Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > comp.soft-sys.math.mathematica > #6532 > unrolled thread
| Started by | Ralph Dratman <ralph.dratman@gmail.com> |
|---|---|
| First post | 2011-11-19 11:50 +0000 |
| Last post | 2011-11-20 10:42 +0000 |
| Articles | 2 — 2 participants |
Back to article view | Back to comp.soft-sys.math.mathematica
Timing graphics in the real world Ralph Dratman <ralph.dratman@gmail.com> - 2011-11-19 11:50 +0000
Re: Timing graphics in the real world "Oleksandr Rasputinov" <oleksandr_rasputinov@hmamail.com> - 2011-11-20 10:42 +0000
| From | Ralph Dratman <ralph.dratman@gmail.com> |
|---|---|
| Date | 2011-11-19 11:50 +0000 |
| Subject | Timing graphics in the real world |
| Message-ID | <ja856s$iiq$1@smc.vnet.net> |
I am trying to find out why a piece of 3D graphics code takes a long
time when scaled up to a lot of little objects. For this purpose I
begin by timing ten runs of a small instance. Here's what my function
calls look like:
noneOfThat[] :=
Module[{}, {drawSolidCube[], drawSolidCube[], drawSolidCube[],
drawSolidCube[], drawSolidCube[], drawSolidCube[], drawSolidCube[],
drawSolidCube[], drawSolidCube[], drawSolidCube[]}]
where SolidCube is actually not solid, but rather an aggregate of
about 20 cubelets per image in this first test case.
AbsoluteTiming[noneOfThat[]]
{0.072105, ...}
or if you prefer,
Timing[noneOfThat[]]
{0.071197, ...}
The timing answers are consistent within about 10% across several tries. Nice.
Just one problem: here in my universe, on a Core Duo Mac Mini at 2.4
GHz with about 2 GB of free memory, the actual process described above
takes about 10 seconds. The cores stay at around 50% usage (each)
while this is happening. Drawing is taking about 140 times longer than
Mathematica says!
Next I try the same task with a 2x increase in linear size of the
cube, leading to an 8x increase in volume and number of objects, with
again 10 repetitions in straight-line Mathematica code.
AbsoluteTiming then says 0.62970 while Timing says 0.574067, both in
the neighborhood of 8 times the earlier figures. Real-world time is
about 85 seconds, also in line with the 8x scaling.
Again the processors seem to be about half busy, and once again it
takes about 140 times as long to do the real drawing as Mathematica
reports.
I'm speculating all this means Mathematica is sending my graphics
hardware a bunch of asynchronous drawing calls while simultaneously
announcing, "Done!"
I infer that a faster graphics card might make a world of difference.
To make that possible, I could switch to a Windows or Linux system if
necessary.
If possible, I would like the new real-world drawing time to be at
least ten times better than the figures reported here.
Comments, please? I would be particularly interested to learn about
any specific hardware that might improve the speed.
Thank you very much.
Ralph Dratman
[toc] | [next] | [standalone]
| From | "Oleksandr Rasputinov" <oleksandr_rasputinov@hmamail.com> |
|---|---|
| Date | 2011-11-20 10:42 +0000 |
| Message-ID | <jaalhq$18h$1@smc.vnet.net> |
| In reply to | #6532 |
On Sat, 19 Nov 2011 11:50:52 -0000, Ralph Dratman <ralph.dratman@gmail.com> wrote: > I am trying to find out why a piece of 3D graphics code takes a long > time when scaled up to a lot of little objects. > > [...] > > Drawing is taking about 140 times longer than > Mathematica says! > > [...] > > Ralph Dratman > The reason for this is that Timing and AbsoluteTiming reflect only the time taken for the kernel to build the Graphics expression. The actual rendering of the graphic is done by the front end, and although this takes most of the total time it is not reflected in the reported timings. Certainly the rendering speed for complicated Graphics3D expressions is not particularly impressive, and in the past there have been faster alternative renderers capable of interpreting the kernel's Graphics output, such as JavaView (http://www.javaview.de/mathematica/index.html; unfortunately limited to pre-version 6 Graphics expressions), but I don't know of any such that will understand the output from more recent Mathematica kernels. What, if anything, will speed up the front end's rendering, I do not know--this is heavily dependent on implementation details that are likely known only to WRI.
[toc] | [prev] | [standalone]
Back to top | Article view | comp.soft-sys.math.mathematica
csiph-web