CPU load remains high after previewing but not after render
Yes, display performance becomes bearable when preview faceting is kept
very low. I figured I was running into a bug in OpenSCAD or my graphics
driver as the display is quite responsive at vastly higher numbers of faces
generated by a render.
On Sun, Aug 4, 2024 at 10:20 PM Todd Allen speedebikes@gmail.com wrote:
Yes, just checked it is doing software graphics. It is possible to use
GPU graphics but it looks like I have a few hoops to jump through. Why is
displaying a preview so much more demanding than displaying a render?
Anyway for now I'll see if I can get by just making previews very coarse
grained.
On Sun, Aug 4, 2024 at 10:06 PM Chun Kit LAM john.lck40@gmail.com wrote:
Do you have GPU passthrough for your VM? When running under
virtualization, it may run under software rendering. And I doubt if
Fusion360 is much more graphically demanding, openscad preview render is
kind of like unoptimized ray-tracing (from my understanding) which is quite
heavy even for good GPUs, and it will definitely stress your CPU when run
under software rendering.
On 5/8/2024 10:43, Todd Allen wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running under Qubes OS
4.2 with pvh virtualization and seamless integration enabled.
I get the same behavior with both a week old unstable release and the
last stable release 2024.01.06 of OpenSCAD, both are .appimage files.
I think the issue may be a graphics driver as the problem got much worse
after switching from a 1080p monitor to a 2560x1440 monitor.
I have a Windows VM which I rarely use to run Fusion360 which is a much
more graphically demanding app and it runs well. I should try OpenSCAD
there to help isolate if the problem is with the linux graphics driver.
Curiously though when I render the video is smooth and fairly responsive
under linux and it is only when looking at a preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26, chamfer=0.5,
inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer : inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer : outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d0.5+$wall_t0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d0.5) yrot(90)
down($pipe_d0.5) join_prism(joined_shape,base="cylinder",base_d=$pipe_d,
length=h0.6, fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct,
circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h1.04) {
difference() {
zrot_copies(n=3) xrot(ang) down(h0.5+$corner_cut) leg(h,
$pipe_d, $outer_chamfer, $inner_chamfer);
zrot_copies(n=3) xrot(ang) down(h*0.5+$corner_cut)
leg_core(h, $rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if parameters
change this will need tweaking, probably make these values enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45)
cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss <
discuss@lists.openscad.org> wrote:
What is the openscad version, as well as OS and hardware spec? Nightly
version is using a different preview pipeline. And this somehow makes me
wonder if you are using software rendering instead of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss, <
discuss@lists.openscad.org> wrote:
I use the render key but the other two are no longer of much use. If I
can I'll rewrite the <export STL> key to do a render with $export=true
which I can test in script to go higher res and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss <
discuss@lists.openscad.org> wrote:
F5, F6 and F7 do preview, render and export but I think F7 is STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via Discuss <
discuss@lists.openscad.org> wrote:
When working with complex high facet models I've turned off automatic
preview and just use rendering because previewing overheats my CPU and
makes it nearly impossible to adjust the viewport as much of the interface
becomes so sluggish it is practically frozen. I had been taking advantage
of the $preview variable to set the other resolution adjusting vars such as
$fs to minimize resolution for previewing and have a high resolution for
rendering. But now I have been manually changing vars to only have high
res for renders before exporting. What I'd really love to have is a way to
do single keystrokes for a preview render, a full render and an export
render (as an obj and not an stl). I expect I will have to build my own
from source to achieve this but thought I would ask in case I am missing
something and other solutions are possible.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
If you look at the paper of OpenCSG, which is the backend for preview,
you will see that the Goldfeather algorithm has a complexity of O(n^2).
Even with depth complexity optimization, it is still O(kn). And you need
to do that for each pixel (AFAIK). This is why it is slow.
I think you will have better luck running the actual render and do not
use preview. This may be slower at first, but it does not require
running an expensive shader every frame.
On 5/8/2024 11:39, Todd Allen wrote:
Yes, display performance becomes bearable when preview faceting is
kept very low. I figured I was running into a bug in OpenSCAD or my
graphics driver as the display is quite responsive at vastly higher
numbers of faces generated by a render.
On Sun, Aug 4, 2024 at 10:20 PM Todd Allen speedebikes@gmail.com wrote:
Yes, just checked it is doing software graphics. It is possible
to use GPU graphics but it looks like I have a few hoops to jump
through. Why is displaying a preview so much more demanding than
displaying a render? Anyway for now I'll see if I can get by just
making previews very coarse grained.
On Sun, Aug 4, 2024 at 10:06 PM Chun Kit LAM
<john.lck40@gmail.com> wrote:
Do you have GPU passthrough for your VM? When running under
virtualization, it may run under software rendering. And I
doubt if Fusion360 is much more graphically demanding,
openscad preview render is kind of like unoptimized
ray-tracing (from my understanding) which is quite heavy even
for good GPUs, and it will definitely stress your CPU when run
under software rendering.
On 5/8/2024 10:43, Todd Allen wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running
under Qubes OS 4.2 with pvh virtualization and seamless
integration enabled.
I get the same behavior with both a week old unstable release
and the last stable release 2024.01.06 of OpenSCAD, both are
.appimage files.
I think the issue may be a graphics driver as the problem got
much worse after switching from a 1080p monitor to a
2560x1440 monitor.
I have a Windows VM which I rarely use to run Fusion360 which
is a much more graphically demanding app and it runs well. I
should try OpenSCAD there to help isolate if the problem is
with the linux graphics driver. Curiously though when I
render the video is smooth and fairly responsive under linux
and it is only when looking at a preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26,
chamfer=0.5, inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer :
inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer :
outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d*0.5+$wall_t*0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet*0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d*0.5)
yrot(90) down($pipe_d*0.5)
join_prism(joined_shape,base="cylinder",base_d=$pipe_d,
length=h*0.6, fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct,
circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h*1.04) {
difference() {
zrot_copies(n=3) xrot(ang)
down(h*0.5+$corner_cut) leg(h, $pipe_d, $outer_chamfer,
$inner_chamfer);
zrot_copies(n=3) xrot(ang)
down(h*0.5+$corner_cut) leg_core(h, $rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if
parameters change this will need tweaking, probably make
these values enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45)
cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss
<discuss@lists.openscad.org> wrote:
What is the openscad version, as well as OS and hardware
spec? Nightly version is using a different preview
pipeline. And this somehow makes me wonder if you are
using software rendering instead of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss,
<discuss@lists.openscad.org> wrote:
I use the render key but the other two are no longer
of much use. If I can I'll rewrite the <export STL>
key to do a render with $export=true which I can
test in script to go higher res and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss
<discuss@lists.openscad.org> wrote:
F5, F6 and F7 do preview, render and export but
I think F7 is STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via
Discuss <discuss@lists.openscad.org> wrote:
When working with complex high facet models
I've turned off automatic preview and just
use rendering because previewing overheats
my CPU and makes it nearly impossible to
adjust the viewport as much of the interface
becomes so sluggish it is practically
frozen. I had been taking advantage of the
$preview variable to set the other
resolution adjusting vars such as $fs to
minimize resolution for previewing and have
a high resolution for rendering. But now I
have been manually changing vars to only
have high res for renders before exporting.
What I'd really love to have is a way to do
single keystrokes for a preview render, a
full render and an export render (as an obj
and not an stl). I expect I will have to
build my own from source to achieve this but
thought I would ask in case I am missing
something and other solutions are possible.
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email todiscuss-leave@lists.openscad.org
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
Also make sure you have large cache sizes in preferences, re-render will be quicker.
From: Chun Kit LAM via Discuss [mailto:discuss@lists.openscad.org]
Sent: Mon, 5 Aug 2024 14:48
To: Todd Allen
Cc: OpenSCAD general discussion Mailing-list; Chun Kit LAM
Subject: [OpenSCAD] Re: CPU load remains high after previewing but not after render
If you look at the paper of OpenCSG, which is the backend for preview, you will see that the Goldfeather algorithm has a complexity of O(n^2). Even with depth complexity optimization, it is still O(kn). And you need to do that for each pixel (AFAIK). This is why it is slow.
I think you will have better luck running the actual render and do not use preview. This may be slower at first, but it does not require running an expensive shader every frame.
On 5/8/2024 11:39, Todd Allen wrote:
Yes, display performance becomes bearable when preview faceting is kept very low. I figured I was running into a bug in OpenSCAD or my graphics driver as the display is quite responsive at vastly higher numbers of faces generated by a render.
On Sun, Aug 4, 2024 at 10:20 PM Todd Allen speedebikes@gmail.com wrote:
Yes, just checked it is doing software graphics. It is possible to use GPU graphics but it looks like I have a few hoops to jump through. Why is displaying a preview so much more demanding than displaying a render? Anyway for now I'll see if I can get by just making previews very coarse grained.
On Sun, Aug 4, 2024 at 10:06 PM Chun Kit LAM john.lck40@gmail.com wrote:
Do you have GPU passthrough for your VM? When running under virtualization, it may run under software rendering. And I doubt if Fusion360 is much more graphically demanding, openscad preview render is kind of like unoptimized ray-tracing (from my understanding) which is quite heavy even for good GPUs, and it will definitely stress your CPU when run under software rendering.
On 5/8/2024 10:43, Todd Allen wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running under Qubes OS 4.2 with pvh virtualization and seamless integration enabled.
I get the same behavior with both a week old unstable release and the last stable release 2024.01.06 of OpenSCAD, both are .appimage files.
I think the issue may be a graphics driver as the problem got much worse after switching from a 1080p monitor to a 2560x1440 monitor.
I have a Windows VM which I rarely use to run Fusion360 which is a much more graphically demanding app and it runs well. I should try OpenSCAD there to help isolate if the problem is with the linux graphics driver. Curiously though when I render the video is smooth and fairly responsive under linux and it is only when looking at a preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26, chamfer=0.5, inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer : inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer : outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d0.5+$wall_t0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d0.5) yrot(90) down($pipe_d0.5) join_prism(joined_shape,base="cylinder",base_d=$pipe_d, length=h0.6, fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct, circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h1.04) {
difference() {
zrot_copies(n=3) xrot(ang) down(h0.5+$corner_cut) leg(h, $pipe_d, $outer_chamfer, $inner_chamfer);
zrot_copies(n=3) xrot(ang) down(h*0.5+$corner_cut) leg_core(h, $rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if parameters change this will need tweaking, probably make these values enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45) cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss discuss@lists.openscad.org wrote:
What is the openscad version, as well as OS and hardware spec? Nightly version is using a different preview pipeline. And this somehow makes me wonder if you are using software rendering instead of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss, discuss@lists.openscad.org wrote:
I use the render key but the other two are no longer of much use. If I can I'll rewrite the <export STL> key to do a render with $export=true which I can test in script to go higher res and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss discuss@lists.openscad.org wrote:
F5, F6 and F7 do preview, render and export but I think F7 is STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via Discuss discuss@lists.openscad.org wrote:
When working with complex high facet models I've turned off automatic preview and just use rendering because previewing overheats my CPU and makes it nearly impossible to adjust the viewport as much of the interface becomes so sluggish it is practically frozen. I had been taking advantage of the $preview variable to set the other resolution adjusting vars such as $fs to minimize resolution for previewing and have a high resolution for rendering. But now I have been manually changing vars to only have high res for renders before exporting. What I'd really love to have is a way to do single keystrokes for a preview render, a full render and an export render (as an obj and not an stl). I expect I will have to build my own from source to achieve this but thought I would ask in case I am missing something and other solutions are possible.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
I put a render() call around bits that cause the preview frame rate to
drop. It is much less of a problem with the new rendering system.
On Mon, 5 Aug 2024 at 07:19, Michael Marx (spintel) via Discuss <
discuss@lists.openscad.org> wrote:
Also make sure you have large cache sizes in preferences, re-render will
be quicker.
From: Chun Kit LAM via Discuss [mailto:discuss@lists.openscad.org]
Sent: Mon, 5 Aug 2024 14:48
To: Todd Allen
Cc: OpenSCAD general discussion Mailing-list; Chun Kit LAM
Subject: [OpenSCAD] Re: CPU load remains high after previewing but not
after render
If you look at the paper of OpenCSG, which is the backend for preview, you
will see that the Goldfeather algorithm has a complexity of O(n^2). Even
with depth complexity optimization, it is still O(kn). And you need to do
that for each pixel (AFAIK). This is why it is slow.
I think you will have better luck running the actual render and do not use
preview. This may be slower at first, but it does not require running an
expensive shader every frame.
On 5/8/2024 11:39, Todd Allen wrote:
Yes, display performance becomes bearable when preview faceting is kept
very low. I figured I was running into a bug in OpenSCAD or my graphics
driver as the display is quite responsive at vastly higher numbers of faces
generated by a render.
On Sun, Aug 4, 2024 at 10:20 PM Todd Allen speedebikes@gmail.com wrote:
Yes, just checked it is doing software graphics. It is possible to use
GPU graphics but it looks like I have a few hoops to jump through. Why is
displaying a preview so much more demanding than displaying a render?
Anyway for now I'll see if I can get by just making previews very coarse
grained.
On Sun, Aug 4, 2024 at 10:06 PM Chun Kit LAM john.lck40@gmail.com wrote:
Do you have GPU passthrough for your VM? When running under
virtualization, it may run under software rendering. And I doubt if
Fusion360 is much more graphically demanding, openscad preview render is
kind of like unoptimized ray-tracing (from my understanding) which is quite
heavy even for good GPUs, and it will definitely stress your CPU when run
under software rendering.
On 5/8/2024 10:43, Todd Allen wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running under Qubes OS
4.2 with pvh virtualization and seamless integration enabled.
I get the same behavior with both a week old unstable release and the last
stable release 2024.01.06 of OpenSCAD, both are .appimage files.
I think the issue may be a graphics driver as the problem got much worse
after switching from a 1080p monitor to a 2560x1440 monitor.
I have a Windows VM which I rarely use to run Fusion360 which is a much
more graphically demanding app and it runs well. I should try OpenSCAD
there to help isolate if the problem is with the linux graphics driver.
Curiously though when I render the video is smooth and fairly responsive
under linux and it is only when looking at a preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26, chamfer=0.5,
inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer : inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer : outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d0.5+$wall_t0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d0.5) yrot(90)
down($pipe_d0.5) join_prism(joined_shape,base="cylinder",base_d=$pipe_d,
length=h0.6, fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct, circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h1.04) {
difference() {
zrot_copies(n=3) xrot(ang) down(h0.5+$corner_cut) leg(h,
$pipe_d, $outer_chamfer, $inner_chamfer);
zrot_copies(n=3) xrot(ang) down(h*0.5+$corner_cut) leg_core(h,
$rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if parameters
change this will need tweaking, probably make these values enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45)
cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss <
discuss@lists.openscad.org> wrote:
What is the openscad version, as well as OS and hardware spec? Nightly
version is using a different preview pipeline. And this somehow makes me
wonder if you are using software rendering instead of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss, <
discuss@lists.openscad.org> wrote:
I use the render key but the other two are no longer of much use. If I
can I'll rewrite the <export STL> key to do a render with $export=true
which I can test in script to go higher res and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss <
discuss@lists.openscad.org> wrote:
F5, F6 and F7 do preview, render and export but I think F7 is STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via Discuss <
discuss@lists.openscad.org> wrote:
When working with complex high facet models I've turned off automatic
preview and just use rendering because previewing overheats my CPU and
makes it nearly impossible to adjust the viewport as much of the interface
becomes so sluggish it is practically frozen. I had been taking advantage
of the $preview variable to set the other resolution adjusting vars such as
$fs to minimize resolution for previewing and have a high resolution for
rendering. But now I have been manually changing vars to only have high
res for renders before exporting. What I'd really love to have is a way to
do single keystrokes for a preview render, a full render and an export
render (as an obj and not an stl). I expect I will have to build my own
from source to achieve this but thought I would ask in case I am missing
something and other solutions are possible.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
Additionally, CPU usage is constantly higher, WHILE you are in measuring
mode.
During measuring mode, OpenSCAD is constantly evaluating, which part of
your design is below the mouse cursor.
Press "ESC" to finish measuring mode and return to low cpu usage.
On Mon, Aug 5, 2024 at 9:15 AM nop head via Discuss <
discuss@lists.openscad.org> wrote:
I put a render() call around bits that cause the preview frame rate to
drop. It is much less of a problem with the new rendering system.
On Mon, 5 Aug 2024 at 07:19, Michael Marx (spintel) via Discuss <
discuss@lists.openscad.org> wrote:
Also make sure you have large cache sizes in preferences, re-render will
be quicker.
From: Chun Kit LAM via Discuss [mailto:discuss@lists.openscad.org]
Sent: Mon, 5 Aug 2024 14:48
To: Todd Allen
Cc: OpenSCAD general discussion Mailing-list; Chun Kit LAM
Subject: [OpenSCAD] Re: CPU load remains high after previewing but not
after render
If you look at the paper of OpenCSG, which is the backend for preview,
you will see that the Goldfeather algorithm has a complexity of O(n^2).
Even with depth complexity optimization, it is still O(kn). And you need to
do that for each pixel (AFAIK). This is why it is slow.
I think you will have better luck running the actual render and do not
use preview. This may be slower at first, but it does not require running
an expensive shader every frame.
On 5/8/2024 11:39, Todd Allen wrote:
Yes, display performance becomes bearable when preview faceting is kept
very low. I figured I was running into a bug in OpenSCAD or my graphics
driver as the display is quite responsive at vastly higher numbers of faces
generated by a render.
On Sun, Aug 4, 2024 at 10:20 PM Todd Allen speedebikes@gmail.com wrote:
Yes, just checked it is doing software graphics. It is possible to use
GPU graphics but it looks like I have a few hoops to jump through. Why is
displaying a preview so much more demanding than displaying a render?
Anyway for now I'll see if I can get by just making previews very coarse
grained.
On Sun, Aug 4, 2024 at 10:06 PM Chun Kit LAM john.lck40@gmail.com
wrote:
Do you have GPU passthrough for your VM? When running under
virtualization, it may run under software rendering. And I doubt if
Fusion360 is much more graphically demanding, openscad preview render is
kind of like unoptimized ray-tracing (from my understanding) which is quite
heavy even for good GPUs, and it will definitely stress your CPU when run
under software rendering.
On 5/8/2024 10:43, Todd Allen wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running under Qubes OS
4.2 with pvh virtualization and seamless integration enabled.
I get the same behavior with both a week old unstable release and the
last stable release 2024.01.06 of OpenSCAD, both are .appimage files.
I think the issue may be a graphics driver as the problem got much worse
after switching from a 1080p monitor to a 2560x1440 monitor.
I have a Windows VM which I rarely use to run Fusion360 which is a much
more graphically demanding app and it runs well. I should try OpenSCAD
there to help isolate if the problem is with the linux graphics driver.
Curiously though when I render the video is smooth and fairly responsive
under linux and it is only when looking at a preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26, chamfer=0.5,
inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer : inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer : outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d0.5+$wall_t0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d0.5) yrot(90)
down($pipe_d0.5) join_prism(joined_shape,base="cylinder",base_d=$pipe_d,
length=h0.6, fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct,
circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h1.04) {
difference() {
zrot_copies(n=3) xrot(ang) down(h0.5+$corner_cut) leg(h,
$pipe_d, $outer_chamfer, $inner_chamfer);
zrot_copies(n=3) xrot(ang) down(h*0.5+$corner_cut)
leg_core(h, $rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if parameters
change this will need tweaking, probably make these values enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45)
cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss <
discuss@lists.openscad.org> wrote:
What is the openscad version, as well as OS and hardware spec? Nightly
version is using a different preview pipeline. And this somehow makes me
wonder if you are using software rendering instead of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss, <
discuss@lists.openscad.org> wrote:
I use the render key but the other two are no longer of much use. If I
can I'll rewrite the <export STL> key to do a render with $export=true
which I can test in script to go higher res and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss <
discuss@lists.openscad.org> wrote:
F5, F6 and F7 do preview, render and export but I think F7 is STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via Discuss <
discuss@lists.openscad.org> wrote:
When working with complex high facet models I've turned off automatic
preview and just use rendering because previewing overheats my CPU and
makes it nearly impossible to adjust the viewport as much of the interface
becomes so sluggish it is practically frozen. I had been taking advantage
of the $preview variable to set the other resolution adjusting vars such as
$fs to minimize resolution for previewing and have a high resolution for
rendering. But now I have been manually changing vars to only have high
res for renders before exporting. What I'd really love to have is a way to
do single keystrokes for a preview render, a full render and an export
render (as an obj and not an stl). I expect I will have to build my own
from source to achieve this but thought I would ask in case I am missing
something and other solutions are possible.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
On Sun, 2024-08-04 at 21:43 -0500, Todd Allen via Discuss wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running under Qubes
OS 4.2 with pvh virtualization and seamless integration enabled.
I get the same behavior with both a week old unstable release and the
last stable release 2024.01.06 of OpenSCAD, both are .appimage files.
AMD Ryzen 5 5600G with Radeon Graphics
Ubuntu MATE 22.04 LTS
OpenSCAD version 2024.03.02
Your program:
Compile and preview finished.
Total rendering time: 0:00:00.235
Compiling design (CSG Tree generation)...
Rendering Polygon Mesh using Manifold...
Geometries in cache: 66
Geometry cache size in bytes: 3443032
CGAL Polyhedrons in cache: 23
CGAL cache size in bytes: 0
Total rendering time: 0:00:00.777
I think the issue may be a graphics driver as the problem got much
worse after switching from a 1080p monitor to a 2560x1440 monitor.
I think you're right.
I have a Windows VM which I rarely use to run Fusion360 which is a
much more graphically demanding app and it runs well. I should try
OpenSCAD there to help isolate if the problem is with the linux
graphics driver. Curiously though when I render the video is smooth
and fairly responsive under linux and it is only when looking at a
preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26,
chamfer=0.5, inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer :
inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer :
outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d0.5+$wall_t0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d0.5) yrot(90)
down($pipe_d0.5)
join_prism(joined_shape,base="cylinder",base_d=$pipe_d, length=h0.6,
fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct,
circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h1.04) {
difference() {
zrot_copies(n=3) xrot(ang) down(h0.5+$corner_cut) leg(h,
$pipe_d, $outer_chamfer, $inner_chamfer);
zrot_copies(n=3) xrot(ang) down(h*0.5+$corner_cut)
leg_core(h, $rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if
parameters change this will need tweaking, probably make these values
enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45)
cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss
discuss@lists.openscad.org wrote:
What is the openscad version, as well as OS and hardware spec?
Nightly version is using a different preview pipeline. And this
somehow makes me wonder if you are using software rendering instead
of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss,
discuss@lists.openscad.org wrote:
I use the render key but the other two are no longer of much
use. If I can I'll rewrite the <export STL> key to do a render
with $export=true which I can test in script to go higher res
and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss
discuss@lists.openscad.org wrote:
F5, F6 and F7 do preview, render and export but I think F7 is
STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via Discuss
discuss@lists.openscad.org wrote:
When working with complex high facet models I've turned off
automatic preview and just use rendering because previewing
overheats my CPU and makes it nearly impossible to adjust
the viewport as much of the interface becomes
so sluggish it is practically frozen. I had been taking
advantage of the $preview variable to set the other
resolution adjusting vars such as $fs to minimize
resolution for previewing and have a high resolution for
rendering. But now I have been manually changing vars to
only have high res for renders before exporting. What I'd
really love to have is a way to do single keystrokes for a
preview render, a full render and an export render (as an
obj and not an stl). I expect I will have to build my own
from source to achieve this but thought I would ask in case
I am missing something and other solutions are possible.
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
_______________________________________________
OpenSCAD mailing list
To unsubscribe send an email to
discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
Thanks! I never thought to try that. When I read about render() in the
documentation it wasn't obvious to me when it would be useful and I never
used it. But after a couple test uses I expect I will use it a lot.
On Mon, Aug 5, 2024 at 2:15 AM nop head via Discuss <
discuss@lists.openscad.org> wrote:
I put a render() call around bits that cause the preview frame rate to
drop. It is much less of a problem with the new rendering system.
On Mon, 5 Aug 2024 at 07:19, Michael Marx (spintel) via Discuss <
discuss@lists.openscad.org> wrote:
Also make sure you have large cache sizes in preferences, re-render will
be quicker.
From: Chun Kit LAM via Discuss [mailto:discuss@lists.openscad.org]
Sent: Mon, 5 Aug 2024 14:48
To: Todd Allen
Cc: OpenSCAD general discussion Mailing-list; Chun Kit LAM
Subject: [OpenSCAD] Re: CPU load remains high after previewing but not
after render
If you look at the paper of OpenCSG, which is the backend for preview,
you will see that the Goldfeather algorithm has a complexity of O(n^2).
Even with depth complexity optimization, it is still O(kn). And you need to
do that for each pixel (AFAIK). This is why it is slow.
I think you will have better luck running the actual render and do not
use preview. This may be slower at first, but it does not require running
an expensive shader every frame.
On 5/8/2024 11:39, Todd Allen wrote:
Yes, display performance becomes bearable when preview faceting is kept
very low. I figured I was running into a bug in OpenSCAD or my graphics
driver as the display is quite responsive at vastly higher numbers of faces
generated by a render.
On Sun, Aug 4, 2024 at 10:20 PM Todd Allen speedebikes@gmail.com wrote:
Yes, just checked it is doing software graphics. It is possible to use
GPU graphics but it looks like I have a few hoops to jump through. Why is
displaying a preview so much more demanding than displaying a render?
Anyway for now I'll see if I can get by just making previews very coarse
grained.
On Sun, Aug 4, 2024 at 10:06 PM Chun Kit LAM john.lck40@gmail.com
wrote:
Do you have GPU passthrough for your VM? When running under
virtualization, it may run under software rendering. And I doubt if
Fusion360 is much more graphically demanding, openscad preview render is
kind of like unoptimized ray-tracing (from my understanding) which is quite
heavy even for good GPUs, and it will definitely stress your CPU when run
under software rendering.
On 5/8/2024 10:43, Todd Allen wrote:
Ryzen 9 6900HX w/ AMD Radeon 680M graphics, 32 GB ram
OS is a fully updated Arch linux virtual machine running under Qubes OS
4.2 with pvh virtualization and seamless integration enabled.
I get the same behavior with both a week old unstable release and the
last stable release 2024.01.06 of OpenSCAD, both are .appimage files.
I think the issue may be a graphics driver as the problem got much worse
after switching from a 1080p monitor to a 2560x1440 monitor.
I have a Windows VM which I rarely use to run Fusion360 which is a much
more graphically demanding app and it runs well. I should try OpenSCAD
there to help isolate if the problem is with the linux graphics driver.
Curiously though when I render the video is smooth and fairly responsive
under linux and it is only when looking at a preview the system chokes.
Here's a sample script which causes this:
include <BOSL2/std.scad>
include <BOSL2/rounding.scad>
$fn= $preview ? 32 : 256;
eps = 0.002;
module setenv_RodFrame(rod_d=10, wall_t=2.5, corner_size=26, chamfer=0.5,
inner_chamfer, outer_chamfer) {
$rod_d = rod_d;
$wall_t = wall_t;
$corner_size = corner_size;
$inner_chamfer = is_undef(inner_chamfer) ? chamfer : inner_chamfer;
$outer_chamfer = is_undef(outer_chamfer) ? chamfer : outer_chamfer;
$pipe_d = $rod_d+2*$wall_t;
$flange_t = $wall_t;
$flange_fillet = 1.25*$wall_t;
$corner_cut = $rod_d0.5+$wall_t0.8;
children();
}
module leg(h, d, cb, ct) {
joined_shape = rect([h-$flange_fillet0.6,$flange_t]);
union() {
zcyl(h=h, d=d, chamfer1=cb, chamfer2=ct, circum=true);
for(a = [-45,-135]) zrot(a) right($pipe_d0.5) yrot(90)
down($pipe_d0.5) join_prism(joined_shape,base="cylinder",base_d=$pipe_d,
length=h0.6, fillet=$flange_fillet, n=24);
}
}
module leg_core(h, d, cb, ct) {
down(eps) zcyl(h=h+10*eps, d=d, chamfer1=cb, chamfer2=ct,
circum=true);
}
module corner() {
ang = acos(1/sqrt(3));
h = $corner_size-$corner_cut;
top_half(s=200) up(h1.04) {
difference() {
zrot_copies(n=3) xrot(ang) down(h0.5+$corner_cut) leg(h,
$pipe_d, $outer_chamfer, $inner_chamfer);
zrot_copies(n=3) xrot(ang) down(h*0.5+$corner_cut)
leg_core(h, $rod_d, -$outer_chamfer,0);
// trim bottom outer edge of flange
// figuring out a formula too hard for now so if parameters
change this will need tweaking, probably make these values enviroment vars
zrot_copies(n=3) fwd(14) down(h) xrot(45)
cuboid([26,3,7],anchor=BACK);
}
}
}
// EXAMPLE invocation
setenv_RodFrame() corner();
On Sun, Aug 4, 2024 at 7:07 PM Chun Kit LAM via Discuss <
discuss@lists.openscad.org> wrote:
What is the openscad version, as well as OS and hardware spec? Nightly
version is using a different preview pipeline. And this somehow makes me
wonder if you are using software rendering instead of GPU rendering.
On 5/8/2024 07:05, Sanjeev Prabhakar via Discuss wrote:
Can you share a picture of your model?
On Mon, 5 Aug, 2024, 2:20 am Todd Allen via Discuss, <
discuss@lists.openscad.org> wrote:
I use the render key but the other two are no longer of much use. If I
can I'll rewrite the <export STL> key to do a render with $export=true
which I can test in script to go higher res and then export an OBJ.
On Sun, Aug 4, 2024 at 11:57 AM nop head via Discuss <
discuss@lists.openscad.org> wrote:
F5, F6 and F7 do preview, render and export but I think F7 is STL only.
On Sun, 4 Aug 2024 at 17:44, Todd Allen via Discuss <
discuss@lists.openscad.org> wrote:
When working with complex high facet models I've turned off automatic
preview and just use rendering because previewing overheats my CPU and
makes it nearly impossible to adjust the viewport as much of the interface
becomes so sluggish it is practically frozen. I had been taking advantage
of the $preview variable to set the other resolution adjusting vars such as
$fs to minimize resolution for previewing and have a high resolution for
rendering. But now I have been manually changing vars to only have high
res for renders before exporting. What I'd really love to have is a way to
do single keystrokes for a preview render, a full render and an export
render (as an obj and not an stl). I expect I will have to build my own
from source to achieve this but thought I would ask in case I am missing
something and other solutions are possible.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
Thanks everyone for all the tips and suggestions! What previously I feared
was a bug in my video driver or OpenSCAD needing possibly drastic work
arounds such as avoiding preview is merely the normal behavior that is
trivial to cope with now that I understand it. The issue was only that
refreshing the viewport without using GPU graphics is very CPU intensive
and the load increases as a product of increasing facets in the model AND
the number of pixels in the viewport. I was frustrated by the OpenSCAD
interface becoming unresponsive when previews were too demanding but in my
case my OS's windows manager hotkeys are unaffected and I can readily
unmaximize or shrink the window with the viewport to resolve inadvertent
overloads. It would be nice in OpenSCAD if there were hotkeys that could
adjust the sizes of editor and viewport when docked or if those windows
were more fully independent as the console window is when undocked but even
without either of those I can keep previewing in my standard workflow.
There is an even more trivial escape for inadvertent preview overloads.
Pressing <F10> (View Surfaces) changes the view to a massively more CPU
friendly display refresh.
On Tue, Aug 6, 2024 at 9:04 AM Todd Allen speedebikes@gmail.com wrote:
Thanks everyone for all the tips and suggestions! What previously I
feared was a bug in my video driver or OpenSCAD needing possibly drastic
work arounds such as avoiding preview is merely the normal behavior that is
trivial to cope with now that I understand it. The issue was only that
refreshing the viewport without using GPU graphics is very CPU intensive
and the load increases as a product of increasing facets in the model AND
the number of pixels in the viewport. I was frustrated by the OpenSCAD
interface becoming unresponsive when previews were too demanding but in my
case my OS's windows manager hotkeys are unaffected and I can readily
unmaximize or shrink the window with the viewport to resolve inadvertent
overloads. It would be nice in OpenSCAD if there were hotkeys that could
adjust the sizes of editor and viewport when docked or if those windows
were more fully independent as the console window is when undocked but even
without either of those I can keep previewing in my standard workflow.