<Guest85>
color([0.1,0.2,0.5, 0.9]) // dark bluish panels
<Guest85>
difference() {
<Guest85>
// Shell made of subdivided icosphere
<Guest85>
// (Simplified: represent as a hollow sphere with panel texture implied)
<Guest85>
cube(battery_size, center=true);
<Guest85>
}
<Guest85>
}
bitbasher has joined #openscad
Guest85 has quit [Quit: Client closed]
mmu_man has joined #openscad
mmu_man has quit [Ping timeout: 245 seconds]
sculptor has joined #openscad
sculptor has quit [Changing host]
sculptor has joined #openscad
sculptor has quit [Client Quit]
sculptor has joined #openscad
<bitbasher>
do we know who is providing the OpenSCAD Playground .. i just tested Jordan's Hello There animation but it complains the objecxt() is not a known function
ToAruShiroiNeko has quit [Ping timeout: 276 seconds]
To_Aru_Shiroi_Ne has joined #openscad
mmu_man has quit [Ping timeout: 252 seconds]
To_Aru_Shiroi_Ne has quit [Read error: Connection reset by peer]
bitbasher has quit [Ping timeout: 245 seconds]
ToAruShiroiNeko has joined #openscad
HumanG331 has quit [Ping timeout: 252 seconds]
mmu_man has joined #openscad
RichardPotthoff has joined #openscad
RichardPotthoff has quit [Changing host]
RichardPotthoff has joined #openscad
mmu_man has quit [Ping timeout: 244 seconds]
mmu_man has joined #openscad
bitbasher has joined #openscad
<bitbasher>
cool .. but he wont want to update to a dev snapshot .. i will just make the examples available on our own docs page .. specifically the animation by Jordan
<teepee>
I always wanted a simple self hosted version on our file server
<teepee>
but I don't know when I'll find the time to look into this. too many open topics...
<TylerTork>
OK I am perplexed. Windows PC with Intel Xeon @ 3.5 GHz, 64 GB RAM renders a model in 15s. Same script on Intel i5 TylerTork 1 GHz and 8GB RAM takes 9s. Backend is Manifold for both. What's going on?
<teepee>
NUMA maybe?
<TylerTork>
I don't know what that is and an internet search is not helping
<teepee>
is it a multi-core xeon, e.g. with more than one cpu?
<teepee>
there are such systems where each cpu core has local memory but the OS sees it combined
<teepee>
if the OS is dumb enough to load programs into the ram of the other core, things get really a performance hit
<teepee>
NUMA = non uniform memory architecture
<TylerTork>
E5-1650 v2 is a 6-core processor. The OS is Windows 10 pro. Is it that dumb?
<teepee>
windows is generally bad at numa, I guess it normally relies on the applications to handle this
<teepee>
no idea of that improved lately
<teepee>
but i guess if that's a single cpu with 6 cores, it's not a NUMA case
<teepee>
I suppose that mostly applies to boards with more than one processor slot
<TylerTork>
Okay, so do you have other theories? It's not the same daily version of OpenSCAD but I would not expect the latest version to be a lot slower than an older one...? You do benchmarks, right?
<teepee>
no
<TylerTork>
no you don't do benchmarks? Or no a newer version should not be slower?
<teepee>
no benchmarks
<teepee>
I would not expect newer versions to be much slower, but depending on the specific model there could be special cases
<TylerTork>
that's disappointing. I guess I'd best update the software on the weaker system and see whether that slows it down.
<teepee>
yup, using the latest on both is probably a useful check
<TylerTork>
OK, installing the latest did slow it down by about 10%. This is 0401 versus 0725. That doesn't account for all the difference.
<TylerTork>
I'm using BOSL2 and I also installed the latest of that, to match with the newer system. No difference in timing. I bought this box specifically to get better performance in OpenSCAD and it really irks me that it's much worse. Any ideas? Anyone?
<teepee>
random test: try disabling threads or whatever that is called
<teepee>
Intel® Hyper-Threading Technology
<TylerTork>
is that an OS function or an OpenSCAD setting?
<teepee>
OS or BIOS I would assume
<TylerTork>
Id OpenSCAD a single-threaded task?
<teepee>
no, if you have Manifold enabled, it can use multiple threads
<TylerTork>
then it seems to me disabling HyperThreading would make things much worse.
marcus has quit [Remote host closed the connection]
<teepee>
no, hyperthreading is faking twice the number of real threads
<teepee>
sharing cpu resource in the hope those 2 threads can work in parallel
marcus has joined #openscad
<teepee>
those 2 "hyper" threads collide on the same physical resources that can cause issues
<teepee>
if you disable hyperthreads, you get 6 real thread, one for each cpu core
<TylerTork>
I seem to hear you saying OpenSCAD doesn't handle hyperthreading very well.
<teepee>
I have no idea, but it is a possibility as I would assume the calculations that happen are pretty similar for all the processing
<TylerTork>
currently I have twice as many logical processors as cores
<teepee>
yep, half of them are fake :)
<TylerTork>
All right, I guess I'll give it a go. Seems like six should be enough for any sensible person anyway
<teepee>
you can always switch it on again after the test
<teepee>
cpus are crazy complex, so *really* knowing what goes on is not easy
TylerTork has quit [Quit: Client closed]
TylerTork has joined #openscad
<TylerTork>
Ok I disabled hyperthreading on the new supposedly faster system and I am back with a report. That did help, but I'm still seeing faster rendering on the 1GHz system than the multi-core 3.3Ghz system when what I hoped for was at least a tripling of speed on the new system. So -- anyone else have any ideas?
<teepee>
buy more 1ghz systems :)
<teepee>
no, no more ideas for today, I'm off to get some sleep
<bitbasher>
night
<teepee>
what's the current numbers with the hyperthreading change and version sync?
<TylerTork>
Total rendering time: 8.644 versus 5.979
<sculptor>
robin hood code - it punishes the users with more expensive systems