Empathy List Archives

R

rew

Sun, Nov 15, 2015 10:28 AM

I upgraded my machine a while ago, and didn't work with openscad for a while.
But now my new machine has lousy 3D performance in OpenScad. Using the new
video card instead of the onboard video has improved the video playback
performance a lot (I was lazy putting it in, and I had output, so why
bother).

The Video card has

01:00.0 VGA compatible controller: NVIDIA Corporation GF119 [GeForce GT 610]
(rev a1)

and I'm running ubuntu 15.10.

I was just at a friend, and his computer was always terribly slow compared
to mine. Now it's the other way around. At work the intern (me too actually)
was expecting his old workstation to be a bit laggy for 3D openscad work,
but again: Much better performance than my workstation now

The symptoms are that even reasonably simple models drop the redraw
framerate to below 5-10FPS and a "swing around" to view the other side of
the model does not follow the mouse movement, but buffers a stream of mouse
events resulting in a several seconds of "slideshow watching" /after/
releasing the mouse!

So: What can I do to diagnose this? Or is the GF119 just a terrible graphics
chip?

I've just googled for some benchmark results, and my GF119 should perform
1.5 times better than my friend's G84GL / Quadro FX 1700.

--
View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505.html
Sent from the OpenSCAD mailing list archive at Nabble.com.

I upgraded my machine a while ago, and didn't work with openscad for a while. But now my new machine has lousy 3D performance in OpenScad. Using the new video card instead of the onboard video has improved the video playback performance a lot (I was lazy putting it in, and I had output, so why bother). The Video card has 01:00.0 VGA compatible controller: NVIDIA Corporation GF119 [GeForce GT 610] (rev a1) and I'm running ubuntu 15.10. I was just at a friend, and his computer was always terribly slow compared to mine. Now it's the other way around. At work the intern (me too actually) was expecting his old workstation to be a bit laggy for 3D openscad work, but again: Much better performance than my workstation now The symptoms are that even reasonably simple models drop the redraw framerate to below 5-10FPS and a "swing around" to view the other side of the model does not follow the mouse movement, but buffers a stream of mouse events resulting in a several seconds of "slideshow watching" /after/ releasing the mouse! So: What can I do to diagnose this? Or is the GF119 just a terrible graphics chip? I've just googled for some benchmark results, and my GF119 should perform 1.5 times better than my friend's G84GL / Quadro FX 1700. -- View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505.html Sent from the OpenSCAD mailing list archive at Nabble.com.

TP

Torsten Paul

Sun, Nov 15, 2015 2:52 PM

On 11/15/2015 11:28 AM, rew wrote:

So: What can I do to diagnose this? Or is the GF119 just a
terrible graphics chip?

No, the graphics chip is probably not to blame. There is an issue
that seems to come from a combination of graphics driver, Qt5
and how OpenSCAD draws the screen.

Currently every mouse event triggers a redraw, due to some changes
in Qt that might be synchronized with the screen refresh causing
the delayed render until everything is processed (well, that's the
guess at least).

We tried to disable the buffer sync in Qt but that does not seem
to work always (or maybe in newer versions?).

You could check if there's an option in the graphics driver to
disable the screen refresh sync. Maybe that helps.

Also: The when compiling from source, the debug version tends to
cause an additional slowdown. Generating the make files with
"qmake CONFIG-=debug" can help with that. For that reason the
nightly snapshots are built with that option too.

ciao,
Torsten.

On 11/15/2015 11:28 AM, rew wrote: > So: What can I do to diagnose this? Or is the GF119 just a > terrible graphics chip? > No, the graphics chip is probably not to blame. There is an issue that seems to come from a combination of graphics driver, Qt5 and how OpenSCAD draws the screen. Currently every mouse event triggers a redraw, due to some changes in Qt that might be synchronized with the screen refresh causing the delayed render until everything is processed (well, that's the guess at least). We tried to disable the buffer sync in Qt but that does not seem to work always (or maybe in newer versions?). You could check if there's an option in the graphics driver to disable the screen refresh sync. Maybe that helps. Also: The when compiling from source, the debug version tends to cause an additional slowdown. Generating the make files with "qmake CONFIG-=debug" can help with that. For that reason the nightly snapshots are built with that option too. ciao, Torsten.

R

rew

Sun, Nov 15, 2015 6:15 PM

Update: Compiled from source, With the "DEBUG" disabled as instructed.

Same slowness/laggyness. Openscad seems to be requiring about 10% of my CPU
during the "laggy updates".... it's not starved for CPU. (the X server does
not show up in "top").

--
View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14516.html
Sent from the OpenSCAD mailing list archive at Nabble.com.

Update: Compiled from source, With the "DEBUG" disabled as instructed. Same slowness/laggyness. Openscad seems to be requiring about 10% of my CPU during the "laggy updates".... it's not starved for CPU. (the X server does not show up in "top"). -- View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14516.html Sent from the OpenSCAD mailing list archive at Nabble.com.

R

rew

Sun, Nov 15, 2015 8:15 PM

Update:
There is "sync-to-vblank" enabled in my nouveau driver. It is said that this
can be disabled by setting
Option "GLXVBlank" "Off"
in xorg.conf, which on modern systems is "any file in
/usr/share/X11/xorg.conf.d", but I haven't been able to make a difference.
On startup the X server still reports stuff about the syncing being enabled.
I have installed the old Ubuntu 14.04 Trusty Tahr. Trusty indeed: the
openscad from there is older, but has reasonable performance. Not great, but
reasonable.

For me I have a workaround at the moment.

Update2: After installing lots of devel packages (even after apt-get
build-dep openscad), current openscad compiles on ubuntu 14.04, and does not
have the laggyness.

--
View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14519.html
Sent from the OpenSCAD mailing list archive at Nabble.com.

Update: There is "sync-to-vblank" enabled in my nouveau driver. It is said that this can be disabled by setting Option "GLXVBlank" "Off" in xorg.conf, which on modern systems is "any file in /usr/share/X11/xorg.conf.d", but I haven't been able to make a difference. On startup the X server still reports stuff about the syncing being enabled. I have installed the old Ubuntu 14.04 Trusty Tahr. Trusty indeed: the openscad from there is older, but has reasonable performance. Not great, but reasonable. For me I have a workaround at the moment. Update2: After installing lots of devel packages (even after apt-get build-dep openscad), current openscad compiles on ubuntu 14.04, and does not have the laggyness. -- View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14519.html Sent from the OpenSCAD mailing list archive at Nabble.com.

TP

Torsten Paul

Sun, Nov 15, 2015 9:24 PM

On 11/15/2015 09:15 PM, rew wrote:

Update2: After installing lots of devel packages (even after apt-get
build-dep openscad), current openscad compiles on ubuntu 14.04, and
does not have the laggyness.

Is that compiled with Qt4?
(That info is shown in Help->Library Info)

ciao,
Torsten.

On 11/15/2015 09:15 PM, rew wrote: > Update2: After installing lots of devel packages (even after apt-get > build-dep openscad), current openscad compiles on ubuntu 14.04, and > does not have the laggyness. > Is that compiled with Qt4? (That info is shown in Help->Library Info) ciao, Torsten.

R

rew

Wed, Nov 18, 2015 9:42 AM

Hi Torsten,

Yes, that's with QT 4.8.6.

Other hint.... After "rendering" the performance of the QT5 compiled version
is great.

Remember me saying that the trusty version was acceptable, well this is much
much better! I now have a simple model that shows the "seconds of delays"
behavior in preview mode, acceptable performance in the old version, and
great performance after being "rendered".

--
View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14617.html
Sent from the OpenSCAD mailing list archive at Nabble.com.

Hi Torsten, Yes, that's with QT 4.8.6. Other hint.... After "rendering" the performance of the QT5 compiled version is great. Remember me saying that the trusty version was acceptable, well this is much much better! I now have a simple model that shows the "seconds of delays" behavior in preview mode, acceptable performance in the old version, and great performance after being "rendered". -- View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14617.html Sent from the OpenSCAD mailing list archive at Nabble.com.

BP

Barry Peet

Wed, Nov 18, 2015 10:00 PM

LS,

could you please remove me from the mailinglist?

regards Barry

On 18 nov. 2015, at 10:42, rew R.E.Wolff@BitWizard.nl wrote:

Hi Torsten,

Yes, that's with QT 4.8.6.

Other hint.... After "rendering" the performance of the QT5 compiled version
is great.

Remember me saying that the trusty version was acceptable, well this is much
much better! I now have a simple model that shows the "seconds of delays"
behavior in preview mode, acceptable performance in the old version, and
great performance after being "rendered".

--
View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14617.html
Sent from the OpenSCAD mailing list archive at Nabble.com.

OpenSCAD mailing list
Discuss@lists.openscad.org
http://lists.openscad.org/mailman/listinfo/discuss_lists.openscad.org

LS, could you please remove me from the mailinglist? regards Barry > On 18 nov. 2015, at 10:42, rew <R.E.Wolff@BitWizard.nl> wrote: > > Hi Torsten, > > Yes, that's with QT 4.8.6. > > Other hint.... After "rendering" the performance of the QT5 compiled version > is great. > > Remember me saying that the trusty version was acceptable, well this is much > much better! I now have a simple model that shows the "seconds of delays" > behavior in preview mode, acceptable performance in the old version, and > great performance after being "rendered". > > > > > > -- > View this message in context: http://forum.openscad.org/OpenScad-3D-performance-tp14505p14617.html > Sent from the OpenSCAD mailing list archive at Nabble.com. > > _______________________________________________ > OpenSCAD mailing list > Discuss@lists.openscad.org > http://lists.openscad.org/mailman/listinfo/discuss_lists.openscad.org

discuss@lists.openscad.org

OpenScad 3D performance.