I am an OpenScad fan. However, I think my computers are not optimal.

I run it on a Macbook PRO with 2.6Ghz Intel Core i7 with 8GB
and on a Win7 x64 machine. Also with Core i7 and 8GB

Despite that, my redering of complex figures (spheres and hulls and differences and so) takes ages.

So If i would configure a new machine to be optimal for OpenScad usage, what specs would I need to specify ?
Any experienced idea's ?

Thomas

---

www.3daybreaker.blogspot.com

Orca V4.4 rebuild to Ramps with Mk8 and E3D, as well as a Rostock Delta Mini and an OLO in backorder :-)

An is with 8th is plenty of horsepower. What is your value for $fn? If this is too high, it greatly impacts rendering time. I usually use inot the 30 to 50 range.

Certain operations take a lot longer than others. I tend to pay attention to the way I'm writing my code to avoid processor intensive operations or to delay them to a point where they're not subsequently being worked on by other operations (minkowski being a good example).

I work on a variety of MBP and Windows platforms (all i7 with at least 8GB RAM also) and have noticed that whilst hardware does play a part in the performance, it's nowhere near as great an impact as the cleanliness of the code.

Yes
I realize the load for hull and minkowski and stuff, but having said that,
What should I buy?

Intell or Amd
ios or windows or unix flavor
If Microsoft: W7 or W8?
Etc

Maybe specific processor or motherboard for memory drivven performance ..?
What comes next to i7-980?

Edited 1 time(s). Last edit at 09/02/2014 07:40AM by Replace.

---

www.3daybreaker.blogspot.com

Orca V4.4 rebuild to Ramps with Mk8 and E3D, as well as a Rostock Delta Mini and an OLO in backorder :-)

Keep $fn ridiculously low (e.g. maybe 10) while working on a design so that rendering after each change is fast yet still (just) good enough to see what your design looks like, then increase it when the design is complete to give you the final finish. I doubt that any modern OS will make a great deal of difference to the rendering time because the same processor is running essentially the same code, and OS specific code will take an insignificant fraction of the total CPU cycles (unless you have other processor intensive tasks running of course). Any application uses only as much memory as it needs, so increasing beyond that point makes no difference to the time. Having less than the required amount however makes a huge difference because the extra is obtained by paging data to and from your HDD which is very time consuming. I have no idea whether OpenScad can take advantage of multiple cores, though I suspect not. Try to simplify any maths in your design - e.g. use a fixed value of 0.5 instead of sin(30) even if the latter may make the purpose more clear (you could add a note in a comment). Having said that, OpenScad seems to be particularly good at recognising when similar equations are used and re-using pre-computed values rather than performing a completely new calculation every time the same maths expression is encountered, so it is perhaps not as important as it is for other, less intelligent applications.

Dave

I appreciate the code efficiency advice. True,

But that does not really answer my question.

Say that there is code, no matter how efficient.
I want to build a computer for optimum OpenScad usage.
Changing the code does not give me a computer.

So back to the drawing board.
A colleague of mine said yesterday: Don.t focus on the processor but buy a bunch of good Video Cards, plug them parallel in a motherboard and have OpenScad use all GPU's simultaneously.

That seems an advice. But I can not value it.

Will OpenScad use GPU (If present)
Can it use more than one ?

Is it good idea?

Thomas

---

www.3daybreaker.blogspot.com

Orca V4.4 rebuild to Ramps with Mk8 and E3D, as well as a Rostock Delta Mini and an OLO in backorder :-)

I have OpenSCAD installed on one of my SLI rigs and can honestly say that it doesn't appear significantly faster on this than on my MBP. My GPU monitors show no significant increase in activity across even one of the cards, never mind across all of them. If there's an option to tell OpenSCAD to run on my GPUs instead then I'd like to know about it!

EDIT:
Just ran a quick test with one of my early creations that used a lot of minkowski() and hull().

This Windows desktop (a 3.6GHz Core i5 with SLI GTX580) compiled the model in 1m49s.
The MBP (2.2GHz Core i7 with Intel Iris Pro) compiled the model in 2m34s.

The CPUs are of different generations - the Windows desktop is Sandy Bridge and the MBP is Haswell but the speed difference shouldn't be that great - the main improvements across those architectures were in efficiency I think. I'm not sure if OpenSCAD would use the extra HT cores provided by the i7 - I'm not even convinced it uses all the physical ones. The SLI rig had three cards in at one point but I don't remember it being any faster as a result - it was certainly warmer though!

I know it's comparing apples to oranges but I would expect the raw GPU power of the desktop to have a greater speed advantage than it shows here.

Edited 1 time(s). Last edit at 10/21/2014 07:18AM by QuackingPlums.

So,
maybe I should ask for a GPU enabling feature development in Github ?

---

www.3daybreaker.blogspot.com

Orca V4.4 rebuild to Ramps with Mk8 and E3D, as well as a Rostock Delta Mini and an OLO in backorder :-)

I also asked this question on the OpenSCAD mailing list - the consensus is that straight single-thread speed is what will get you the fastest rendering, and that to re-engineer the various dependencies is quite a complex task (and already on the backlog).