Video Performance Tool and Use - GLXGEARS

Problems and successes with specific brands/models of computer video hardware
Post Reply
Message
Author
gcmartin

Video Performance Tool and Use - GLXGEARS

#1 Post by gcmartin »

ImageWhy use it???
I have watched, over the years, results of one tool, “GLXGEARS

noryb009
Posts: 634
Joined: Sat 20 Mar 2010, 22:28

#2 Post by noryb009 »

I think people set it up however they want, then compare it with their last score from the previous puppy. If the score went way down, something might be wrong. If it went way up, something good happened.

I'm not for sure, though.
[url=https://github.com/noryb009/lick/releases/latest]LICK - Install Puppy Linux from Windows[/url]

gcmartin

anomaly in using GLXGEARS

#3 Post by gcmartin »

I have found that GLXGEARS has been used inadvertently in the threads in the Puppy forum. Everyone has very good intentions by demonstrating its running report, but, (AND I DONT MEAN THIS PARAGRAPH IN ANY DEROGATORY SENSE AND NOONE IS TO TAKE IT AS OFFENSIVE; AS, THIS IS JUST SOME INFORMATIVE INFORMATION FOR US TO CONSIDER HOW BEST WE CAN IMPROVE THE USE OF ANY TOOL LIKE THIS), but, the GLXGEARS information is meaningless (AND I HAVE REPORTED THIS IN THE SAME FASHION IN THE PAST AS EVERYONE ELSE).

In essence, we have been reporting GLXGEARS numbers in the past without any "baseline" in which to measure it. Thus, its just a number. And a meaningless one at that.

When I started thinking about this end of last year, I started to think about how to develop a process by which this tool's report could be more meaningful, without being complicated by a complex procedure for data capture, reporting, comparison, and objective analysis of graphic card performance on a given PUP.

2 weeks ago, I thought I was on to a reasonable process to capture, report and compare. But, then last eve, I found that my plan had to be thrown out because there are too many unknowns to give accuracy in the process.

First off, GLXGEARS does not come out-of-the-box with every Puppy distros. ONLY some of them.

Secondly and most importantly, what I assumed to be expected with GLXGEARS was that it had a built-in selection of a measurement that would be the same no matter what PC anyone used it on, and no matter what window manager (assuming you account for this in your process by keeping all measurements categorized by WM) you employed within "X".

But, my investigations showed me that the base assumption was invalid, thus my process was structured on an "invalid base".

The following should make it obvious to everyone why.: Look at the desktop numbers and look at what happen while running with each milestone noted by "<.....".

Code: Select all

sh-4.1# glxgears
2679 frames in 5.2 seconds	< start normal window to desktop
2549 frames in 5.0 seconds
3245 frames in 5.0 seconds
3280 frames in 5.0 seconds
3242 frames in 5.0 seconds
3248 frames in 5.0 seconds
3254 frames in 5.0 seconds
3243 frames in 5.0 seconds
3271 frames in 5.0 seconds
4432 frames in 5.0 seconds	< switch to minimize normal window
6587 frames in 5.0 seconds
6589 frames in 5.0 seconds
6561 frames in 5.0 seconds
6568 frames in 5.0 seconds
6569 frames in 5.0 seconds
6593 frames in 5.0 seconds
6584 frames in 5.0 seconds
5931 frames in 5.0 seconds
3263 frames in 5.0 seconds	< switch normal back to desktop
3240 frames in 5.0 seconds
3236 frames in 5.0 seconds
726 frames in 5.1 seconds	< expand normal size to length of screen
343 frames in 5.5 seconds
322 frames in 5.2 seconds
323 frames in 5.2 seconds
347 frames in 5.6 seconds
341 frames in 5.5 seconds
314 frames in 5.1 seconds
313 frames in 5.1 seconds
590 frames in 5.1 seconds	< switch length of screen to minimize 
1113 frames in 5.0 seconds
1149 frames in 5.2 seconds
1136 frames in 5.1 seconds
1151 frames in 5.0 seconds
1085 frames in 5.1 seconds
1115 frames in 5.0 seconds
1142 frames in 5.2 seconds
1061 frames in 5.4 seconds
311 frames in 5.0 seconds	< switch to full screen
325 frames in 5.3 seconds
320 frames in 5.2 seconds
320 frames in 5.2 seconds
332 frames in 5.6 seconds
316 frames in 5.2 seconds
334 frames in 5.5 seconds
350 frames in 5.3 seconds
602 frames in 5.0 seconds	< switch to minimize full screen
580 frames in 5.2 seconds
594 frames in 5.1 seconds
614 frames in 5.3 seconds
601 frames in 5.2 seconds
610 frames in 5.3 seconds
602 frames in 5.2 seconds
483 frames in 5.2 seconds
680 frames in 5.1 seconds
1113 frames in 5.1 seconds
659 frames in 5.3 seconds
316 frames in 5.1 seconds
171 frames in 5.1 seconds
6051 frames in 5.0 seconds
9362 frames in 5.0 seconds	< reduce screen to cover large gear only
9405 frames in 5.0 seconds
9294 frames in 5.0 seconds
9417 frames in 5.0 seconds
9363 frames in 5.0 seconds
9376 frames in 5.0 seconds

sh-4.1# 
Until we can come up with a way to use GLXGEARS as a graphic measurement tool where it has consistency in measurements within any distro, its information that we gather is useless.

The above results can be done on ANY PUP which has GLXGEARS. It should demonstrate its inconsistency in not just your changing GLXGEARS window size, but also, inconsistency when changing graphics drivers; say from Xvesa to Xorg_High. And also from changing resolution, and from changing any video setting and trying t have a stable base to measure against.

Does anyone have any idea of how this tool can be used in a consistent fashion for data gathering to "measurement comparison"?

Without any consistent measurement tool, its any wonder that the community has some of the video issues we see. Further, I, personally would like to know what tool I can use to let me know if a video change has positively or negatively affected my system's performance and my video's performance. Eye checks and emotional checks were never good indicators, although they give a place to start observations.

Thus, if we are to use any tool to measure video performance on our PUPs, we need to have tool which removes from the user’s hands those things which would affect consistency. This consistency must be in operation, reporting, capture, presentation of comparison, with all information necessary to show that tool use is consistent on each capture

Right now, with GLXGEARS, we do not, yet, possess the proper process for its use such that it should be considered to be effective.

Please offer any knowledge, observations, insights, or tests that you feel would be useful to readers of this thread.


Thanks in advance
Last edited by gcmartin on Wed 27 Apr 2011, 00:56, edited 1 time in total.

User avatar
rjbrewer
Posts: 4405
Joined: Tue 22 Jan 2008, 21:41
Location: merriam, kansas

Re: anomaly in using GLXGEARS

#4 Post by rjbrewer »

gcmartin wrote:
Does anyone have any idea of how this tool can be used in a consistent fashion for data gathering to "measurement comparison"?
Not that I'm aware of.

A little bit of Google searching might have led you to the same
conclusion in the first place.

Must be something better out there.
Best of luck.

Inspiron 700m, Pent.M 1.6Ghz, 1Gb ram.
Msi Wind U100, N270 1.6>2.0Ghz, 1.5Gb ram.
Eeepc 8g 701, 900Mhz, 1Gb ram.
Full installs

User avatar
James C
Posts: 6618
Joined: Thu 26 Mar 2009, 05:12
Location: Kentucky

#5 Post by James C »

Here's a good article on the subject
http://wiki.cchtml.com/index.php/Glxgea ... _Benchmark

User avatar
r1tz
Posts: 162
Joined: Thu 09 Sep 2010, 05:19
Location: In #puppylinux (IRC)

#6 Post by r1tz »

imo, glxgears is to test if you are using the right drivers.

like with my intel gpu, glxgears give a low fps when i use software rendering.

and it gives x10 the number with the correct drivers installed.

gcmartin

#7 Post by gcmartin »

@r1tz I understand that you are using GLXGEARS. Like you, I had "wrongly" assumed that there was some extremely good accuracy in its use. Unfortunately, I was wrong in my assumption.

Anyone of us can continue to use the tool. But, I am bringing some "inexperienced" people (like me) information that indicates that what we thought was a valid tool with valid information cannot be relied upon as we may had come to believe.

I began using it as a newbie as everyone seem to be reporting it with new distros, new driver PETs, new video whatevers. I had assumed that since everyone seem to be reporting its numbers, that this was valid information to report.

I am sharing in this thread, that my assumption has turned out to be invalid; simply put, wrong. Because everyone else was doing it, I began adopting a bad habit in trying to begin to rely on these reports as having some value.

You may want to try these out for yourself so that you can see how your report's numbers can vary wildly, while the performance may be just the opposite of the report's. For example, try running it, then hide the spinning screen. Look at the numbers, then bring the screen back and change that screen size on the desktop while watching the number in the terninal window. You, like the rest of us, will observe fps changes that are dramatic. That's just a simple test using a single driver. Next, change driver to XVesa, and do the same thing. Then imagine what you need to say as you try to realize how to explain the phenomenon, firstly, Then next try to determine how you can test "apples to apples" as you change and test drivers knowing that each driver handles screen differently with difference gradients and different buffers and different software and different.... GLXGEARS doesn't provide you what we would need to do either of these two things I ask you to imagine/determine.

Another simple test I found to help in my findings is to run with your startup screen resolution. Save the fps report. Then change the resolution on the same driver and rerun GLXGEARS. Save these fps numbers, too. Now compare the 2 fps reports.

If you or anyone believes that there is a way to use GLXGEARS for performance comparison, I'm open to any suggestions that we can test.

@James C, that post you send and a link in that post's page is excellent in giving a raw description and a recommendation of some tools which may prove to be better for when we try to measure driver performance in Puppy. Thanks, that is an excellent lead.

Hope this helps

Post Reply