TheseCode to evaluate how fast brainrender is and how it compares to other software.

Note: the exact duration on your system will depend on various factors like if other processes are running or the size of the rendering window/screen.

Note: the duration reported in the results does not include scene creation (scene = Scene()) which takes <2s and mst of this time is spent accessing atlas data.

Machines tested

• [1] macOS - Mojave 10.14.6 - MacBook Pro (15-inch, 2019) - 2.3 GHz Intel Core i9 - Radeon Pro 560X 4 GB GPU
• [2] Ubuntu - 18.04.2 LTS x86_64 - Intel i7-8565U (x) @ 4.5GHz - NO GPU
• [3] Windows 10 - Intel(R) Core i7-7700HQ 2.8GHz - NO GPU
• [4] Windows 10 - Intel(R) Xeon(R) CPU E5-2643 v3 3.4GHz - NVIDIA GeForce GTX 1080 Ti

Tests

N cells

Render N cells + the root object. The timer counts how long it takes to create a Points actor representing N cells, adding this to a scene and rendering the scene full screen/

Slice N cells

Render N cells + the root object and slice them. The timer counts how long it takes to create a Points actor representing N cells, adding this to a scene and slicing the actor + the root mesh through the midline, as well as rendering the scene full screen.

Brain regions

Render >1k brain region meshes. Measures how long it takes to fetch and render the meshes for almost all brain regions in the mouse brain.

Animation

Make a short animation

Volume

Parse a numpy array to create a Volume actor (10 times) and render

Results

Test	Machine	GPU	# actors	# vertices	FPS*	run duration	benchmark file
10k cells	[1]	yes	3	1029324	24.76	0.81s	bm_cells.py
	[2]	No	3	1029324	22.46	1.16s	bm_cells.py
	[3]	No	3	1029324	20.00	1.41s	bm_cells.py
	[4]	Yes	3	1029324	100.00	1.34s	bm_cells.py
100k cells	[1]	yes	3	9849324	18.87	3.23s	bm_cells.py
	[2]	No	3	9849324	14.91	4.34s	bm_cells.py
	[3]	No	3	9849324	0.43	7.94s	bm_cells.py
	[4]	Yes	3	9849324	1.20	1.13s	bm_cells.py
1M cells	[1]	yes	3	98049324	2.65	31.01s	bm_cells.py
	[2]	No	3	98049324	2.55	96.49s	bm_cells.py
	[3]	No	3	98049324	0.03	86.75s	bm_cells.py
	[4]	Yes	3	98049324	0.13	36.57s	bm_cells.py
Slicing 10k cells	[1]	yes	3	237751	37.64	0.96s	bm_cells.py
	[2]	No	3	237751	39.10	1.25s	bm_cells.py
	[3]	No	3	237751	26.32	1.88s	bm_cells.py
	[4]	Yes	3	237751	200.00	1.34s	bm_cells.py
Slicing 100k cells	[1]	yes	3	276092	31.79	7.77s	bm_cells.py
	[2]	No	3	276092	25.98	9.09s	bm_cells.py
	[3]	No	3	276092	21.28	16.88s	bm_cells.py
	[4]	Yes	3	276092	111.11	9.65s	bm_cells.py
Slicing 1M cells	[1]	yes	3	275069	11.23	91.31s	bm_cells.py
	[2]	No	3	275069	5.39	104.79s	bm_cells.py
	[3]	No	3	275069	5.03	158.99s	bm_cells.py
	[4]	Yes	3	275069	37.04	97.43s	bm_cells.py
brain regions	[1]	yes	1678	1864388	9.38	11.78s	bm_brain_regions.py
	[2]	No	1678	1864388	7.61	27.40s	bm_brain_regions.py
	[3]	No	1678	1864388	6.49	46.79s	bm_brain_regions.py
	[4]	Yes	1678	1864388	11.90	35.83s	bm_brain_regions.py
animation	[1]	yes	8	96615	9.91	18.98s	bm_animation.py
	[2]	No	8	96615	22.12	12.63s	bm_animation.py
	[3]	No	8	96615	15.15	11.92s	bm_animation.py
	[4]	Yes	8	96615	47.62	12.29s	bm_animation.py
volume	[1]	yes	12	49324	1.79	2.31s	bm_volume.py
	[2]	No	12	49324	1.66	1.95s	bm_volume.py
	[3]	No	12	49324	3.55	2.15s	bm_volume.py
	[4]	Yes	12	49324	23.26	1.21s	bm_volume.py

• the FPS measured are approximate and are meant only as an indication of the expected performance.

Note: Volume actors don't have a number of "vertices" so are not counted heres

Note: these tests are designed to push brainrender to the limits, in practice several steps can be taken to speed up rendering times and increase FPS. This includes decimating/smoothing Actors to reduce the number of vertices or creating meshes with lower resolution to begin with. For most cases a significant increase in performance can be achieved with no noticeable loss in rendering quality. For instance on computer [4] lowering the resolution of the Points actor from 8 to 4 when rendering 1M points tripled the FPS.