HIGH PERFORMANCE VIDEO ENCODING
USING NVIDIA GPUS
Abhijit Patait
Sr. Manager,
GPU Multimedia SW
AGENDA
Overview GPU Video Encoding
NVIDIA Video Encoding Capabilities
— Kepler vs Maxwell GPU capabilities
— Roadmap
Software API
Performance & Quality
WHY GPU VIDEO ENCODING?
BENEFITS OF ENCODING ON GPU
Low power
— Fixed function hardware
— Reduced memory transfers
Low latency
High performance
Higher density
Scalability
Ease of Programming
— Linux, Windows, C/C++, Application portability
NVIDIA GPU VIDEO ENCODING CAPABILITIES
NVIDIA GPU ENCODING CAPABILITIES
Feature
Benefits
H.264 base, main, high profiles
Wide range of use-cases
High performance (Up to 16x HD)
“Blazing-speed” encoding
YUV 4:2:0 and 4:4:4 support
High quality encoding without chroma subsampling
QP maps
Customizable quality, region of interest encoding
MVC
Full resolution stereo encode
Up to 4096 × 4096 in HW
High resolution encode
API - NV Encode SDK & GRID SDK
Flexible, Win/Linux, DirectX/CUDA
Independent of CUDA
Use CUDA and encode simultaneously
VIDEO ENCODING — KEPLER VS. MAXWELL
Kepler (GK104, GK107, GK106, GK110,
GK208)
Maxwell (GM107)
Planar 4:4:4
Standard 4:4:4 and H.264 lossless encoding
~240 fps 2-pass encoding @ 720p
~500 fps 2-pass encoding @ 720p
GRID K340/K520, K1/K2, Quadro, Tesla
K10/K20
Current and future Maxwell GPU-boards
GeForce – 2 full-speed encode
sessions/GPU
GeForce – 2 full-speed encode
sessions/GPU
NV Encode SDK 1.0, 2.0, 3.0 (Now)
NV Encode SDK 4.0+ (May 2014)
GRID SDK 1.x, 2.2, 2.3 (Now)
GRID SDK 3.0+ (June 2014)
NVIDIA VIDEO ENCODING ROADMAP
Performance improvements
Quality improvements
— 4:4:4 & lossless encoding
— Rate control enhancements
— Adaptive quantization
— ROI, ME-only mode
New video standards
NVENC SOFTWARE APIS
USING NVENC
•
•
•
•
•
No capture
Transcoding
Archiving
Video editing
CUDA pre-process +
encoding
• Granular encoder
settings
• D3D, CUDA interop
Capture +
Encode
GRID SDK
NVENC SDK
Direct
Encode
• Capture + encode
• Optimized for lowlatency apps
• Capture + CUDA preprocess + encoding
• Encoder settings
optimized for streaming
• D3D, CUDA interop
DIRECT ENCODE (NVENC SDK)
Client application
Encoded
bitstream
Initialize, Configure, Encode
NVENC API
Configure HW
CUDA
Driver
NVENC
Driver
DirectX
Driver
HW Encode
NVENC firmware + hardware
CAPTURE AND ENCODE (GRID SDK)
Client application
Encoded
Bitstream
DX/OGL Present
NvFBC/NvIFR
Capture
NVENC
Driver
YUV
DirectX/OGL
Driver
Encode
NVENC Hardware
GPU 3D Engine
NVENC SDK
Available on NVIDIA developer zone
— https://developer.nvidia.com/nvidia-video-codec-sdk
— Current release 3.0
— Release 4.0 in May 2014 with Maxwell support
Interface header, documentation, sample application
— .dll/.so included in the driver
Unified API for Windows and Linux
Works on x86/x64
Various API’s, presets, rate control modes for
— Transcoding
— Video conferencing
— GTC Session S4654
NVENC SDK (CONTD.)
Advantages
— Flexibility
Dynamic resolution/bitrate change
CABAC vs CAVLC; low-level encoder settings, B-frames, sync vs async, custom QP
Linux, Windows, DirectX, CUDA, OGL (via CUDA)
Also works on GeForce hardware (2 sessions/GPU)
— Error concealment
Reference picture invalidation
Intra-refresh
— Quality
Two-pass modes for higher quality
Various presets with quality/performance trade-off
4:4:4 & lossless encoding (Maxwell only)
GRID SDK ENCODE
Available on NVIDIA developer zone
— https://developer.nvidia.com/grid-app-game-streaming
— Current release: 2.2
Interface header, documentation, sample apps
— .dll/.so included in the driver
Windows and Linux
Works on x86/x64
Various presets and API’s for
— Remote graphics (Cloud gaming, remote desktop, capture & stream)
Optimized for low latency
GRID SDK (CONTD.)
Advantages
— Simplicity
Very simple API; single function call for capture + H.264 encode
— Low-latency, high performance
Optimized API
— Error concealment
Reference picture invalidation
Intra-refresh
— Quality
Two-pass modes for higher quality
4:4:4 & lossless encoding (Maxwell only)
PERFORMANCE AND QUALITY
PERFORMANCE – 720P
NVENC Performance at 720p, Low-Latency HP preset
Rate control modes
231 fps
CBR_IFRAME_2PASS
504 fps
232 fps
2_PASS_FRAMESIZE_CAP
Kepler (GRID)
503 fps
Maxwell
232 fps
2_PASS_QUALITY
505 fps
100
200
300
400
500
600
720p Performance (fps)
Performance measured on GRID K520 with GRID SDK NVENC performance benchmarking application
PERFORMANCE – 1080P
NVENC Performance at 1080p, Low-Latency HP preset
Rate control modes
118 fps
CBR_IFRAME_2PASS
239 fps
118 fps
2_PASS_FRAMESIZE_CAP
240 fps
Kepler (GRID)
Maxwell
119 fps
2_PASS_QUALITY
238 fps
50
100
150
200
250
1080p Performance (fps)
Performance measured on GRID K520 with GRID SDK NVENC performance benchmarking application
ENCODING QUALITY VS X264 –
ASSUMPTIONS
Infinite GOP IPPP…
VBV buffer = bitrate/framerate
x264
— Zero latency
— CRF = 24
— Preset = faster
NVENC
— Preset = LOW_LATENCY_HQ
— RC = 2-pass-quality
NVENC/X264 QUALITY COMPARISON
Titan Fall 720p, 5 Mbps, Low-latency HQ
45
1.2
40
1.1
PSNR Y (dB)
35
1
0.9
25
SSIM Y
PSNR Y (dB)
30
20
0.8
SSIM Y
0.7
10
0.6
5
0
0.5
101
201
PSNR x264
SSIM NVENC
SSIM x264
15
1
PSNR NVENC
301
401
501
601
701
801
901
NVENC/X264 QUALITY COMPARISON
Bunny 1080p, 12 Mbps, Low-latency HQ
60
1.5
1.4
50
1.3
40
PSNR Y (dB)
30
1.1
1
20
0.9
SSIM Y
10
0.8
0
0.7
1
101
201
301
401
501
SSIM Y
PSNR Y (dB)
1.2
PSNR NVENC
PSNR x264
SSIM NVENC
SSIM x264
QUALITY COMPARISON – PSNR
PSNR Comparison - x264 vs NVENC
50.00 dB
45.00 dB
40.00 dB
PSNR Y (dB)
35.00 dB
30.00 dB
25.00 dB
20.00 dB
15.00 dB
10.00 dB
5.00 dB
0.00 dB
-5.00 dB
PSNR NVENC
PSNR x264
Bunny
1080p
47.24 dB
43.71 dB
NFS Rivals
720p
34.05 dB
33.18 dB
NFS Rivals
1080p
35.51 dB
34.39 dB
Titan Fall
720p
30.58 dB
29.78 dB
Titan Fall
1080p
28.13 dB
30.63 dB
WoT - 3
1280 × 768
34.15 dB
33.41 dB
WoT - 12
1280 × 768
35.60 dB
34.72 dB
PSNR Difference
3.52 dB
0.87 dB
1.12 dB
0.80 dB
-2.50 dB
0.74 dB
0.87 dB
QUALITY COMPARISON – SSIM
SSIM Comparison - x264 vs NVENC
1.0000
0.8000
SSIM Y
0.6000
0.4000
0.2000
0.0000
-0.2000
SSIM NVENC
0.9874
NFS Rivals
720p
0.9217
SSIM x264
0.9808
0.9103
0.9269
0.8073
0.8567
0.8930
0.9027
0.01
0.01
0.01
0.03
-0.03
0.02
0.01
SSIM Difference
Bunny 1080p
NFS Rivals
1080p
0.9388
Titan Fall
720p
0.8350
Titan Fall
1080p
0.8309
WoT - 3
1280 × 768
0.9101
WoT - 12
1280 × 768
0.9169
QUESTIONS?