“Efficiency Unleashed: The Next-gen NXP i.MX 95 Applications Processor for Embedded Vision,” a Presentation from NXP Semiconductors
•
0 likes•31 views
For the full video of this presentation, please visit: https://www.edge-ai-vision.com/2024/06/efficiency-unleashed-the-next-gen-nxp-i-mx-95-applications-processor-for-embedded-vision-a-presentation-from-nxp-semiconductors/ James Prior, Senior Product Manager at NXP Semiconductors, presents the “Efficiency Unleashed: The Next-gen NXP i.MX 95 Applications Processor for Embedded Vision” tutorial at the May 2024 Embedded Vision Summit. Machine vision is the most obvious way to help humans live better, enabling hundreds of applications spanning security, monitoring, inspection and more. Modern edge processors need private on-device and scalable hybrid machine learning capabilities to offer enough longevity to stay relevant in industrial and commercial IoT markets. In this talk, Prior presents the upcoming i.MX 95 family of applications processors. The i.MX 95 features a new, self-developed neural processing unit from NXP—the eIQ Neutron NPU. Designed to scale from today’s conventional neural networks to tomorrow’s transformer-based models, the eIQ Neutron NPU scalable architecture delivers edge AI capabilities at high efficiency with award-winning tools, combined with chip-level security and privacy features. The i.MX 95 applications processor family features powerful processing and vision capabilities combined with safety, security and expandable high-speed interfaces.
Recommended
More Related Content
Similar to “Efficiency Unleashed: The Next-gen NXP i.MX 95 Applications Processor for Embedded Vision,” a Presentation from NXP Semiconductors
Similar to “Efficiency Unleashed: The Next-gen NXP i.MX 95 Applications Processor for Embedded Vision,” a Presentation from NXP Semiconductors (20)
More from Edge AI and Vision Alliance
More from Edge AI and Vision Alliance (20)
Recently uploaded
Recently uploaded (20)
“Efficiency Unleashed: The Next-gen NXP i.MX 95 Applications Processor for Embedded Vision,” a Presentation from NXP Semiconductors
- 1. Efficiency Unleashed: The Next-Gen NXP i.MX 95 Applications Processor for Embedded Vision James Prior Senior Product Manager BL SCE, NXP Semiconductors
- 2. • Enable real-time analytics and actuation Not hampered by network latency • On-board machine learning Precise and fast detection, classification, adaptation • Reduce data center traffic Only process and store relevant data • Reduce network cost Shield cloud from large part of raw data • Safeguard privacy Transmit semantic rather than raw data • Increase security Resilient to offline conditions Edge AI Needs More Efficient ML/AI © 2024 NXP Semiconductors, Inc. 2
- 3. AI/ML Enabled Applications with eIQ® Neutron NPU 3 © 2024 NXP Semiconductors, Inc.
- 4. First i.MX Application Processor with eIQ® Neutron NPU 4 © 2024 NXP Semiconductors, Inc. Machine Vision Capability High-Performance Compute Real-Time Compute Rich High Speed I/O Display & Multimedia Capability Advanced Security Enhanced Reliability Functional Safety
- 5. • Acting on market feedback to deliver scalability & consistency • Optimized for Edge AI inference workload performance & power efficiency • Supports major NN structures (CNN, MLP, RNN, LSTM, TCN, and more) • LLM support enablement: LLAMA v2 & Blenderbot • Flexibility to tune solution to meet customer needs • Hardware scales from 32 Ops/cycle to 2k Ops/cycle • Software support is unified and consistent NXP eIQ® Neutron NPU Acceleration for Edge AI & Industry 4.0 Workloads 5 © 2024 NXP Semiconductors, Inc.
- 6. NXP eIQ® Neutron NPU Performance 6 © 2024 NXP Semiconductors, Inc. Benchmark / Performance (Inferences/Sec) i.MX 8M Plus VeriSilicon VIP8000 2.3 TOPS1 NXP i.MX 93 ARM ETHOS U65 0.5 TOPS2 NXP i.MX 95 eIQ Neutron N3-1024S 2 TOPS 3 MobileNet-v1 368 236 11124 MobileNet-v2 332 282 721 Inception v3 30 30 101 ResNet50-v1 60 20 125 SSD-MobileNet_v2 137 76 350 Performance relative to Neutron N3-1024S 0.48 0.35 1 Efficiency relative to Neutron N3-1024S 0.41 0.71 1 1: From NXP i.MX8M Plus (measured) 2: From benchmark document by Arm shared with NXP. 3: Projected based on cycle-accurate simulations. 4. Measured on i.MX 95 A1 pre-validation silicon. Preliminary results subject to change.
- 7. i.MX 95 Embedded Vision Function Mapping 7 © 2024 NXP Semiconductors, Inc.
- 8. Embedded Developers ML Experts Data Scientists Bring Your Own Model Bring Your Own Data eIQ® ML Software Development Environment Extensible workflow tools for dataset management, model training, optimization, deployment, and profiling Arm Cortex-A, GPU, DSP, NPU, eIQ Neutron NPU Arm® Cortex®-M, DSP, eIQ Neutron NPU Inference TFLite eIQ Toolkit TFLite for Microcontrollers AUDIO / VOICE VISION TIME SERIES STATE MONITOR PERSON DETECTION App Code Hub IDENTITY RECOGNITION and more …. Pipelines Inference engine support for compute engines (CPU, NPU, GPU, DSP) available on NXP MPU and MCU Command Line Graphical UI or
- 9. i.MX 95 Vision Processing Pipeline 9 © 2024 NXP Semiconductors, Inc. Up to Single 12MP high resolution camera - 4096x3072p30 / 3820x2160p60
- 10. Powerful Immersive Graphics 10 © 2024 NXP Semiconductors, Inc. Arm® Mali™ G310 3D GPU • Superb code portability & ease of use • Strong app ecosystem • Android Play store • Industry support by HMI engines • Valhall architecture with tile-based rendering • Modern API support including Vulkan, OpenGL • Area- & energy-efficiency focus • QoS support with finer granularity • Enhanced, separate 2D GPU enables safety-context streams • Supports (de)warp function • Higher pixel throughput @ 1300 mega-pixel/sec (vs. NXP i.MX 8M Plus)
- 11. 3D GPU Benchmarks 11 © 2024 NXP Semiconductors, Inc. Specification/Benchmark i.MX8 QXP/DXP GC7000 Lite i.MX8DM GC7000XSVX i.MX8QM 2xGC7000XSVX i.MX 95 Mali-G310 v2 GPU Clock Freq [core/shader; MHz] 700 / 850 800 / 1000 800 / 1000 1000 GFLOPS (FP32/FP16) 51.2 / 25.6 64 / 128 128 / 256 64/128 Triangle Fill-rate (Mega-triangles/sec) 234 267 267 + 267 (dual) 267 (bridged) 400 Pixel Fill rate (Giga-pixels/Sec) 1.4 1.6 3.2 4 GFXBench30 Manhattan offscreen1920x1080 (FPS) 5.4 10.7 16.3 20.3 GFXBench31 Manhattan31 offscreen (FPS) 3.2 6.9 12 13 GLBenchmark27 TREX Onscreen (FPS) 18 26 39 35 Antutu v8.4.x (Terracotta-Vulkan) score TBD 522 534 984 Antutu v8.4.x (Refinery- OpenGLES31 + AEP) score N/A 6849 7603 18261 Preliminary results obtained by NXP internal team; subject to change. Platform configuration and BSP/software version may offer different results
- 12. New NXP Image Signal Processor (ISP) Optimized for Machine Vision 12 © 2024 NXP Semiconductors, Inc. • Capture RGB + IR images simultaneously for better low-light image capture, small camera modules, and enhanced facial recognition • Up to 12MP high-resolution camera • 20 bit-per-pixel pipeline supports up to 8 sensors w/aggregate 500 MPixel/s throughput • HDR combining of 2 exposures • Support color, monochrome & RGB-IR sensors • Advance de-noising and edge enhancement for low-light conditions • Single camera memory-to-memory or streaming processing Image Sensor ISP & ISI Image Sensor Image Sensor Image Sensor(s) 2D GPU SRAM DRAM
- 13. How To Get NXP i.MX 95 Evaluation Kits? Apply for Toradex Verdin EVK early access now 13 © 2024 NXP Semiconductors, Inc. Empowering developers to kickstart projects and smoothly transition from proof-of-concept to large-scale production Machine Vision Capabilities Quad-Lane MIPI CSI (up to 2x w/o DSI) High-Performance Compute Featuring a 16GB LPDDR5 RAM Energy Flex Architecture Power Monitoring of SoM and Carrier Board Control & Optimization of SoC voltage rails Rich set of Interfaces 10-GbE port | 2x GbE ports with TSN capability USB 3.0 Type C | 2x USB 2.0 Type A | 2x CAN-FD Enhanced Reliability and Functional Safety FuSa-dedicated EEPROM | OctalSPI Flash for fast boot Expansibility and Connectivity mPCIe and M.2 Key E slots for Cellular Modems, WiFi-6/6E etc. Micro-SD Card slot | 8x ADC Inputs | 4x I2C via headers Display and Multimedia Quad-Lane MIPI DSI | Octal-Lane LVDS Audio Codec on-board | PDM mics
- 14. • NXP i.MX 95 Family of Applications Processors • For Auto Digital Cluster; Industry 4.0 & HMI ; Commercial IoT Smart Devices & Gateways; & more • New NXP eIQ Neutron NPU • Premium Arm® Mali™ Graphics • 4K Vision Processing • Functional Safety • ISO 26262 (ASIL-B) / IEC-61508 (SIL 2) • Early Access Program Open Now • Sampling 2Q24 • Production 2H25 First i.MX Application Processor with eIQ® Neutron NPU 14 © 2024 NXP Semiconductors, Inc.
- 15. Where to find more info on NXP i.MX 95 & AI/ML 2024 Embedded Vision Summit Additional Info: i.MX 95 In Action For eCockpit with Generative AI Visit The NXP booth #503 15 © 2024 NXP Semiconductors, Inc. NXP i.MX 95 Product Page https://www.nxp.com/imx95 Introducing NXP eIQ® Neutron NPU Blog https://www.nxp.com/company/blog/intr oducing-the-nxp-eiq-neutron-neural- processing-unit-npu:BL-INTRODUCING- THE-NXP-EIQ-NPU Toradex Verdin i.MX 95 Evaluation Kit https://www.toradex.com/computer-on- modules/verdin-arm-family/nxp-imx95- evaluation-kit