Kalles Fraktaler 2.14

  • 137 Replies
  • 5799 Views

0 Members and 1 Guest are viewing this topic.

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #120 on: January 01, 2020, 09:17:22 PM »
By request (in a previous version) I made Ctrl-S/Save generate timestamped files to avoid accidental overwrite, you can change this in the settings (Advanced -> Save overwrites existing file)
Never seen such behavior on any software. "Save (Ctrl-S)" always saves current work in current file, making a silent backup sometimes (like the *~ Emacs files), not demoting original file to backup as done here. Maybe consider normalizing this?

Regarding kf400 blobs, below same thing rendered in UF without anything special (Resc=1e6 (or 1e26, no difference), standard 64bit floats); no circles.
So it must be related to PT somehow? In the KF example kf400 I posted the white blobs are not there when rendered at 640X360, only when you render at high resolution (I did 20000^2) do they show up.

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #121 on: January 01, 2020, 10:04:42 PM »
Here's another good testing location, all with no-guess/glitch-hitol, with floatexp, longdouble, double and double + argminZ reference.

UF seems to fail completely on this one (blank image result).

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #122 on: January 04, 2020, 09:39:30 PM »
This one (tried all permutations of settings) is fine zooming out just a little bit.

Offline claude

  • *
  • 3f
  • ******
  • Posts: 1373
    • mathr.co.uk
« Reply #123 on: January 04, 2020, 09:52:47 PM »
I ran into similar issue at 4-fold embedded Julia set. Don't know the cause.

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #124 on: January 13, 2020, 10:03:22 PM »
Could you remind me what "size n m" means which is printed in offline renders?
I'm sure you've mentioned it here but hard to find. Maybe add to KF docs?

Offline claude

  • *
  • 3f
  • ******
  • Posts: 1373
    • mathr.co.uk
« Reply #125 on: January 14, 2020, 12:04:10 AM »
There are two overlapping printouts, which is causing the confusion I think.  There is a status message printed every second or two (using carriage return to go to start of line overwrite so as not to flood the terminal with a sequence of lines) showing progress within each reference, then when each reference starts there is another printout (which sometimes has the chopped remainder of the previous status message left over at the right hand end of the screen: this part is meaningless).

I think (can't check right now) that the per-reference printout has the approximate size (in pixels) of the glitch region in which it is putting a new reference, so usually this number decreases, and if you have "ignore small glitches" enabled then the image should be finished by the time it gets to 1.

If the overlap is too annoying and you don't need the regular status updates you can disable the more frequent status messages with the log level control (--help should list the available levels, not sure if all of them change something at the moment).

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #126 on: January 14, 2020, 12:58:25 AM »
Not confusing, I just would like to know what "size 177 4342734 means for example.
And the big number at the end of the progress display that's updated per glitch?
When you get around it can you document it?
How to turn on/off is already well documented.

Offline claude

  • *
  • 3f
  • ******
  • Posts: 1373
    • mathr.co.uk
« Reply #127 on: January 14, 2020, 03:06:23 AM »
Here's an example of what I mean.  Left-over truncated (thus mostly incomplete/meaningless) per-second status messages I have made bold, the rest is per-reference / one-off messages:

Quote
kf 2.14.8 (c) 2013-2017 Karl Runmo, (c) 2017-2019 Claude Heiland-Allen
loading location test.kfr
rendering at 640x360
reference 1
reference 2 at (268,311) size 48 T   230400
reference 3 at (354,151) size 29 T     2240
reference 4 at (188,127) size 25 T     1567
reference 5 at (279,188) size 22 T     1565
reference 6 at (311,137) size 15 T      534
reference 7 at (328,222) size 5
reference 8 at (457,119) size 5  T      154
reference 9 at (313,138) size 3  T      134
reference 10 at (186,127) size 2
reference 11 at (356,150) size 3 T       30
reference 12 at (310,215) size 2 T       16
reference 13 at (325,171) size 2
no more glitches R  69%  A   3%  T        2
colouring final image
saving TIFF test.tif
all done, exiting

The line:
Quote
reference 8 at (457,119) size 5
means a glitch with area 5 pixels is the largest, to solve the remaining glitches KF is adding a reference at pixel coordinates (457,119)

The rest of the line is truncated stuff:
Quote
  T      154
depending on image size/reference position/reference count even more of the string can be truncated, so the T might be missing or even the first digits of the following number.

An example of a non-truncated per-second status message sequence is:
Quote
P   0%  G   0%  R  73%  A  24%  T   230400
 P   0%  G   0%  R  73%  A  56%  T   230400
 P  14%  G   0%  R  73%  A  61%  T   230400
 P  32%  G   1%  R  73%  A  61%  T   230400
 P  50%  G   3%  R  73%  A  61%  T   230400
 P  70%  G   6%  R  73%  A  61%  T   230400
 P  88%  G   7%  R  73%  A  61%  T   230400
reference 2 at (268,311) size 48
 P   0%  G   0%  R  66%  A  15%  T     2240

P is percentage of remaining pixels that have been (re)computed with this reference
G is percentage of remaining pixels that have been guessed (neighbouring pixels have the same integer iteration count, pixel is interpolated)
R is percentage of maximum iterations that have been computed for reference
A is percentage of maximum iterations that have been computed for series approximation
T is total number of pixels that are to be computed with this reference

I don't think this changed between kf-2.14.8 and kf-2.14.9

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #128 on: January 14, 2020, 05:42:27 PM »
Aha, so I should have been confused but wasn't, which was my confusion.
It's fun to look at these numbers when rendering.

Offline Chronicler1701

  • *
  • Fractal Freshman
  • *
  • Posts: 7
« Reply #129 on: February 08, 2020, 12:52:54 AM »
I’m wondering how many cores/threads the program can utilize during a render.  AMD just released their 64c/128t Threadripper 3990X, so I’m curious to know if KF could take full advantage of this monster of a CPU.

Offline claude

  • *
  • 3f
  • ******
  • Posts: 1373
    • mathr.co.uk
« Reply #130 on: February 08, 2020, 12:38:38 PM »
KF should be able to use all the cores/threads if they are supported by Windows or Wine. But the only fully parallel part of the algorithm is per-pixel perturbation iterations, the reference calculations use at most a few threads (only more than 1 at deeper zooms).  Likewise Newton zooming uses only a few threads, the work is inherently sequential.

The coordination between threads is using atomic operations (e.g. an atomic counter for "next pixel to render"), not sure how performance scales to very high core counts.

For tasks like rendering zoom videos you might get more throughput running more than one KF at a time (combining the output may be awkward though) especially at locations dominated by reference calculations.

Offline CFJH

  • *
  • Fractal Frogurt
  • ******
  • Posts: 483
« Reply #131 on: February 08, 2020, 11:46:47 PM »
here's what I've observed:
calculation of a reference: 3 threads
Calculation "Newton Rhaspon" 4 threads
Calculation "zoom out" with "reuse reference" average 1.5 Threads (the calculation there the "A" counts seems to be single threaded)

all at higher zooms above 1e30000

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #132 on: February 09, 2020, 08:51:19 AM »
I’m wondering how many cores/threads the program can utilize during a render.  AMD just released their 64c/128t Threadripper 3990X, so I’m curious to know if KF could take full advantage of this monster of a CPU.
If you get to try it please post results.

A while ago I considered getting the 32c Threadripper but then I read 16 of the CPUs were put on there 50% for show, as some communication hardware on the m-board could not handle more than 16 so the extra 16 would be useless unless your program would be restricted in certain ways and it was not clear to me (or Claude who I asked) if KF would be OK or not.

I never figured it out but just settled for the 16core one to be safe (but slow).

Offline Dinkydau

  • *
  • Fractal Furball
  • ***
  • Posts: 263
    • DeviantART gallery
« Reply #133 on: February 09, 2020, 04:30:09 PM »
I don't have the answer either but some extra information and ideas:

The latest threadripper CPUs (that have names in the 3000 - 3999 range) don't have the problem with different memory latency per numa node anymore, so there's no need to worry about that.

Performance of parallel workloads scales linearly with the number of cores as long as there are no other bottlenecks, so the question is if there are such bottlenecks. Memory speed may be important, because perturbation inference calculations make use of the orbit of the reference pixel. The orbit can take up gigabytes of RAM when the iteration count is high and so it can't fit in the cache and has to be read from memory by 128 threads in parallel. Can the slowest DDR4 memory handle that? I'm not sure. 128 is a lot.

If you use windows you should investigate how to deal with the scheduler because it has problems when you try to use too many threads. An example page with some information: https://www.anandtech.com/show/15483/amd-threadripper-3990x-review/3
Linux supposedly doesn't have that problem.

Offline gerrit

  • *
  • 3f
  • ******
  • Posts: 1979
« Reply #134 on: February 14, 2020, 08:21:41 PM »
Problem with glitch correction. Green lakes should be minis.


xx
Kalles Fraktaler 2.13

Started by claude on Kalles Fraktaler

90 Replies
4320 Views
Last post March 20, 2019, 05:32:26 AM
by claude
xx
Kalles Fraktaler 2 + GMP

Started by claude on Kalles Fraktaler

232 Replies
10452 Views
Last post September 12, 2019, 10:27:53 PM
by claude
xx
Another version of Kalles Fraktaler?

Started by greentexas on Kalles Fraktaler

9 Replies
997 Views
Last post September 22, 2017, 02:59:16 PM
by greentexas
xx
Kalles Fraktaler Movie Maker64

Started by Bill Snowzell on Fractal movie gallery

0 Replies
259 Views
Last post February 01, 2018, 09:29:04 PM
by Bill Snowzell
question
Kalles Fraktaler tiled rendering

Started by Chronicler1701 on Kalles Fraktaler

5 Replies
368 Views
Last post September 06, 2018, 07:43:16 PM
by gerrit