tweaked to make them less likely or you lose interest in playing. This isn't with "known bad" hardware either, a mid-range IBM Thinkpad laptop running World of Warcraft might crash to the Windows "Your video system died and I can't fix it" screen once a week or so, leaving the player to reboot their PC. Players just learn that this is "normal" and put up with it.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 12:48 UTC (Fri) by drag (subscriber, #31333) [Link]

Hopefully improvements in technology will render most of the traditional 'discrete hardware acceleration' mostly obsolete.

If you think about it people have been saying that 'software rendering' is the future for a long time now. And although very slow software rendering is very stable and gives a very good visual image. (if you don't mind slideshows, of course)

Intel seems to want to go with a approach that leads to a massive number of very simple x86-like cores that able to perform massive amounts of parrallel graphic-related calculations very quickly.

Make x86-like to make it easy for normal folks to program. Have them all pipelined together so you can essentially program your own GPU to match what you need. Then since it's generic you can program it for lots of different sorts of stuff like media encoding or maybe OpenRT (realtime raytracing).

Then since they are x86-like it makes sense to me that with some code optimizations you could recompile Mesa to use these devices and get essentially hardware-acceleration speeds with these cores. Pretty much remove most of the distinction between GPU and CPUs.

They are aiming to release a discrete video card in 2009 that is designed to compete head to head with Nvidia. Supposedly up to 16 cores for the high-end model. Several times more powerfull then the current high-end cards.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 20:33 UTC (Fri) by nix (subscriber, #2304) [Link]

Sounds more like a Cell's SPUs than anything else to me, only a SPU with
high-speed access to the video controller.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 22:56 UTC (Fri) by drag (subscriber, #31333) [Link]

Well you know, in terms of cpu technology IBM has always lead the pack.

I just hope that Intel does open up everything enough that just your normal average C hacker will be able to make stuff to configure and run on these cores themselves.

It would be a very cool to be able to leverage your video card's proccessing power to do special things like rendering raytracing, or do media encoding and such things. I think that we'd be a lot further along in terms of 3d graphics and virtual environments, also, if it wasn't for these things being so damn proprietary all the time.

Free software drivers for the Intel 965GM Express Chipset

Posted May 12, 2007 9:36 UTC (Sat) by njs (subscriber, #40338) [Link]

Modern graphics cards AFAICT basically *are* general purpose processors running a custom-built software renderer -- the hardware is a bit different from your average CPU (massively parallel array of little chips with weird instruction sets and cache structure), but whatever. It's a supercomputer, except it retails at ~$1k and it goes in a PCI slot. (I know at least one person who is literally using a high-end NVidia to replace an honest-to-goodness cluster).

It'd be nice if it were possible to get at those massively parallel CPUs without running Windows, though.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 12:48 UTC (Fri) by rankincj (subscriber, #4865) [Link]

Interesting - I own a Radeon 9200 and it has definitely been known to crash X. (celestia, WoW etc.) In fact, the other week it brought my entire PC down with it! However, I have been assuming that this is because I am on Mesa's bleeding 7.0 edge.

May I ask which versions of X and Mesa you are running, please?

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 16:04 UTC (Fri) by AJWM (subscriber, #15888) [Link]

Just as a datapoint, I have a 9250 and haven't had any problems - celestia, flightgear flightsim, etc. are just fine. Mesa 6.2.1 and the card's in an AMD64 box. Using the open source driver, of course. Mostly though I just do 2D, I only picked up the card for flightsim.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 20:35 UTC (Fri) by nix (subscriber, #2304) [Link]

Mesa 6.2.1 is a bit old, isn't it? Do you mean 6.5.1?

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 23:54 UTC (Fri) by AJWM (subscriber, #15888) [Link]

No, 6.2.1 is correct. It's a relatively old install (Suse 10.0) and I haven't got around to updating it yet.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 20:35 UTC (Fri) by nix (subscriber, #2304) [Link]

No trouble with xserver 1.1.1 or 1.2.0 and xf86-video-ati 6.6.3 combined
with Mesa 6.5.0 or 6.5.2. (All compiled from upstream sources.)

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 15:16 UTC (Fri) by bobort (subscriber, #5019) [Link]

We have 2 G965 machines and 7 r200 machines, and they all crash. The r200 bug has been well known for 2-3 years and many others have reported it (it involves the X server spinning the CPU with a particular set of ioctls, it's pretty easy to recognize). It happens much more often when there are 2 DRI apps running simultaneously. I have an r200 at home that I've never seen crash, but I hardly ever use DRI there.

The G965 machines crash 3-5 times/day on occasion when somebody is really using the 3D hard. I've seen several reports of this same crash from others as well (it emits a DRI error before hanging, so it's also pretty easy to spot). I have higher hopes for this to get fixed in the future, G965 is still pretty new. I doubt the r200 crash will ever be fixed at this point, it's probably a hardware bug that ATI isn't disclosing the workaround for.

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 17:02 UTC (Fri) by airlied (subscriber, #9104) [Link]

A lot of the r200 crashes were AGP bugs that the open source world never found about it.. so r200 in Intel AGP might be fine, stick it in via or amd and it sucks..

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 20:37 UTC (Fri) by nix (subscriber, #2304) [Link]

00:01.0 PCI bridge: VIA Technologies, Inc. VT8366/A/7 [Apollo KT266/A/333
AGP]

So it's not just that VIA sucks (although this VIA-based Athlon 4 system
has been persistently prone to spontaneous lockups and is terribly
overheating-sensitive: CPU temp above about 50C means a lockup, so it
needs a hugely overspecced fan. Perhaps I should underclock it.)

Free software drivers for the Intel 965GM Express Chipset

Posted May 11, 2007 23:07 UTC (Fri) by drag (subscriber, #31333) [Link]

If you have the space for it I suggest taking a look at the massive scythe ninja heatsink.
http://www.silentpcreview.com/article251-page1.html

These things were originally designed for passive cooling and are just very massive. Suprisingly they aren't not heavy, you don't have to worry about the weight of the thing tweaking heatsink mounts or anything like that.

I have one on my 2.8ghz dual core Pentium-D proccessor. Pentium-Ds are infamous for being a very hot CPU, essentially being 2 Pentium4 cores.

I made a small duct for it out of stiff paper going back to the rear exhaust fan, which is a 800rpm 120mm fan. Then directly above the cpu I have the intake for the powersupply, which is another slow 120mm fan.

Those two fans are the only fans on my system and they keep my cpu ticking at barely above room tempurature. It's just amazing how efficient these heatpipe heatsinks can be.

The only serious problem is that you need to make sure that adiquate cooling is aviable for the big power conditioners/transistors typically found close to the cpu socket. If you overclock these will get very hot without air directly blowing down on them like you find a traditional cpu heatsink. Small aluminum heatsinks with thermal epoxy are more then adiquate to take care of that.

but I don't overclock so it's a non-issue.

that's another thing I like a lot about the onboard video: No need for a VGA card fan. This does wonders for making things quieter.

My old Dell 600mhz firewall hidden in the closet makes more noise through the door then my desktop currently does when running full-tilt.

Free software drivers for the Intel 965GM Express Chipset

Posted May 23, 2007 12:20 UTC (Wed) by daenzer (subscriber, #7050) [Link]

> The r200 bug has been well known for 2-3 years and many others have
> reported it (it involves the X server spinning the CPU with a particular
> set of ioctls, it's pretty easy to recognize).

Those are the symptoms of a GPU lockup, which can be caused by a lot of different things.

Copyright © 2008, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds