A First Look At The 2010 Linux Graphics Survey Results

[kari]

Nvidia drivers

Sligth correction to the text about nVidia drivers usage. Of the nVidia users, about 13,5 % appear to be using nouveau (or 5 % of total users, of which some 38 % use nVidia).

Quite surprising, really. But when I just installed OpenSuse 11.3 on my play machine, it used nouveau as default, and that probably explains it. Many users get along with what the distro installs as default. Don't know really how many distros use nouveau as default, didn't it used to be nv.

[droidhacker]

It IS the job of the driver to function properly when I plug a totally standard Dell monitor into my computer.

Linux driver code is FILLED with exceptional hardware cases.

"totally standard dell monitor"..... doesn't mean anything.
1) DELL is well known as being the lowest of the low.
2) A defect is a defect is a defect.
3) Sure there are workarounds for specific problems. That doesn't mean that this SHOULD be the case, or that "they" (the driver developers) OWE it to you to fix something that was screwed up by SOMEONE ELSE UNRELATED.

Example: You buy a new car and it burns oil. Is the company that makes the oil responsible for developing a magic oil that won't get burnt by your defective car? No! Is it the responsibility of the company that paves the roads to apply magical no-oil-burning voodoo to the pavement? NO! It is the responsibility of your car's manufacturer to fix the oil burning problem!

The only reason why there are insane workaround in the drivers for specific cards or monitors is because the manufacturers of those cards/monitors can't be trusted to take responsibility for THEIR PROBLEMS. And the reason for this is that the REGULATORY BODIES are run by MORONS who don't understand the technology.

[allquixotic]

The multi-select in most of the questions makes the percentages in the pie charts counter-intuitive. We intuitively expect them to represent one quantity, when in fact, the quantity we expect is not published; instead, a less-useful quantity is published.

What Michael posted in his article on the pie charts (the ones that are multi-select) is best explained by an analogy.

Let's say we have a multi-select question with three checkboxes: A, B, and C. For every user that checks A, throw an apple into a cart. For every user that checks B, throw an orange into the cart. For every use that checks C, throw a tomato into the cart. To derive the percentages that Michael displays, count up all the pieces of fruit total, apples+oranges+tomatoes, then divide the number of apples by that total to get the "percentage" of A, etc.

A few weird facts about this method:

• If everyone threw one apple, one orange and one tomato into the cart, then the result would be 33% for each response, regardless of how many people took the survey. But the actual number of people who threw a tomato into the cart would be 100%; similarly for the other two fruits.
• If no one threw any oranges in the cart, but everyone threw both an apple and a tomato into the cart, then the result would be 50% for responses A and C, and 0% for option B. The 0% is correct, but the actual number of people who checked A and C would be 100%.
• If half the people threw in an apple, everyone threw in an orange, and no one threw in a tomato, then by Michael's calculations, the value for apple (option A) would be 25%, the value for orange (option B) would be 75%, and the value for tomato (option C) would be 0%. But how do we derive the actual percentage of people who threw in which pieces of fruit?

Perhaps confusingly, it is not possible for us to derive the actual percentage of people who threw in a given piece of fruit from the data we have. The critical piece of information we are missing in Michael's article is the number of pieces of fruit, i.e., the total number of times any checkbox was checked for each question. This determines what we have to multiply Michael's percentages by to get the total number of people who checked a given answer. Then, we can come up with a meaningful absolute percentage for a single response by dividing this number by the number of respondents (which Michael declared was 6300).

So for example, the Michael-percentage of licensing/open-source was 9.38%, or 0.0938. That's 938/10000, or 469/5000, or 1876/20000. The divisor, if it were the actual fraction of responses vs. total number of checks, would represent the total number of checks; but since we don't know what "scale" to divide the fraction on, we can't come up with a meaningful answer! We can only take educated guesses.

This particular question, "What are your key interests or concerns with Linux video drivers", had 11 choices for the answer. If the total number of checks were 6300, then each respondent would be checking exactly one checkbox, on average. If the total number of checks were 12600, then each respondent would have checked an average of two checkboxes. Given the fact that there are 11 responses, and many of these issues are popular, I think it is probabilistically likely that the average number of checks is more like 5 or 6, given that Phoronix readers are generally interested in multiple use cases, not just one.

If the average number of checks is 5 or 6, that means the total number of checks would speculatively be around 6300 * 5.5 = 34650 or so. If you work 0.0938 * 34650, you get 3250, which would mean that, hypothetically, 3250 / 6300 = 51.587% of the respondents selected the Licensing/open source checkbox! O_O

This is a far cry from Michael's 9.38%.

On the other hand, if most people only selected one checkbox, that would mean that there would be around 6300 checkboxes checked, so 0.0938 * 6300 = 591... so 591 / 6300 brings us back to our 9.38% that Michael has listed down, except in this hypothetical scenario, 9.38% is the percentage of respondents who selected licensing/open source. Therefore, the percentage of respondents who selected licensing/open source is between 1 and 6 times greater than the 9.38% that Michael posted, depending on the average number of checks that people made.

Consider, hypothetically, a scenario where everyone checks 8 or 9 checkboxes, because people are interested in a wide array of features, and everyone is of the opinion that all of these features "need some work" (and hence they check them because they want manufacturers to pay attention to them). In this case, the percentage of people who checked licensing/open source would be even higher -- somewhere around 75 or 80%! So as you see, rolling up all the options relative to each other is not really meaningful when you have this many options.

My conclusion is that the percentages in these pie charts are almost completely meaningless unless we know the exact number of total checkboxes that were checked for each question. Michael, can you please provide this data?

[droidhacker]

Video acceleration more than twice as interesting as Gallium?! >.<

That's inconclusive. Gallium isn't necessarily the only avenue towards video acceleration. It tends to be the favored approach in terms of shader based video acceleration, but wouldn't apply to UVD or VDPAU hardware. Others probably just don't care HOW you implement it so long as it works.

nVidia has more share of the pie, but all the Ati drivers summed up, have more share of the pie than the nVidia blob plus the FLOSS driver ???

Your addition is a little off there.
You probably added nvidia + nv = total rather than nvidia+nouveau+nv.

HOWEVER, the fact remains that there are apparently more users using AMD drivers than AMD cards (36.32% vs 35.47%). Same with nvidia (38.54% vs 38.12%). I wonder how much of this could be rounding errors and how much of this could be due to hybrid graphics?

Ubuntu users have the largest share of the pie (X.org 1.7.6), however it's only 36,21%, so why does Ubuntu recieve so much cobverage ???

WTF are you talking about? xorg 1.7.6 does NOT correspond to "ubuntu". ANY distro can use ANY version of xorg. Fedora 12 also uses xorg 1.7.6.

13,62% of all people here hook their PC up to their TV? Is this a mistake? HDMI is not TV-out, right?

I didn't like that question at all. Couldn't tell if he meant specifically tv-out as in S-Video, or if he meant "hooked up to a tv" in general, i.e. via HDMI/DVI. I'm sure that it was taken both ways.

65,40% of everyone who filled in the survey 'only' has a single display, however "Which of the display feature(s) do you use?" says 36.61% of the people have Multiple Display Heads.

I am certain that this was a result of misinterpreting the checkboxes as bullet lists. Most people probably picked the ONE feature they use most rather than picking ALL the features that they use.

And note that the information on that question is presented in a completely insane and INCORRECT way... since the options are not relative to each other... but relative to the total number of respondents to the survey, i.e. the results of tvout+hotplugging+multihead+powersaving do NOT add up to 100%. These questions are INDEPENDENT OF EACH OTHER. They should be presented as a BAR GRAPH as a percent of the total number of surveys presented. Also note that since those numbers have been adjusted to add up to 100%, they are all DEFINITELY INCORRECT.

I believe that some of the other questions are affected by similar errors in presentation.

MICHAEL: Could you please post the RAW DATA IN SPREADSHEET FORM so that we can perform a proper analysis?

Then 5.47% of all people use Gnome-Shell / Mutter, while the Gnome people use Gnome because it's so freaking conservative and stable, but then go on and _use_ GS/Mutter. Care to explain why this is nothing but pure fanboyism?

I don't see the point you're making here.
Where do you come up with the idea that people use gnome because its conservative and/or stable? Compared to KDE, I use gnome because IN MY OPINION, KDE is a freakin UGLY BLOATED MESS. The question appeared to be related to use as a PERSONAL PREFERENCE rather than technical reasoning.

To up the WTF stage even higher: people with modern graphics card that are modern enough (PCI-E) have a processor that's modern enough for video playback.

Huh? Da heck you talking about again? Try decoding 1080p on a SEMPRON 2600. Yep, PCI-E and a weakass single-core 1.6 GHz.

Alternatively, go for a bottom end chip that IS capable of decoding said video, like any dual-core ~2.4 GHz. It'll do the job, but it'll be running full blast, making tons of heat and running your cooling fans at dentist drill speeds and making 10x the noise. Try enjoying your movie with that going on! Not everybody running a PCI-E mainboard has that together with a 4 or 6 core CPU.

So why does one care about video offloading is only for HTPC's,

You think that the only place that you play videos is on HTPC's? How about LAPTOPS? When you're out some place and waiting for something (i.e. doctor's office, airplane, etc.) wouldn't it be nice to be able to watch a movie without it draining your battery dead in 10 minutes and making the sounds of a vacuum cleaner?

but suspend & hibernate (what you must have on an HTPC, is 8.34% of interest, while video offloading is 15.52%.

How does being suspended or hibernated help you to play videos?

And we're not there yet; how much percent uses a compositing window manager? 68.65%, while 23.47% says they use Visual Desktop Effects.

The second question was about what you ***ACTIVELY ENGAGE IN***. That means how many people sit there for hours on end just spinning the cube. That and the compositing doesn't necessarily make the desktop effects. I.e. enabling metacity as a compositing window manager doesn't make any difference in visual effects, it just moves the work over to the GPU.

Conclusion: This survey has "Bullshit" spraypainted all over it.

Not entirely, though it is clear that some parts of the planning process were skipped and that the analysis is definitely out to lunch in a very bad way.

[pingufunkybeat]

Reader survey. Please click all that apply:

[ ] I am a human being
[ ] I am male
[ ] I am female
[ ] I like ice-cream
[ ] I am taller than 5 feet


The analysis of our data showed that only 15% of all our readers are human beings.

This is a surprising result. Hrmmmmm. ...Scratch scratch....

[energyman]

I never understood why people deeply care about open source drivers. Does Microsoft spend any of their resources to develop drivers for various hardwares? To me, developing open source drivers makes sense only if the manufacturer isn't willing to provide Linux drivers. IMHO, writing and improving applications that compete and beat similar applications available on other platforms is better use of talented people's skills.

Cheers

what happens if nvidia decides to only support SLES in the future? What about ubuntu? RHEL? gentoo?
(well, gentoo would have the least problems), Slackware, fedora?

Binary drivers are an unsatisfactory short term solution. You can not depend your future on them.
Microsoft is a bad example because they are a completely different entity.

Not caring about drivers is short sightened - even idiotic.

[droidhacker]

what happens if nvidia decides to only support SLES in the future? What about ubuntu? RHEL? gentoo?
(well, gentoo would have the least problems), Slackware, fedora?

Binary drivers are an unsatisfactory short term solution. You can not depend your future on them.
Microsoft is a bad example because they are a completely different entity.

Not caring about drivers is short sightened - even idiotic.

Microshod is a GREAT example of this....
Back in the days when you used MS, how many times did you buy some piece of hardware to find that it simply could not work on the next version of windoze?

I'll give you an example of such hardware: brother mfc4650. Dead on any microshod newer than win98.

[mutlu_inek]

WRONG and MISLEADING Graphs and Text

Michael,

By now you must have read some of the comments in this thread and thus have realized that you seriously distort and misrepresent some of the data by using pie charts even though the total number of answers does not add up to 100%. This is a grave mistake that leads to serious misinterpretation of the data, not only by the casual reader, but also by you, as can be seen in your interpretations of the charts. This is easily visible in your chart about the "key interests or concerns related to Linux video drivers," in which you claim that "less than 10% of the users are interested in licensing or the open-source status of a driver." This is simply WRONG.

Since you should have become aware of this by now, I would like to ask you to CORRECT this news article. Please stop spreading wrong 'information' about open source drivers on your (unfortunately very important) website. Such FUD may have serious repercussions, since this is what people will keep in mind. Not everyone reads the comments or understands how to use or read charts.

To repeat: PLEASE CORRECT THE ARTICLE!

[Chewi]

I too am amazed that Michael has not so much as commented on this yet.

[allquixotic]

I think Michael is busy with XDS proceedings... he's probably (close to) finding himself in France, so he's got more going on than usual. He's got a lot on his plate. Give him time.

Before this turns into a Michael bashing thread, can people please assume good faith on Michael's part, that he accidentally misinterpreted the results based on his pie charts? Give him the benefit of the doubt that he isn't trying to misrepresent the situation. And give him an extra amount of time to fix the article because he is busy with a significant yearly event right now.

Yes, it's wrong and should be fixed, but no need to get up in arms over it.