9 comments

[ 3.9 ms ] story [ 31.9 ms ] thread
Man, this was my last hope at having decently-performing graphics hardware with well-written and well-supported Linux drivers.

Maybe someday AMD will release enough information about ATi's cards that the open-source drivers will become as reliable and featureful as Intel's...

For some AMD cards, the documentation is fairly complete. I think the main problem is that end-to-end support (OpenGL -> X server integration -> kernel module) is simply a huge task these days, and the Mesa codebase in particular is massive. I'm considering helping out in that effort, as my background in working with game console GPUs might be an advantage. Depends how much time and money I can realistically invest in it.
Not sure what you mean, NVIDIA hardware works great on Linux and I use it on a daily basis.

Are you referring to the fact that NVIDIA's drivers aren't open-source? Well, there is no reason to expect that Larrabee's would have been.

In any case, there is probably nothing interesting in the driver source code anyway. (A lot of "copy this buffer of instructions to the GPU, this buffer back" and not much else.)

My understanding is that a lot of new X11 features (such as XRandR, GEM memory managment, kernel mode-setting and so forth) are added to the core and to the open-source drivers, and closed-source drivers lag behind significantly. Also, I've a friend with an nVidia-based laptop, and suspend/resume apparently works about one time in ten, while it worked flawlessly on my Intel-based laptop.

Also, apart from the Paulsbro/GMA500 fiasco (which was contracted out to a third party), all Intel's graphics hardware has had open-source drivers written by Intel themselves, so it doesn't seem too extreme to expect similar support for Larrabee.

> it doesn't seem too extreme to expect similar support for Larrabee

You have a point, but I think the reason NVIDIA doesn't open-source the drivers is the same as the reason Larrabee probably wouldn't open-source them - i.e. fear of releasing the details of the GPU design to the competition. Current Intel's graphics has nothing to hide from NVIDIA, but Lrrabee would. But of course, this is just speculation, especially now that no consumer Larrabee is on the horizon...

That's too bad. More competition in the consumer market would probably have been good for consumers and for the pace of technological progress.
The article mentions a shift in focus, from consumer to HPC/Development.

I doubt it's gone forever, maybe just delayed.

I think the consumer market for Larabee is much smaller than HPC, so it makes sense for Intel to chase that money instead.

Isn't the great thing about consumer GPU processing like Nvidia's Cuda that it's pretty much everywhere, even in consumer devices? So that some research guy can convert his number-crunching code to run in parallel with Cuda right on his own workstation and see if it works.

If it does, then he can later put it up and running on a more efficient server or cluster. But initially he gets to try if X times speed up works for him, without getting any special hardware just to try out.

Disappointing but not terribly surprising. The programming expertise just isn't there for consumer applications, yet, whereas for HPC many problems come close to being embarrassingly parallel.

Oh well. Another time, perhaps.