Rendering Pipeline Comparison (2022.07.27)

Environment

Type	Name
OS	Windows 10 Pro 64
Processor	Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz (16 CPUs), ~2.9GHz
Memory	32768MB RAM
Device	NVIDIA GeFore RTX 3080
VRAM	10077MB
Display	1920 × 1200 (32 bit) (59 Hz)
Build Configuration	Debug / Windows

Comparison Methods

Bandwidth Comparison^Thibieroz04

\textrm{Bandwidth}_{60\textrm{fps}} = \left(\textrm{W} \times \textrm{H} \times \left[ \textrm{MRT}_{\textrm{BPP}} \times   
 n_{\textrm{MRT}} \times n + \textrm{Z}_\textrm{BPP} \times \textrm{Overdraw} + \textrm{T}_\textrm{BPP} \times \textrm{T}_\textrm{B} + n \times \left(2 \times \textrm{BB}_\textrm{BPP} + \textrm{T}_\textrm{S} \times \textrm{T}_\textrm{BPP} \right ) \right ] + \textrm{C}_\textrm{Geometry} \right ) \times 60 \textrm{Bytes} / \textrm{s}

Storage Comparison

Normal^Koonce07
- R8G8B8A8 vs R10G10B10R2 vs R16G16F vs R16G16B16A16F
  - Deferring Cost
  - Decoding Cost
  - Sampling and Storage
  - Quality
  - Free Components

Duration

Frame Duration

FrameDurationBarAll

FrameDurationBarDeferred

FrameDurationBarDeferredNoDefault

FrameDurationBarForward

FrameDurationBarForwardNoDefault

Light Phase Duration

LightPhaseDurationBarAll

LightPhaseDurationBarAllNoDefault

LightPhaseDurationBarTiled

LightPhaseDurationBarClustered

Render Color Duration

RenderColorDurationBarAll

RenderColorDurationBarDeferred

RenderColorDurationBarDeferredNoDefault

RenderColorDurationBarForward

RenderColorDurationBarForwardNoDefault

RenderColorDurationBarTiled

RenderColorDurationBarTiledNoShading

RenderColorDurationBarClustered

Tile / Cluster Assignment Duration

TileClusterAssignmentDurationBarAll

TileClusterAssignmentDurationBarDeferred

TileClusterAssignmentDurationBarDeferredNoShading

TileClusterAssignmentDurationBarForward

TileClusterAssignmentDurationBarTiled

TileClusterAssignmentDurationBarClustered

Bandwidth

Frame

Pipeline	Frame
	DRAM Read/Write Utilization			DRAM Activity	L1 Cache	L2 Cache
	Percentage utilization of DRAM reads	Percentage utilization of DRAM writes	Total DRAM Read/Write Utilization	Percentage of memory cycles that a read or write request to DRAM was active	Read/write utilization	Read/write utilization
Forward	16.333333	1	17.333333	17.666667	53.666667	13.666667
Forward+	35.333333	2	37.333333	37.333333	24.333333	26.666667
Forward+ 2.5D Culling	35.666667	2	37.666667	37.666667	24.666667	26.666667
Forward+ 2.5D, AABB-based Culling	38.666667	2	40.666667	41	18.666667	29
Forward Clustered	34	2	36	36.333333	23.333333	26
Deferred	20	1	21	21	53	17
Deferred Tiled	37	2	39	39	21	28
Deferred Tiled 2.5D Culling	36.333333	2	38.333333	39.333333	20.333333	27.333333
Deferred Tiled 2.5D, AABB-based Culling	39	2	41	41.666667	15.666667	29.666667
Deferred Tiled (DICE)	36.666667	2	38.666667	39	21	27.666667
Deferred Tiled (DICE) 2.5D Culling	36.333333	2	38.333333	39.333333	21.333333	27.333333
Deferred Tiled (DICE) 2.5D, AABB-based Culling	39.666667	2	41.666667	42	16	29.666667
Deferred Tiled (Intel)	36.666667	2	38.666667	39.333333	21.333333	27.666667
Deferred Clustered	36	2	38	38.666667	20.666667	28

Geometry Phase

Pipeline	Geometry Phase
	DRAM Read/Write Utilization			DRAM Activity	L1 Cache	L2 Cache
	Percentage utilization of DRAM reads	Percentage utilization of DRAM writes	Total DRAM Read/Write Utilization	Percentage of memory cycles that a read or write request to DRAM was active	Read/write utilization	Read/write utilization
Deferred	19.666667	21.333333	41	41.333333	59.666667	42.666667
Deferred Tiled	19.666667	21.666667	41.333333	42	60.333333	43
Deferred Tiled 2.5D Culling	19	21.666667	40.666667	41	60.333333	42.666667
Deferred Tiled 2.5D, AABB-based Culling	18.333333	21	39.333333	39.333333	58.666667	41.333333
Deferred Tiled (DICE)	18.666667	21.666667	40.333333	41	60.333333	42.666667
Deferred Tiled (DICE) 2.5D Culling	19	21.666667	40.666667	41.333333	60.333333	42.666667
Deferred Tiled (DICE) 2.5D, AABB-based Culling	18.666667	21.666667	40.333333	41	60.666667	43
Deferred Tiled (Intel)	18	21	39	39.666667	58.666667	41.666667
Deferred Clustered	18	21	39	39.666667	58.666667	41.666667

Light Phase

Pipeline	Lighting Phase
	DRAM Read/Write Utilization			DRAM Activity	L1 Cache	L2 Cache
	Percentage utilization of DRAM reads	Percentage utilization of DRAM writes	Total DRAM Read/Write Utilization	Percentage of memory cycles that a read or write request to DRAM was active	Read/write utilization	Read/write utilization
Deferred	0	0	0	0	85	3.333333
Deferred Tiled	3	0	3	4	94.666667	4.666667
Deferred Tiled 2.5D Culling	3.333333	0	3.333333	4	94.333333	4.666667
Deferred Tiled 2.5D, AABB-based Culling	4	1	5	5	94.666667	5.333333
Deferred Tiled (DICE)	2	0	2	3	95.666667	4
Deferred Tiled (DICE) 2.5D Culling	2.333333	0	2.333333	3.333333	94.333333	4
Deferred Tiled (DICE) 2.5D, AABB-based Culling	4	1	5	5	93.333333	5
Deferred Tiled (Intel)	2	0	2	3	97	4
Deferred Clustered	3	0	3	4	97	5

Render Color

Pipeline	Render Color
	DRAM Read/Write Utilization			DRAM Activity	L1 Cache	L2 Cache
	Percentage utilization of DRAM reads	Percentage utilization of DRAM writes	Total DRAM Read/Write Utilization	Percentage of memory cycles that a read or write request to DRAM was active	Read/write utilization	Read/write utilization
Forward	0	0	0	0.666667	82.666667	3.333333
Forward+	2	1	3	3.666667	89.333333	5
Forward+ 2.5D Culling	2.333333	1	3.333333	4	92.666667	5.333333
Forward+ 2.5D, AABB-based Culling	3	2	5	5	91	7
Forward Clustered	2.666667	1	3.666667	4.666667	92.666667	6.333333

Tile / Cluster Assignment

Pipeline	Tile / Cluster Assignment
	DRAM Read/Write Utilization			DRAM Activity	L1 Cache	L2 Cache
	Percentage utilization of DRAM reads	Percentage utilization of DRAM writes	Total DRAM Read/Write Utilization	Percentage of memory cycles that a read or write request to DRAM was active	Read/write utilization	Read/write utilization
Forward+	1.333333	1	2.333333	3.333333	23.333333	12
Forward+ 2.5D Culling	1.333333	1	2.333333	2.666667	26	11
Forward+ 2.5D, AABB-based Culling	3.666667	2	5.666667	5.666667	41.666667	14
Forward Clustered	0	4	4	5	43	18
Deferred Tiled	0	1	1	2	24	12
Deferred Tiled 2.5D Culling	0	1	1	2	26	11
Deferred Tiled 2.5D, AABB-based Culling	1	2	3	3	43	14
Deferred Tiled (DICE)	2	0	2	3	95.666667	4
Deferred Tiled (DICE) 2.5D Culling	2.333333	0	2.333333	3.333333	94.333333	4
Deferred Tiled (DICE) 2.5D, AABB-based Culling	4	1	5	5	93.333333	5
Deferred Tiled (Intel)	2	0	2	3	97	4
Deferred Clustered	0.333333	4	4.333333	5	42.333333	18

Shadow Maps

Lights Scalability

^Lauritzen10

Frame Duration

FrameDurationAll

FrameDurationDeferred

FrameDurationDeferredNoDefault

FrameDurationForward

FrameDurationForwardNoDefault

FrameDurationTiled

FrameDurationTiled2_5D

FrameDurationTiled2_5DAABB

FrameDurationClustered

Light Phase Duration

LightPhaseDurationDeferred

LightPhaseDurationDeferredNoDefault

LightPhaseDurationTiled

LightPhaseDurationTiled2_5D

LightPhaseDurationTiled2_5DAABB

LightPhaseDurationClustered

Render Color Duration

RenderColorDurationAll

RenderColorDurationDeferred

RenderColorDurationDeferredNoDefault

RenderColorDurationForward

RenderColorDurationForwardNoDefault

RenderColorDurationTiled

RenderColorDurationTiled2_5D

RenderColorDurationTiled2_5DAABB

RenderColorDurationClustered

Tile / Cluster Assignment Duration

TileClusterAssignmentDurationAll

TileClusterAssignmentDurationDeferred

TileClusterAssignmentDurationForward

TileClusterAssignmentDurationTiled

TileClusterAssignmentDurationTiled2_5D

TileClusterAssignmentDurationTiled2_5DAABB

TileClusterAssignmentDurationClustered

GBuffer: Fat Buffer vs Thin Buffer^Kaplanyan10

Color Space Encoding^Kaplanyan10

Normal Encoding^Kaplanyan10

Frame Duration

NormalCompressionFrameDurationNoDefault

Render Color Duration

NormalCompressionRenderColorDurationNoDefault

Geometry Phase Duration

NormalCompressionGeometryPhaseDurationNoDefault

Lighting Phase Duration

NormalCompressionLightingPhaseDurationNoDefault

False Positive Rate

Forward+

Name		Duration (ms)	Bandwidth
			DRAM RW Utilization			DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
			% Utilization of DRAM Reads	% Utilization of DRAM Writes	Total DRAM RW Utilization	DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
Scene Render	Particle Update	0.091776 0.095168 0.097408 =0.094784	0 0 1 =0.333333	0	0 0 1 =0.333333	0 0 2 =0.666667	0	0 0 1 =0.333333
	Z PrePass	0.083200 0.081600 0.082240 =0.082347	25 24 24 =24.333333	9 6 6 =7	34 30 30 =31.333333	34 31 31 =32	3	34 33 32 =33
	Generate SSAO	0.156384 0.155328 0.156512 =0.156075	8 9 10 =9	7 7 8 =7.333333	15 16 18 =16.333333	15 17 19 =17	47 47 48 =47.333333	32 34 33 =33
	Fill Light Grid	0.060064 0.059296 0.061600 =0.060320	1 1 5 =2.333333	3 3 2 =2.666667	4 4 7 =5	4 4 8 =5.333333	43 43 41 =42.333333	22 22 21 =21.666667
	Main Render	0.031232 0.030720 0.030816 =0.030923	0	0	0	0	0	13
	Render Shadow Map	0.107104 0.090272 0.089888 =0.095755	19 21 21 =20.333333	8 9 9 =8.666667	27 30 30 =29	28 30 30 =29.333333	3	26 28 28 =27.333333
	Render Color	1.317600 1.318592 1.318752 =1.318315	3	2	5	5	91 92 83 =88.666667	7 7 6 =6.666667
	Generate Camera Velocity	0.059232 0.058464 0.059392 =0.059029	8 7 10 =8.333333	5	13 12 15 =13.333333	13 13 15 =13.666667	32 33 32 =32.333333	21 20 20 =20.333333
	Temporal Resolve	0.135456 0.134752 0.137024 =0.135744	42 39 45 =42	17 19 17 =17.666667	59 58 62 =59.666667	60 58 62 =60	68 68 67 =67.666667	69 67 74 =70
	Particle Render	0.085088 0.083296 0.085344 =0.084576	3 3 5 =3.666667	1	4 4 6 =4.666667	4 5 7 =5.333333	2	3 3 4 =3.333333
	Motion Blur	0.052416 0.055840 0.055616 =0.054624	44 50 49 =47.666667	15 14 14 =14.333333	59 64 63 =62	59 64 64 =62.333333	29 28 28 =28.333333	46 47 47 =46.666667
	Total	2.186400 2.170144 2.181408	9 8 9 =8.666667	4	13 12 13 =12.666667	13 13 14 =13.333333	66 65 65 =65.333333	16 15 16 =15.666667

Forward+ 2.5D Culling with AABB-based Culling

Name		Duration (ms)	Bandwidth
			DRAM RW Utilization			DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
			% Utilization of DRAM Reads	% Utilization of DRAM Writes	Total DRAM RW Utilization	DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
Scene Render	Particle Update	0.091584 0.092256 0.091296 =0.091712	0 3 0 =1	0	0 3 0 =1	0 4 0 =1.333333	0	0 2 1 =1
	Z PrePass	0.083328 0.081248 0.081120 =0.081898	25 24 24 =24.333333	9 10 10 =9.666667	34	34 35 34 =34.333333	3	34 33 32 =33
	Generate SSAO	0.161536 0.159520 0.154848 =0.158634	8	7	15	15 15 16 =15.333333	48	33
	Fill Light Grid	0.059744 0.059776 0.059584 =0.059701	1	3	4	4	43	22
	Main Render	0.030752 0.030720 0.030720 0.030731	0	0	0	0	0	13
	Render Shadow Map	0.106656 0.089920 0.089760 =0.095445	19 21 21 =20.333333	7 9 9 =8.333333	26 30 30 =28.666667	27 30 30 =29	3	26 28 28 =28.666667
	Render Color	1.308896 1.311968 1.316896 =1.312587	3 3 4 =3.333333	2	5 5 6 =5.333333	5 5 6 =5.333333	92 92 84 89.333333	7
	Generate Camera Velocity	0.058944 0.058464 0.058624 =0.058677	6 8 8 =7.333333	5 5 4 =4.666667	11 13 12 =12	12 13 13 =12.666667	32 32 33 =32.666667	19 21 22 =20.666667
	Temporal Resolve	0.135200 0.134208 0.135776 =0.135061	44 38 43 =41.666667	17	61 55 60 =58.666667	61 56 60 =59	67 68 67 =67.333333	69 66 72 =69
	Particle Render	0.083584 0.083776 0.083936 =0.083765	3 5 3 =3.666667	1	4 6 4 =4.666667	5 7 4 =5.333333	2	3
	Motion Blur	0.052544 0.052224 0.052768 =0.052512	45 46 45 =45.333333	15 14 15 =14.666667	60	61 60 60 =60.333333	29 30 29 =29.333333	46 47 46 =46.333333
	Total	2.186400 2.161280 2.162144 =2.169941	8	4	12	12	62 65 62 =63	15

Forward Clustered

Name		Duration (ms)	Bandwidth
			DRAM RW Utilization			DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
			% Utilization of DRAM Reads	% Utilization of DRAM Writes	Total DRAM RW Utilization	DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
Scene Render	Particle Update	0.094080 0.098720 0.090016 =0.094272	0	0	0	0	0	0
	Z PrePass	0.083136 0.080256 0.081248 =0.081546	25 24 24 =24.333333	6 6 10 =7.333333	31 30 34 =31.666667	32 31 35 =32.666667	3	34 33 33 =33.333333
	Generate SSAO	0.158592 0.155296 0.167648 =0.160512	7 8 8 =7.666667	7 8 7 =7.333333	14 16 15 =15	14 17 15 =15.333333	47	31 31 32 =31.333333
	Fill Light Grid	0.454208 0.450720 0.454272 =0.453067	0	1	1	1	38 39 38 =38.333333	21
	Main Render	0.031392 0.030752 0.031168 =0.031104	3 0 0 =1	3 0 0 =1	6 0 0 =2	7 0 0 =2.333333	0	17 15 15 =15.666667
	Render Shadow Map	0.106048 0.089184 0.089056 =0.094763	19 21 21 =20.333333	8 9 9 =8.666667	27 30 30 =29	28 30 31 =29.666667	3	26 28 28 =27.333333
	Render Color	2.018272 2.024256 2.027040 =2.023189	2	1	3	3	89 88 91 =89.333333	5
	Generate Camera Velocity	0.059264 0.058496 0.059264 =0.059008	6 10 12 =9.333333	5 4 4 =4.333333	11 14 16 =13.666667	12 15 16 =14.333333	32	18 21 23 =20.666667
	Temporal Resolve	0.136320 0.134176 0.136032 =0.135509	43 37 43 =41	16 17 16 =16.333333	59 54 59 =57.333333	60 55 60 =58.333333	67 68 67 =67.333333	69 65 74 =69.333333
	Particle Render	0.083680 0.084256 0.082656 =0.083531	3 5 3 =3.666667	1 2 1 =1.333333	4 7 4 =5	5 7 5 =6.333333	2	3
	Motion Blur	0.063776 0.054368 0.054528 =0.057557	45 55 45 =48.333333	20 17 20 =19	65 72 65 =67.333333	66 72 66 =68	29 26 29 =28	40 42 41 =41
	Total	3.295520 3.267296 3.279776 =3.280864	5 5 6 =5.333333	2 3 3 =2.666667	7 8 9 =8	8 9 9 =8.666667	70 67 68 =68.333333	13

Deferred Tiled

Name		Duration (ms)	Bandwidth
			DRAM RW Utilization			DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
			% Utilization of DRAM Reads	% Utilization of DRAM Writes	Total DRAM RW Utilization	DRAM Activity	L1 Cache RW Utilization	L2 Cache RW Utilization
Scene Render	Particle Update	0.094080 0.098720 0.090016 =0.094272	0	0	0	0	0	0
	Z PrePass	0.083136 0.080256 0.081248 =0.081546	25 24 24 =24.333333	6 6 10 =7.333333	31 30 34 =31.666667	32 31 35 =32.666667	3	34 33 33 =33.333333
	Generate SSAO	0.158592 0.155296 0.167648 =0.160512	7 8 8 =7.666667	7 8 7 =7.333333	14 16 15 =15	14 17 15 =15.333333	47	31 31 32 =31.333333
	Fill Light Grid	0.454208 0.450720 0.454272 =0.453067	0	1	1	1	38 39 38 =38.333333	21
	Main Render	0.031392 0.030752 0.031168 =0.031104	3 0 0 =1	3 0 0 =1	6 0 0 =2	7 0 0 =2.333333	0	17 15 15 =15.666667
	Render Shadow Map	0.106048 0.089184 0.089056 =0.094763	19 21 21 =20.333333	8 9 9 =8.666667	27 30 30 =29	28 30 31 =29.666667	3	26 28 28 =27.333333
	Render Color	2.018272 2.024256 2.027040 =2.023189	2	1	3	3	89 88 91 =89.333333	5
	Generate Camera Velocity	0.059264 0.058496 0.059264 =0.059008	6 10 12 =9.333333	5 4 4 =4.333333	11 14 16 =13.666667	12 15 16 =14.333333	32	18 21 23 =20.666667
	Temporal Resolve	0.136320 0.134176 0.136032 =0.135509	43 37 43 =41	16 17 16 =16.333333	59 54 59 =57.333333	60 55 60 =58.333333	67 68 67 =67.333333	69 65 74 =69.333333
	Particle Render	0.083680 0.084256 0.082656 =0.083531	3 5 3 =3.666667	1 2 1 =1.333333	4 7 4 =5	5 7 5 =6.333333	2	3
	Motion Blur	0.063776 0.054368 0.054528 =0.057557	45 55 45 =48.333333	20 17 20 =19	65 72 65 =67.333333	66 72 66 =68	29 26 29 =28	40 42 41 =41
	Total	2.051136	11	5	16	16	56	18

Optimization

Locate the bottleneck of the pipeline
Optimize that stage

Pipelines to optimize:

Forward+
Forward Clustered
Deferred Tiled
Deferred Clustered
Deferred Thin G-Buffer Tiled
Deferred Thin G-Buffer Clustered

Locating the Bottleneck

Finding bottlenecks
- Set up several tests where each test decreases the amount of work a particular stage performs
  - If one of these test causes the frames per second the increase, the bottleneck stage has been found.

Multiplatform GPU High-Level Optimization Guidelines^{SousaKasyanSchulz12}

Generalize and always optimize for the worst case scenario
1. Discover the biggest bottlenecks and address them by tackling the biggest time consumer. This means avoiding partial optimizations.
  - Ex) Crysis 1: If the camera was static, then motion blur was disabled. If the camera was moving fast, then motion blur was enabled. This kind of bad optimization strategy resulted in big performance peaks and an inconsistent frame rate.
2. Once done, repeat ad nauseam!
Don’t repeat work or do unnecessary work. For example:
1. Don’t down-sample full-screen color targets or depth targets multiple times for different postprocessing functinos
2. Minimize the number of memory transfers, render target clears, and any redundant full-screen passes
3. Such repeated or redundant work adds up very quickly
  - Ex) A full-screen pass a 720 p costs ca. 0.25 ms on the Xbox 360 and ca. 0.4 ms on PS3. It is very easy to spend many milliseconds in a wasteful manner.
4. Batch as much as possible in a single pass
Take advantage of interframe coherency. Amortize costs across frames:
1. This can provide a significant gain if done carefully, talking performance peaks and multi-GPU systems into account
2. Distribute costs evenly
  - Ex) If the HUD updates every nth frame, then every n + 1-th frame update some similar-costing render technique
3. For screen-space ambient occlusion(SSAO) and the like, the cost can be distributed across frames
In the end, the key words for the most cases are: “share, share, share.” Share as many computations and as much bandwidth as is reasonably possible in a single pass.

Multiplatform Optimization: Best Practice^{SousaKasyanSchulz12}

Timers that showed where the GPU costs were located
- Shadows, lighting, post processes
Visualization tools
- Lighting, scene overdraw visualization

References

Deferred Shading with Multiple Render Targets. Nicolas Thibieroz, PowerVR Technologies / AMD. ShaderX2.
Deferred Shading in Tabula Rasa. Rusty Koonce, NCSoft Corporation / Facebook. GPU Gems 3.
CryENGINE 3: Reaching the Speed of Light. Anton Kaplanyan, Crytek / Intel Corporation. SIGGRAPH 2010: Advances in Real-Time Rendering in Games Course.
Deferred Rendering for Current and Future Rendering Pipelines. Andrew Lauritzen, Intel Corporation. SIGGRAPH 2010: Beyond Programmable Shader Course.
Rendering Tech of Space Marine. Pope Kim, Relic Entertainment / POCU. Daniel Barrero, Relic Entertainment. KGC 2011.
Tiled Shading. Ola Olsson, Chalmers University of Technology / Epic Games. Ulf Assarsson, Chalmers University of Technology. Journal of Graphics, GPU, and Game Tools.
CryENGINE 3: Three Years of Work in Review. Tiago Sousa, Crytek / id Software. Nickolay Kasyan, Crytek / AMD. Nicolas Schulz, Crytek. GPU Pro 3.

import matplotlib.pyplot as plt
import numpy as np

x = np.arange(0, 3 * np.pi, 0.1)
y = np.sin(x)

x = ["16/8/8", "32/24/8", "64/56/8", "128/56/8"]
forward_plus_y = [1.293643, 1.402208, 1.691659, 1.991093]
forward_plus_culling_y = [1.253589, 1.318080, 1.516608, 1.718923]
forward_clustered_y = [2.174741, 2.318197, 2.950581, 3.268107]
deferred_tiled_y = [1.370539, 1.453547, 1.709749, 1.972149]
deferred_tiled_culling_y = [1.321376, 1.372097, 1.554155, 1.718635]
deferred_tiled_dice_y = [1.579851, 1.630304, 1.849120, 2.059509]
deferred_tiled_dice_culling_y = [1.367307, 1.450763, 1.607851, 1.739381]
deferred_clustered = [1.369045, 2.354485, 2.935424, 3.519659]

plt.plot(x, forward_plus_y, label="Forward+")
plt.plot(x, forward_plus_culling_y, label="Forward+ 2.5D, AABB Culling")
plt.plot(x, forward_clustered_y, label="Forward Clustered")
plt.plot(x, deferred_tiled_y, label="Deferred Tiled")
plt.plot(x, deferred_tiled_culling_y, label="Deferred Tiled 2.5D, AABB Culling")
plt.plot(x, deferred_tiled_dice_y, label="Deferred Tiled DICE")
plt.plot(x, deferred_tiled_dice_culling_y, label="Deferred Tiled DICE 2.5D, AABB Culling")
plt.plot(x, deferred_clustered, label="Deferred Clustered")
plt.xlabel("Number of Lights (Point/Cone/Cone Shadowed)")
plt.ylabel("Scene Render Duration (ms)")
plt.legend()
plt.show()

Rendering Pipeline Comparison (2022.07.27)

Environment

Comparison Methods

Bandwidth ComparisonThibieroz04

Storage Comparison

Duration

Frame Duration

Light Phase Duration

Render Color Duration

Tile / Cluster Assignment Duration

Bandwidth

Frame

Geometry Phase

Light Phase

Render Color

Tile / Cluster Assignment

Shadow Maps

Lights Scalability

Frame Duration

Light Phase Duration

Render Color Duration

Tile / Cluster Assignment Duration

GBuffer: Fat Buffer vs Thin BufferKaplanyan10

Color Space EncodingKaplanyan10

Normal EncodingKaplanyan10

Frame Duration

Render Color Duration

Geometry Phase Duration

Lighting Phase Duration

False Positive Rate

Forward+

Forward+ 2.5D Culling with AABB-based Culling

Forward Clustered

Deferred Tiled

Optimization

Locating the Bottleneck

Multiplatform GPU High-Level Optimization GuidelinesSousaKasyanSchulz12

Multiplatform Optimization: Best PracticeSousaKasyanSchulz12

References

Bandwidth Comparison^Thibieroz04

GBuffer: Fat Buffer vs Thin Buffer^Kaplanyan10

Color Space Encoding^Kaplanyan10

Normal Encoding^Kaplanyan10

Multiplatform GPU High-Level Optimization Guidelines^{SousaKasyanSchulz12}

Multiplatform Optimization: Best Practice^{SousaKasyanSchulz12}