KF performance benchmarks

  • 15 Replies
  • 717 Views

0 Members and 1 Guest are viewing this topic.

Offline claude

  • *
  • 3f
  • ******
  • Posts: 1371
    • mathr.co.uk
« Reply #15 on: June 04, 2018, 05:36:42 PM »
Can't you interleave the reference calculation with the rest? Use the idle processors to start computing first the SA and when that's done the image, while the reference orbit gets calculated? This way there is no waste of resources.

Yes actually the SA could be combined with reference orbit, but the image loop is (for each pixel ; for each iteration) not (for each iteration ; for each pixel) so couldn't be interleaved with reference (the way it is now is much more efficient (memory bandwidth is slow) but requires the whole reference to be ready by the time pixel iterations start).  And you can't start calculating the next reference until the pixels are complete for you to analyze the glitches to know where to put it...


xx
benchmarks

Started by claude on Mandelbulber

8 Replies
534 Views
Last post May 29, 2019, 07:00:08 PM
by claude
clip
SIMD benchmarks

Started by claude on Kalles Fraktaler

2 Replies
115 Views
Last post September 08, 2019, 10:45:25 PM
by claude
xx
M3D Processor Benchmarks

Started by Fitz on Mandelbulb3d

5 Replies
373 Views
Last post October 25, 2018, 08:16:44 PM
by Sabine62
xx
(Fixed precision) floating point performance: Best datatype?

Started by marcm200 on Programming

18 Replies
424 Views
Last post January 30, 2020, 08:11:23 PM
by marcm200
xx
curious for GPU benchmarks of "Ryzen with Vega"

Started by hobold on Programming

0 Replies
366 Views
Last post February 12, 2018, 09:31:39 PM
by hobold