CM
charles meyer
Mon, Feb 17, 2025 3:51 PM
Hi All,
I hope this is a permitted post re: hardware?
I'm helping a small public library develop their Maker Space which includes
3D Printers and Cricut machines.
They use mostly Tinkercad, some meshmixer and occasionally OpenScad for the
3D printer.
The thought is toy star teaching praorrs how to video edit in DR and use
PhotoShop or Affinity Publihser for graphic editing.
I found this writing....
Professional or Workstation Graphics Cards are special types of graphics
cards that are used in workstations and high-end servers. These graphics
cards pack huge power and are optimized for CAD, CAE, and heavy-duty
graphics applications. They are used mainly in scientific applications,
studios and work involving the use of heavy Graphics applications for 3D
modeling, 3D rendering, etc. or to compute a massive number of graphics
calculations.
Workstation Graphics Cards look the same as the normal graphics cards from
the outside but their working is very much different. These graphics cards
may have the same internal GPU architecture
https://graphicscardhub.com/nvidia-pascal-vs-amd-polaris-gpu-architecture/
but they are optimized for professional graphics applications and some of
them may contain extra hardware components or features for processing
additional complex graphics work. They are also equipped with larger memory
as compared to consumer graphics cards. You can find a maximum of 32 GB of
memory in a workstation graphics card but the highest memory present
in the topmost
gaming graphics card https://graphicscardhub.com/ultimate-graphics-card/
is 12GB. Also, the memory present in most workstation graphics cards is ECC
memory which helps in producing accurate results. Workstation Graphics
Cards are also known as Enterprise graphics cards.
Have you found in your 3D printing CAD it was needed or helpful to
have an Enterprise
graphics cards or Professional or Workstation Graphics Card?
I don't know what this author means with the reference to heavy Graphics
applications for 3D modeling, 3D rendering?
Thank you so much,
Charles.
Hi All,
I hope this is a permitted post re: hardware?
I'm helping a small public library develop their Maker Space which includes
3D Printers and Cricut machines.
They use mostly Tinkercad, some meshmixer and occasionally OpenScad for the
3D printer.
The thought is toy star teaching praorrs how to video edit in DR and use
PhotoShop or Affinity Publihser for graphic editing.
I found this writing....
Professional or Workstation Graphics Cards are special types of graphics
cards that are used in workstations and high-end servers. These graphics
cards pack huge power and are optimized for CAD, CAE, and heavy-duty
graphics applications. They are used mainly in scientific applications,
studios and work involving the use of heavy Graphics applications for 3D
modeling, 3D rendering, etc. or to compute a massive number of graphics
calculations.
Workstation Graphics Cards look the same as the normal graphics cards from
the outside but their working is very much different. These graphics cards
may have the same internal GPU architecture
<https://graphicscardhub.com/nvidia-pascal-vs-amd-polaris-gpu-architecture/>
but they are optimized for professional graphics applications and some of
them may contain extra hardware components or features for processing
additional complex graphics work. They are also equipped with larger memory
as compared to consumer graphics cards. You can find a maximum of 32 GB of
memory in a workstation graphics card but the highest memory present
in the topmost
gaming graphics card <https://graphicscardhub.com/ultimate-graphics-card/>
is 12GB. Also, the memory present in most workstation graphics cards is ECC
memory which helps in producing accurate results. Workstation Graphics
Cards are also known as Enterprise graphics cards.
Have you found in your 3D printing CAD it was needed or helpful to
have an Enterprise
graphics cards or Professional or Workstation Graphics Card?
I don't know what this author means with the reference to heavy Graphics
applications for 3D modeling, 3D rendering?
Thank you so much,
Charles.
JB
Jordan Brown
Mon, Feb 17, 2025 4:53 PM
I don't think anybody minds an occasional tangentially-related topic, as
long as it doesn't take over. And if you spun it as "graphics cards for
OpenSCAD", it would be totally on-topic.
My experience is that it doesn't matter much. I buy cheap computers,
and OpenSCAD has performed fine for me, at least in the graphics areas.
I won't say that I've never had laggy video updates, but it's pretty
rare. I upgrade from the base models for three parameters: RAM (the
more the merrier), persistent storage (128GB doesn't cut it), and (for
laptops) display size.
The parts of OpenSCAD that seem performance-sensitive are rendering and,
especially with BOSL2, execution. Neither is aided by a better GPU.
(This should, by the way, not be read as a criticism of BOSL2. It's
doing incredible stuff and it's no surprise that it takes time to do
it.) If those are your performance complaints, get a better CPU. If
your performance complaint is that rotating the model is laggy, a hotter
GPU might help.
I haven't priced high-end graphics cards, but I believe they are more
expensive than midrange systems... and for the environment you describe,
if you have a choice between one computer with a high-end graphics card
and two more modest computers, I know which answer I'd pick.
I don't think anybody minds an occasional tangentially-related topic, as
long as it doesn't take over. And if you spun it as "graphics cards for
OpenSCAD", it would be totally on-topic.
My experience is that it doesn't matter much. I buy cheap computers,
and OpenSCAD has performed fine for me, at least in the graphics areas.
I won't say that I've never had laggy video updates, but it's pretty
rare. I upgrade from the base models for three parameters: RAM (the
more the merrier), persistent storage (128GB doesn't cut it), and (for
laptops) display size.
The parts of OpenSCAD that seem performance-sensitive are rendering and,
especially with BOSL2, execution. Neither is aided by a better GPU.
(This should, by the way, not be read as a criticism of BOSL2. It's
doing incredible stuff and it's no surprise that it takes time to do
it.) If those are your performance complaints, get a better CPU. If
your performance complaint is that rotating the model is laggy, a hotter
GPU might help.
I haven't priced high-end graphics cards, but I believe they are more
expensive than midrange systems... and for the environment you describe,
if you have a choice between one computer with a high-end graphics card
and two more modest computers, I know which answer I'd pick.
CK
Chun Kit LAM
Mon, Feb 17, 2025 5:18 PM
From my understanding, high-end graphics cards don't help with mesh
processing or NURBS-based designs. It is useful if you want to render
designs to 2D, especially when you need photorealistic rendering (e.g.
blender). Simulation may also require workstation GPUs for acceleration
(see below). Rotating a rendered model is rarely a bottleneck for modern
graphics cards (or even just integrated GPUs) because rasterization is
typically fast enough.
I think in the foreseeable future, open source CAD tools will not
optimize for GPU processing because consumer grade GPUs don't provide
double precision floating point operations, GPUs companies force you to
buy workstation GPUs for that. Most open source developers probably
don't have access to workstation grade GPUs (at least for the manifold
project, we don't) and most users don't have them, so it is unlikely we
will try to optimize for that.
For better performance, buying better CPUs (with more threads) and more
RAM is probably the way to go. 8 threads can probably saturate openscad
and manifold, at least for now. For RAM, I will recommend at least 32GB
nowadays.
If the project is not in a hurry, I will recommend you to wait a bit.
AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
probably similar to mid-range GPUs nowadays (that costs >= $300), but
should be much cheaper. Another issue with mid-range GPUs is that they
usually have very limited VRAM, so some tasks can't even run on the GPU
due to VRAM limits. For iGPU, this is not as much as an issue because
you can allocate more RAM to the iGPU. I think this should come out in a
few months or so.
On 2/18/25 00:53, Jordan Brown via Discuss wrote:
I don't think anybody minds an occasional tangentially-related topic,
as long as it doesn't take over. And if you spun it as "graphics
cards for OpenSCAD", it would be totally on-topic.
My experience is that it doesn't matter much. I buy cheap computers,
and OpenSCAD has performed fine for me, at least in the graphics
areas. I won't say that I've never had laggy video updates, but it's
pretty rare. I upgrade from the base models for three parameters:
RAM (the more the merrier), persistent storage (128GB doesn't cut it),
and (for laptops) display size.
The parts of OpenSCAD that seem performance-sensitive are rendering
and, especially with BOSL2, execution. Neither is aided by a better
GPU. (This should, by the way, not be read as a criticism of BOSL2.
It's doing incredible stuff and it's no surprise that it takes time to
do it.) If those are your performance complaints, get a better CPU.
If your performance complaint is that rotating the model is laggy, a
hotter GPU might help.
I haven't priced high-end graphics cards, but I believe they are more
expensive than midrange systems... and for the environment you
describe, if you have a choice between one computer with a high-end
graphics card and two more modest computers, I know which answer I'd pick.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
From my understanding, high-end graphics cards don't help with mesh
processing or NURBS-based designs. It is useful if you want to render
designs to 2D, especially when you need photorealistic rendering (e.g.
blender). Simulation may also require workstation GPUs for acceleration
(see below). Rotating a rendered model is rarely a bottleneck for modern
graphics cards (or even just integrated GPUs) because rasterization is
typically fast enough.
I think in the foreseeable future, open source CAD tools will not
optimize for GPU processing because consumer grade GPUs don't provide
double precision floating point operations, GPUs companies force you to
buy workstation GPUs for that. Most open source developers probably
don't have access to workstation grade GPUs (at least for the manifold
project, we don't) and most users don't have them, so it is unlikely we
will try to optimize for that.
For better performance, buying better CPUs (with more threads) and more
RAM is probably the way to go. 8 threads can probably saturate openscad
and manifold, at least for now. For RAM, I will recommend at least 32GB
nowadays.
If the project is not in a hurry, I will recommend you to wait a bit.
AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
probably similar to mid-range GPUs nowadays (that costs >= $300), but
should be much cheaper. Another issue with mid-range GPUs is that they
usually have very limited VRAM, so some tasks can't even run on the GPU
due to VRAM limits. For iGPU, this is not as much as an issue because
you can allocate more RAM to the iGPU. I think this should come out in a
few months or so.
On 2/18/25 00:53, Jordan Brown via Discuss wrote:
> I don't think anybody minds an occasional tangentially-related topic,
> as long as it doesn't take over. And if you spun it as "graphics
> cards for OpenSCAD", it would be totally on-topic.
>
> My experience is that it doesn't matter much. I buy cheap computers,
> and OpenSCAD has performed fine for me, at least in the graphics
> areas. I won't say that I've never had laggy video updates, but it's
> pretty rare. I upgrade from the base models for three parameters:
> RAM (the more the merrier), persistent storage (128GB doesn't cut it),
> and (for laptops) display size.
>
> The parts of OpenSCAD that seem performance-sensitive are rendering
> and, especially with BOSL2, execution. Neither is aided by a better
> GPU. (This should, by the way, not be read as a criticism of BOSL2.
> It's doing incredible stuff and it's no surprise that it takes time to
> do it.) If those are your performance complaints, get a better CPU.
> If your performance complaint is that rotating the model is laggy, a
> hotter GPU might help.
>
> I haven't priced high-end graphics cards, but I believe they are more
> expensive than midrange systems... and for the environment you
> describe, if you have a choice between one computer with a high-end
> graphics card and two more modest computers, I know which answer I'd pick.
>
>
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email to discuss-leave@lists.openscad.org
GB
Glenn Butcher
Mon, Feb 17, 2025 8:30 PM
For a time, manifold had a build option that incorporated NVIDIA's CUDA
GPU library. I tried it, but the Ubuntu CUDA packages wanted to change
out my kernel and that made a holy mess of my machine.
On 2/17/2025 10:18 AM, Chun Kit LAM via Discuss wrote:
From my understanding, high-end graphics cards don't help with mesh
processing or NURBS-based designs. It is useful if you want to render
designs to 2D, especially when you need photorealistic rendering (e.g.
blender). Simulation may also require workstation GPUs for
acceleration (see below). Rotating a rendered model is rarely a
bottleneck for modern graphics cards (or even just integrated GPUs)
because rasterization is typically fast enough.
I think in the foreseeable future, open source CAD tools will not
optimize for GPU processing because consumer grade GPUs don't provide
double precision floating point operations, GPUs companies force you
to buy workstation GPUs for that. Most open source developers probably
don't have access to workstation grade GPUs (at least for the manifold
project, we don't) and most users don't have them, so it is unlikely
we will try to optimize for that.
For better performance, buying better CPUs (with more threads) and
more RAM is probably the way to go. 8 threads can probably saturate
openscad and manifold, at least for now. For RAM, I will recommend at
least 32GB nowadays.
If the project is not in a hurry, I will recommend you to wait a bit.
AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
probably similar to mid-range GPUs nowadays (that costs >= $300), but
should be much cheaper. Another issue with mid-range GPUs is that they
usually have very limited VRAM, so some tasks can't even run on the
GPU due to VRAM limits. For iGPU, this is not as much as an issue
because you can allocate more RAM to the iGPU. I think this should
come out in a few months or so.
On 2/18/25 00:53, Jordan Brown via Discuss wrote:
I don't think anybody minds an occasional tangentially-related topic,
as long as it doesn't take over. And if you spun it as "graphics
cards for OpenSCAD", it would be totally on-topic.
My experience is that it doesn't matter much. I buy cheap computers,
and OpenSCAD has performed fine for me, at least in the graphics
areas. I won't say that I've never had laggy video updates, but it's
pretty rare. I upgrade from the base models for three parameters:
RAM (the more the merrier), persistent storage (128GB doesn't cut
it), and (for laptops) display size.
The parts of OpenSCAD that seem performance-sensitive are rendering
and, especially with BOSL2, execution. Neither is aided by a better
GPU. (This should, by the way, not be read as a criticism of BOSL2.
It's doing incredible stuff and it's no surprise that it takes time
to do it.) If those are your performance complaints, get a better
CPU. If your performance complaint is that rotating the model is
laggy, a hotter GPU might help.
I haven't priced high-end graphics cards, but I believe they are more
expensive than midrange systems... and for the environment you
describe, if you have a choice between one computer with a high-end
graphics card and two more modest computers, I know which answer I'd
pick.
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
For a time, manifold had a build option that incorporated NVIDIA's CUDA
GPU library. I tried it, but the Ubuntu CUDA packages wanted to change
out my kernel and that made a holy mess of my machine.
On 2/17/2025 10:18 AM, Chun Kit LAM via Discuss wrote:
> From my understanding, high-end graphics cards don't help with mesh
> processing or NURBS-based designs. It is useful if you want to render
> designs to 2D, especially when you need photorealistic rendering (e.g.
> blender). Simulation may also require workstation GPUs for
> acceleration (see below). Rotating a rendered model is rarely a
> bottleneck for modern graphics cards (or even just integrated GPUs)
> because rasterization is typically fast enough.
>
> I think in the foreseeable future, open source CAD tools will not
> optimize for GPU processing because consumer grade GPUs don't provide
> double precision floating point operations, GPUs companies force you
> to buy workstation GPUs for that. Most open source developers probably
> don't have access to workstation grade GPUs (at least for the manifold
> project, we don't) and most users don't have them, so it is unlikely
> we will try to optimize for that.
>
> For better performance, buying better CPUs (with more threads) and
> more RAM is probably the way to go. 8 threads can probably saturate
> openscad and manifold, at least for now. For RAM, I will recommend at
> least 32GB nowadays.
>
> If the project is not in a hurry, I will recommend you to wait a bit.
> AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
> probably similar to mid-range GPUs nowadays (that costs >= $300), but
> should be much cheaper. Another issue with mid-range GPUs is that they
> usually have very limited VRAM, so some tasks can't even run on the
> GPU due to VRAM limits. For iGPU, this is not as much as an issue
> because you can allocate more RAM to the iGPU. I think this should
> come out in a few months or so.
>
> On 2/18/25 00:53, Jordan Brown via Discuss wrote:
>> I don't think anybody minds an occasional tangentially-related topic,
>> as long as it doesn't take over. And if you spun it as "graphics
>> cards for OpenSCAD", it would be totally on-topic.
>>
>> My experience is that it doesn't matter much. I buy cheap computers,
>> and OpenSCAD has performed fine for me, at least in the graphics
>> areas. I won't say that I've never had laggy video updates, but it's
>> pretty rare. I upgrade from the base models for three parameters:
>> RAM (the more the merrier), persistent storage (128GB doesn't cut
>> it), and (for laptops) display size.
>>
>> The parts of OpenSCAD that seem performance-sensitive are rendering
>> and, especially with BOSL2, execution. Neither is aided by a better
>> GPU. (This should, by the way, not be read as a criticism of BOSL2.
>> It's doing incredible stuff and it's no surprise that it takes time
>> to do it.) If those are your performance complaints, get a better
>> CPU. If your performance complaint is that rotating the model is
>> laggy, a hotter GPU might help.
>>
>> I haven't priced high-end graphics cards, but I believe they are more
>> expensive than midrange systems... and for the environment you
>> describe, if you have a choice between one computer with a high-end
>> graphics card and two more modest computers, I know which answer I'd
>> pick.
>>
>>
>> _______________________________________________
>> 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
JD
John David
Mon, Feb 17, 2025 11:25 PM
I have no end of heartache with NVidia's CUDA. I have the same thing on
the server I built for my graduate research. Ended up never being able to
use the Tesla K40. I had the bad luck of purchasing the K40 a couple of
weeks before the K80 came out. They dropped the K40 support very quickly,
and it messed with my ability to maintain the machine with it. Sigh...
Anyway want a K4)?...
On Mon, Feb 17, 2025 at 3:30 PM Glenn Butcher via Discuss <
discuss@lists.openscad.org> wrote:
For a time, manifold had a build option that incorporated NVIDIA's CUDA
GPU library. I tried it, but the Ubuntu CUDA packages wanted to change
out my kernel and that made a holy mess of my machine.
On 2/17/2025 10:18 AM, Chun Kit LAM via Discuss wrote:
From my understanding, high-end graphics cards don't help with mesh
processing or NURBS-based designs. It is useful if you want to render
designs to 2D, especially when you need photorealistic rendering (e.g.
blender). Simulation may also require workstation GPUs for
acceleration (see below). Rotating a rendered model is rarely a
bottleneck for modern graphics cards (or even just integrated GPUs)
because rasterization is typically fast enough.
I think in the foreseeable future, open source CAD tools will not
optimize for GPU processing because consumer grade GPUs don't provide
double precision floating point operations, GPUs companies force you
to buy workstation GPUs for that. Most open source developers probably
don't have access to workstation grade GPUs (at least for the manifold
project, we don't) and most users don't have them, so it is unlikely
we will try to optimize for that.
For better performance, buying better CPUs (with more threads) and
more RAM is probably the way to go. 8 threads can probably saturate
openscad and manifold, at least for now. For RAM, I will recommend at
least 32GB nowadays.
If the project is not in a hurry, I will recommend you to wait a bit.
AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
probably similar to mid-range GPUs nowadays (that costs >= $300), but
should be much cheaper. Another issue with mid-range GPUs is that they
usually have very limited VRAM, so some tasks can't even run on the
GPU due to VRAM limits. For iGPU, this is not as much as an issue
because you can allocate more RAM to the iGPU. I think this should
come out in a few months or so.
On 2/18/25 00:53, Jordan Brown via Discuss wrote:
I don't think anybody minds an occasional tangentially-related topic,
as long as it doesn't take over. And if you spun it as "graphics
cards for OpenSCAD", it would be totally on-topic.
My experience is that it doesn't matter much. I buy cheap computers,
and OpenSCAD has performed fine for me, at least in the graphics
areas. I won't say that I've never had laggy video updates, but it's
pretty rare. I upgrade from the base models for three parameters:
RAM (the more the merrier), persistent storage (128GB doesn't cut
it), and (for laptops) display size.
The parts of OpenSCAD that seem performance-sensitive are rendering
and, especially with BOSL2, execution. Neither is aided by a better
GPU. (This should, by the way, not be read as a criticism of BOSL2.
It's doing incredible stuff and it's no surprise that it takes time
to do it.) If those are your performance complaints, get a better
CPU. If your performance complaint is that rotating the model is
laggy, a hotter GPU might help.
I haven't priced high-end graphics cards, but I believe they are more
expensive than midrange systems... and for the environment you
describe, if you have a choice between one computer with a high-end
graphics card and two more modest computers, I know which answer I'd
pick.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
I have no end of heartache with NVidia's CUDA. I have the same thing on
the server I built for my graduate research. Ended up never being able to
use the Tesla K40. I had the bad luck of purchasing the K40 a couple of
weeks before the K80 came out. They dropped the K40 support very quickly,
and it messed with my ability to maintain the machine with it. Sigh...
Anyway want a K4)?...
On Mon, Feb 17, 2025 at 3:30 PM Glenn Butcher via Discuss <
discuss@lists.openscad.org> wrote:
> For a time, manifold had a build option that incorporated NVIDIA's CUDA
> GPU library. I tried it, but the Ubuntu CUDA packages wanted to change
> out my kernel and that made a holy mess of my machine.
>
> On 2/17/2025 10:18 AM, Chun Kit LAM via Discuss wrote:
> > From my understanding, high-end graphics cards don't help with mesh
> > processing or NURBS-based designs. It is useful if you want to render
> > designs to 2D, especially when you need photorealistic rendering (e.g.
> > blender). Simulation may also require workstation GPUs for
> > acceleration (see below). Rotating a rendered model is rarely a
> > bottleneck for modern graphics cards (or even just integrated GPUs)
> > because rasterization is typically fast enough.
> >
> > I think in the foreseeable future, open source CAD tools will not
> > optimize for GPU processing because consumer grade GPUs don't provide
> > double precision floating point operations, GPUs companies force you
> > to buy workstation GPUs for that. Most open source developers probably
> > don't have access to workstation grade GPUs (at least for the manifold
> > project, we don't) and most users don't have them, so it is unlikely
> > we will try to optimize for that.
> >
> > For better performance, buying better CPUs (with more threads) and
> > more RAM is probably the way to go. 8 threads can probably saturate
> > openscad and manifold, at least for now. For RAM, I will recommend at
> > least 32GB nowadays.
> >
> > If the project is not in a hurry, I will recommend you to wait a bit.
> > AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
> > probably similar to mid-range GPUs nowadays (that costs >= $300), but
> > should be much cheaper. Another issue with mid-range GPUs is that they
> > usually have very limited VRAM, so some tasks can't even run on the
> > GPU due to VRAM limits. For iGPU, this is not as much as an issue
> > because you can allocate more RAM to the iGPU. I think this should
> > come out in a few months or so.
> >
> > On 2/18/25 00:53, Jordan Brown via Discuss wrote:
> >> I don't think anybody minds an occasional tangentially-related topic,
> >> as long as it doesn't take over. And if you spun it as "graphics
> >> cards for OpenSCAD", it would be totally on-topic.
> >>
> >> My experience is that it doesn't matter much. I buy cheap computers,
> >> and OpenSCAD has performed fine for me, at least in the graphics
> >> areas. I won't say that I've never had laggy video updates, but it's
> >> pretty rare. I upgrade from the base models for three parameters:
> >> RAM (the more the merrier), persistent storage (128GB doesn't cut
> >> it), and (for laptops) display size.
> >>
> >> The parts of OpenSCAD that seem performance-sensitive are rendering
> >> and, especially with BOSL2, execution. Neither is aided by a better
> >> GPU. (This should, by the way, not be read as a criticism of BOSL2.
> >> It's doing incredible stuff and it's no surprise that it takes time
> >> to do it.) If those are your performance complaints, get a better
> >> CPU. If your performance complaint is that rotating the model is
> >> laggy, a hotter GPU might help.
> >>
> >> I haven't priced high-end graphics cards, but I believe they are more
> >> expensive than midrange systems... and for the environment you
> >> describe, if you have a choice between one computer with a high-end
> >> graphics card and two more modest computers, I know which answer I'd
> >> pick.
> >>
> >>
> >> _______________________________________________
> >> 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
>
P
pca006132
Tue, Feb 18, 2025 1:48 AM
Yes, but we later found that the benefit of using the GPU is not that big
due to frequent transfer between the cpu and gpu making it pcie throughput
bottlenecked. Also, most users of the library don't want to deal with cuda.
We ended up just removing it all together.
On Tue, Feb 18, 2025, 4:30 AM Glenn Butcher via Discuss <
discuss@lists.openscad.org> wrote:
For a time, manifold had a build option that incorporated NVIDIA's CUDA
GPU library. I tried it, but the Ubuntu CUDA packages wanted to change
out my kernel and that made a holy mess of my machine.
On 2/17/2025 10:18 AM, Chun Kit LAM via Discuss wrote:
From my understanding, high-end graphics cards don't help with mesh
processing or NURBS-based designs. It is useful if you want to render
designs to 2D, especially when you need photorealistic rendering (e.g.
blender). Simulation may also require workstation GPUs for
acceleration (see below). Rotating a rendered model is rarely a
bottleneck for modern graphics cards (or even just integrated GPUs)
because rasterization is typically fast enough.
I think in the foreseeable future, open source CAD tools will not
optimize for GPU processing because consumer grade GPUs don't provide
double precision floating point operations, GPUs companies force you
to buy workstation GPUs for that. Most open source developers probably
don't have access to workstation grade GPUs (at least for the manifold
project, we don't) and most users don't have them, so it is unlikely
we will try to optimize for that.
For better performance, buying better CPUs (with more threads) and
more RAM is probably the way to go. 8 threads can probably saturate
openscad and manifold, at least for now. For RAM, I will recommend at
least 32GB nowadays.
If the project is not in a hurry, I will recommend you to wait a bit.
AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
probably similar to mid-range GPUs nowadays (that costs >= $300), but
should be much cheaper. Another issue with mid-range GPUs is that they
usually have very limited VRAM, so some tasks can't even run on the
GPU due to VRAM limits. For iGPU, this is not as much as an issue
because you can allocate more RAM to the iGPU. I think this should
come out in a few months or so.
On 2/18/25 00:53, Jordan Brown via Discuss wrote:
I don't think anybody minds an occasional tangentially-related topic,
as long as it doesn't take over. And if you spun it as "graphics
cards for OpenSCAD", it would be totally on-topic.
My experience is that it doesn't matter much. I buy cheap computers,
and OpenSCAD has performed fine for me, at least in the graphics
areas. I won't say that I've never had laggy video updates, but it's
pretty rare. I upgrade from the base models for three parameters:
RAM (the more the merrier), persistent storage (128GB doesn't cut
it), and (for laptops) display size.
The parts of OpenSCAD that seem performance-sensitive are rendering
and, especially with BOSL2, execution. Neither is aided by a better
GPU. (This should, by the way, not be read as a criticism of BOSL2.
It's doing incredible stuff and it's no surprise that it takes time
to do it.) If those are your performance complaints, get a better
CPU. If your performance complaint is that rotating the model is
laggy, a hotter GPU might help.
I haven't priced high-end graphics cards, but I believe they are more
expensive than midrange systems... and for the environment you
describe, if you have a choice between one computer with a high-end
graphics card and two more modest computers, I know which answer I'd
pick.
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
Yes, but we later found that the benefit of using the GPU is not that big
due to frequent transfer between the cpu and gpu making it pcie throughput
bottlenecked. Also, most users of the library don't want to deal with cuda.
We ended up just removing it all together.
On Tue, Feb 18, 2025, 4:30 AM Glenn Butcher via Discuss <
discuss@lists.openscad.org> wrote:
> For a time, manifold had a build option that incorporated NVIDIA's CUDA
> GPU library. I tried it, but the Ubuntu CUDA packages wanted to change
> out my kernel and that made a holy mess of my machine.
>
> On 2/17/2025 10:18 AM, Chun Kit LAM via Discuss wrote:
> > From my understanding, high-end graphics cards don't help with mesh
> > processing or NURBS-based designs. It is useful if you want to render
> > designs to 2D, especially when you need photorealistic rendering (e.g.
> > blender). Simulation may also require workstation GPUs for
> > acceleration (see below). Rotating a rendered model is rarely a
> > bottleneck for modern graphics cards (or even just integrated GPUs)
> > because rasterization is typically fast enough.
> >
> > I think in the foreseeable future, open source CAD tools will not
> > optimize for GPU processing because consumer grade GPUs don't provide
> > double precision floating point operations, GPUs companies force you
> > to buy workstation GPUs for that. Most open source developers probably
> > don't have access to workstation grade GPUs (at least for the manifold
> > project, we don't) and most users don't have them, so it is unlikely
> > we will try to optimize for that.
> >
> > For better performance, buying better CPUs (with more threads) and
> > more RAM is probably the way to go. 8 threads can probably saturate
> > openscad and manifold, at least for now. For RAM, I will recommend at
> > least 32GB nowadays.
> >
> > If the project is not in a hurry, I will recommend you to wait a bit.
> > AMD's new APU (for laptops/mini PC) is rumored to have a good iGPU,
> > probably similar to mid-range GPUs nowadays (that costs >= $300), but
> > should be much cheaper. Another issue with mid-range GPUs is that they
> > usually have very limited VRAM, so some tasks can't even run on the
> > GPU due to VRAM limits. For iGPU, this is not as much as an issue
> > because you can allocate more RAM to the iGPU. I think this should
> > come out in a few months or so.
> >
> > On 2/18/25 00:53, Jordan Brown via Discuss wrote:
> >> I don't think anybody minds an occasional tangentially-related topic,
> >> as long as it doesn't take over. And if you spun it as "graphics
> >> cards for OpenSCAD", it would be totally on-topic.
> >>
> >> My experience is that it doesn't matter much. I buy cheap computers,
> >> and OpenSCAD has performed fine for me, at least in the graphics
> >> areas. I won't say that I've never had laggy video updates, but it's
> >> pretty rare. I upgrade from the base models for three parameters:
> >> RAM (the more the merrier), persistent storage (128GB doesn't cut
> >> it), and (for laptops) display size.
> >>
> >> The parts of OpenSCAD that seem performance-sensitive are rendering
> >> and, especially with BOSL2, execution. Neither is aided by a better
> >> GPU. (This should, by the way, not be read as a criticism of BOSL2.
> >> It's doing incredible stuff and it's no surprise that it takes time
> >> to do it.) If those are your performance complaints, get a better
> >> CPU. If your performance complaint is that rotating the model is
> >> laggy, a hotter GPU might help.
> >>
> >> I haven't priced high-end graphics cards, but I believe they are more
> >> expensive than midrange systems... and for the environment you
> >> describe, if you have a choice between one computer with a high-end
> >> graphics card and two more modest computers, I know which answer I'd
> >> pick.
> >>
> >>
> >> _______________________________________________
> >> 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
>
