MM
Michael Marx (spintel)
Sat, Oct 11, 2025 12:45 AM
I'm considering getting a new PC.
Have any recent developments changed any consideration of OpenSCAD H/W use?
Particularly GPU related.
Is it still basically CPU bound (presuming lots of memory as that seems cheap these days)?
With Manifold, how do the preferences/advance:
a. CGAL Cache size, and
b. PolySet Cache size,
relate?
I presume, get heaps of memory and make them big still applies?
I think I picked up that with Manifold, CGAL is still used for some functions (minkowski) and for
some geometry complications?
I'm considering getting a new PC.
Have any recent developments changed any consideration of OpenSCAD H/W use?
Particularly GPU related.
Is it still basically CPU bound (presuming lots of memory as that seems cheap these days)?
With Manifold, how do the preferences/advance:
a. CGAL Cache size, and
b. PolySet Cache size,
relate?
I presume, get heaps of memory and make them big still applies?
I think I picked up that with Manifold, CGAL is still used for some functions (minkowski) and for
some geometry complications?
MK
Marius Kintel
Sat, Oct 11, 2025 1:56 AM
The main difference is that we have multi-threading now, which means more cores is good, and as a result more memory will be used.
GPU utilization hasn’t changed. If anything it’s less important right now as rendering is starting to compete with preview.
a. CGAL Cache size, and
b. PolySet Cache size,
These are still not implemented correctly for Manifold. Needs fixing.
I think I picked up that with Manifold, CGAL is still used for some functions (minkowski) and for some geometry complications?
Minkowski of concave operands still goes through the expensive CGAL code path.
-Marius
The main difference is that we have multi-threading now, which means more cores is good, and as a result more memory will be used.
GPU utilization hasn’t changed. If anything it’s less important right now as rendering is starting to compete with preview.
> a. CGAL Cache size, and
> b. PolySet Cache size,
These are still not implemented correctly for Manifold. Needs fixing.
> I think I picked up that with Manifold, CGAL is still used for some functions (minkowski) and for some geometry complications?
>
Minkowski of concave operands still goes through the expensive CGAL code path.
-Marius
P
pca006132
Sat, Oct 11, 2025 2:08 AM
Other than the core count, memory frequency/latency matters for big models.
I don't have data about whether latency or bandwidth matters more, but they
are important.
On Sat, Oct 11, 2025, 09:56 Marius Kintel via Discuss <
discuss@lists.openscad.org> wrote:
The main difference is that we have multi-threading now, which means more
cores is good, and as a result more memory will be used.
GPU utilization hasn’t changed. If anything it’s less important right now
as rendering is starting to compete with preview.
a. CGAL Cache size, and
b. PolySet Cache size,
These are still not implemented correctly for Manifold. Needs fixing.
I think I picked up that with Manifold, CGAL is still used for some
functions (minkowski) and for some geometry complications?
Minkowski of concave operands still goes through the expensive CGAL code
path.
-Marius
OpenSCAD mailing list
To unsubscribe send an email to discuss-leave@lists.openscad.org
Other than the core count, memory frequency/latency matters for big models.
I don't have data about whether latency or bandwidth matters more, but they
are important.
On Sat, Oct 11, 2025, 09:56 Marius Kintel via Discuss <
discuss@lists.openscad.org> wrote:
> The main difference is that we have multi-threading now, which means more
> cores is good, and as a result more memory will be used.
>
> GPU utilization hasn’t changed. If anything it’s less important right now
> as rendering is starting to compete with preview.
>
> a. CGAL Cache size, and
> b. PolySet Cache size,
>
>
> These are still not implemented correctly for Manifold. Needs fixing.
>
> I think I picked up that with Manifold, CGAL is still used for some
> functions (minkowski) and for some geometry complications?
>
>
> Minkowski of concave operands still goes through the expensive CGAL code
> path.
>
> -Marius
>
>
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email to discuss-leave@lists.openscad.org
JB
Jordan Brown
Sat, Oct 11, 2025 2:22 AM
I've been pretty happy with my year-or-two-ago, mid-range PC:
- Intel Core I7-13700 (2100MHz, 16 cores)
- 64GB
- 1TB SSD (I think)
- Intel UHD Graphics 770 (builtin)
- NVIDIA GeForce GT1030 (cheap add-in card for second monitor)
I think the base system cost me somewhat less than US$1000, plus ram and
the graphics card.
I honestly don't know how it's changed OpenSCAD performance, because I
wasn't finding OpenSCAD performance to be painful on my previous
system. But it really sped up OpenSCAD builds, like 5x to 10x. I don't
notice a performance difference between the two monitors, suggesting
that the two GPUs aren't radically different.
And I really like my two 43" TCL 4K televisions that I use as monitors...
I should also note that I don't play video games, so don't really stress
the display hardware.
I've been pretty happy with my year-or-two-ago, mid-range PC:
* Intel Core I7-13700 (2100MHz, 16 cores)
* 64GB
* 1TB SSD (I think)
* Intel UHD Graphics 770 (builtin)
* NVIDIA GeForce GT1030 (cheap add-in card for second monitor)
I think the base system cost me somewhat less than US$1000, plus ram and
the graphics card.
I honestly don't know how it's changed OpenSCAD performance, because I
wasn't finding OpenSCAD performance to be painful on my previous
system. But it really sped up OpenSCAD builds, like 5x to 10x. I don't
notice a performance difference between the two monitors, suggesting
that the two GPUs aren't radically different.
And I really like my two 43" TCL 4K televisions that I use as monitors...
I should also note that I don't play video games, so don't really stress
the display hardware.
JB
Jon Bondy
Sat, Oct 11, 2025 7:50 AM
When did we get multi-threading?
On 10/10/2025 9:56 PM, Marius Kintel via Discuss wrote:
The main difference is that we have multi-threading now, which means
more cores is good, and as a result more memory will be used.
--
This email has been checked for viruses by AVG antivirus software.
www.avg.com
When did we get multi-threading?
On 10/10/2025 9:56 PM, Marius Kintel via Discuss wrote:
> The main difference is that we have multi-threading now, which means
> more cores is good, and as a result more memory will be used.
>
>
--
This email has been checked for viruses by AVG antivirus software.
www.avg.com
MM
Michael Marx (spintel)
Sat, Oct 11, 2025 9:04 AM
What is the basis of the threading, what sort of numbers happen for complex models?
I just had a fiddle, with old/example024.scad, n=4, to generate a decent load. Manifold ~8m F6.
I see 12 active threads during render, I have 12 LPUs* so I presume OpenSCAD/Manifold uses the available CPU thread count.
*Logical Processing Units, i5-12450H, 4 P-cores (8 threads) + 4 E-cores (+4 threads) = 12 threads.
(Tho I did notice non-trivial periods where it went single threading)
Looks like AMD (no E-cores) Ryzen7/9 is the pick.
I presume nobody is thinking that NPUs will have a role? 16-bit would be useless I imagine.
From: Marius Kintel via Discuss [mailto:discuss@lists.openscad.org]
The main difference is that we have multi-threading now, which means more cores is good, and as a result more memory will be used.
What is the basis of the threading, what sort of numbers happen for complex models?
I just had a fiddle, with old/example024.scad, n=4, to generate a decent load. Manifold ~8m F6.
I see 12 active threads during render, I have 12 LPUs* so I presume OpenSCAD/Manifold uses the available CPU thread count.
*Logical Processing Units, i5-12450H, 4 P-cores (8 threads) + 4 E-cores (+4 threads) = 12 threads.
(Tho I did notice non-trivial periods where it went single threading)
Looks like AMD (no E-cores) Ryzen7/9 is the pick.
I presume nobody is thinking that NPUs will have a role? 16-bit would be useless I imagine.
_____
From: Marius Kintel via Discuss [mailto:discuss@lists.openscad.org]
The main difference is that we have multi-threading now, which means more cores is good, and as a result more memory will be used.
P
pca006132
Sat, Oct 11, 2025 9:47 AM
You have to explicitly target GPU/NPU to utilize them. They are
heterogeneous accelerators (instead of CPU SIMD). There is no plan to
target them for manifold, at least for now. GPU acceleration will be
interesting (mainly for the much larger memory bandwidth), but that will
needs a rewrite so there is no plan for it.
On 10/11/25 17:04, Michael Marx (spintel) via Discuss wrote:
What is the basis of the threading, what sort of numbers happen for
complex models?
I just had a fiddle, with old/example024.scad, n=4, to generate a
decent load. Manifold ~8m F6.
I see 12 active threads during render, I have 12 LPUs* so I presume
OpenSCAD/Manifold uses the available CPU thread count.
*Logical Processing Units, i5-12450H, 4 P-cores (8 threads) + 4
E-cores (+4 threads) = 12 threads.
(Tho I did notice non-trivial periods where it went single threading)
Looks like AMD (no E-cores) Ryzen7/9 is the pick.
I presume nobody is thinking that NPUs will have a role? 16-bit would
be useless I imagine.
*From:*Marius Kintel via Discuss [mailto:discuss@lists.openscad.org]
The main difference is that we have multi-threading now, which means
more cores is good, and as a result more memory will be used.
OpenSCAD mailing list
To unsubscribe send an email todiscuss-leave@lists.openscad.org
You have to explicitly target GPU/NPU to utilize them. They are
heterogeneous accelerators (instead of CPU SIMD). There is no plan to
target them for manifold, at least for now. GPU acceleration will be
interesting (mainly for the much larger memory bandwidth), but that will
needs a rewrite so there is no plan for it.
On 10/11/25 17:04, Michael Marx (spintel) via Discuss wrote:
>
> What is the basis of the threading, what sort of numbers happen for
> complex models?
>
> I just had a fiddle, with old/example024.scad, n=4, to generate a
> decent load. Manifold ~8m F6.
> I see 12 active threads during render, I have 12 LPUs* so I presume
> OpenSCAD/Manifold uses the available CPU thread count.
> *Logical Processing Units, i5-12450H, 4 P-cores (8 threads) + 4
> E-cores (+4 threads) = 12 threads.
>
> (Tho I did notice non-trivial periods where it went single threading)
>
> Looks like AMD (no E-cores) Ryzen7/9 is the pick.
>
> I presume nobody is thinking that NPUs will have a role? 16-bit would
> be useless I imagine.
>
> ------------------------------------------------------------------------
>
> *From:*Marius Kintel via Discuss [mailto:discuss@lists.openscad.org]
>
> The main difference is that we have multi-threading now, which means
> more cores is good, and as a result more memory will be used.
>
>
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email todiscuss-leave@lists.openscad.org
MM
Michael Marx (spintel)
Sat, Oct 11, 2025 12:48 PM
With Manifold, in Nightly builds, needs:
Preferences/Advanced/3D_Rendering/Backend [Manifold (new/fast)]
-----Original Message-----
From: Jon Bondy via Discuss [mailto:discuss@lists.openscad.org]
Sent: Saturday, October 11, 2025 6:50 PM
To: OpenSCAD general discussion Mailing-list
Cc: Jon Bondy
Subject: [OpenSCAD] Re: Latest OpenSCAD (incl. Manifold) & PC configuration?
When did we get multi-threading?
On 10/10/2025 9:56 PM, Marius Kintel via Discuss wrote:
The main difference is that we have multi-threading now, which means
more cores is good, and as a result more memory will be used.
With Manifold, in Nightly builds, needs:
Preferences/Advanced/3D_Rendering/Backend [Manifold (new/fast)]
> -----Original Message-----
> From: Jon Bondy via Discuss [mailto:discuss@lists.openscad.org]
> Sent: Saturday, October 11, 2025 6:50 PM
> To: OpenSCAD general discussion Mailing-list
> Cc: Jon Bondy
> Subject: [OpenSCAD] Re: Latest OpenSCAD (incl. Manifold) & PC configuration?
>
> When did we get multi-threading?
>
> On 10/10/2025 9:56 PM, Marius Kintel via Discuss wrote:
> > The main difference is that we have multi-threading now, which means
> > more cores is good, and as a result more memory will be used.
> >
> >
>
> --
> This email has been checked for viruses by AVG antivirus software.
> www.avg.com
> _______________________________________________
> OpenSCAD mailing list
> To unsubscribe send an email to discuss-leave@lists.openscad.org
JB
Jon Bondy
Sat, Oct 11, 2025 1:04 PM
Thanks. Seems like that is the default, since I would not have known
how to change that setting.
On 10/11/2025 8:48 AM, Michael Marx (spintel) via Discuss wrote:
With Manifold, in Nightly builds, needs:
Preferences/Advanced/3D_Rendering/Backend [Manifold (new/fast)]
-----Original Message-----
From: Jon Bondy via Discuss [mailto:discuss@lists.openscad.org]
When did we get multi-threading?
On 10/10/2025 9:56 PM, Marius Kintel via Discuss wrote:
The main difference is that we have multi-threading now, which means
more cores is good, and as a result more memory will be used.
--
This email has been checked for viruses by AVG antivirus software.
www.avg.com
Thanks. Seems like that is the default, since I would not have known
how to change that setting.
On 10/11/2025 8:48 AM, Michael Marx (spintel) via Discuss wrote:
> With Manifold, in Nightly builds, needs:
> Preferences/Advanced/3D_Rendering/Backend [Manifold (new/fast)]
>
>> -----Original Message-----
>> From: Jon Bondy via Discuss [mailto:discuss@lists.openscad.org]
>>
>> When did we get multi-threading?
>>
>> On 10/10/2025 9:56 PM, Marius Kintel via Discuss wrote:
>>> The main difference is that we have multi-threading now, which means
>>> more cores is good, and as a result more memory will be used.
>>>
--
This email has been checked for viruses by AVG antivirus software.
www.avg.com
MK
Marius Kintel
Sat, Oct 11, 2025 1:07 PM
With Manifold, in Nightly builds, needs:
Preferences/Advanced/3D_Rendering/Backend [Manifold (new/fast)]
It’s been the default for a few months.
-Marius
On Oct 11, 2025, at 08:48, Michael Marx (spintel) via Discuss <discuss@lists.openscad.org> wrote:
>
> With Manifold, in Nightly builds, needs:
> Preferences/Advanced/3D_Rendering/Backend [Manifold (new/fast)]
>
It’s been the default for a few months.
-Marius
