T-Chip Group
Intelligent hardware website
Innovate intelligent hardware and solutions.
Server Business Website
Server products and efficient solutions.
Consumer Business Website
StationPC NAS products and services.

CORE-8550JD4 AI SoM

48 TOPS Computing Power NPU

· High-Performance AI Processor QCS8550  

· Deployment of Large AI Models  

· Supports Ray Tracing Technology  

· 8K Video Encoding and Decoding  

· Qualcomm Cognitive ISP with 36MP Triple Camera  

· 4nm Low-Power Design for Hardware Iteration  

· Compact Form Factor with Gold-Finger Package  

· Comprehensive Development Toolchain

Manual Wiki Download

Qwen

Llama

DeepSeek

Gemma

CORE-8550JD4
CORE-1126BJD4 AI SoM

Powered by the Qualcomm QCS8550 AI processor with a 48 TOPS NPU, it supports mainstream AI models and features the Adreno 740 GPU for ray tracing and 8K video. The cognitive ISP supports up to a 100MP single or triple 36MP camera, all in a compact design with a 260-pin gold finger interface.

AI Processor QCS8550

Features an octa-core Kryo CPU in the QCS8550, powered by the ARM architecture with a clock speed of up to 3.36 GHz, delivering robust support for high-performance computing and multitasking.

Private Deployment

Supports private deployment of large-scale Transformer models (including Qwen, Llama, Gemma, DeepSeek-R1, and Phi series) and mainstream deep learning frameworks like TensorFlow, PyTorch, and ONNX.

Ray Tracing Technology

Equipped with the Adreno 740 GPU, it delivers full ray tracing support and is compatible with OpenGL ES 3.2, Vulkan 1.2, and OpenCL 3.0. Combined with Adreno Neural Processing, it significantly boosts both graphics and AI computing performance.

8K Video Codec

Supports advanced video processing with 8K@60fps/4K@240fps decoding and 8K@30fps/4K@120fps encoding, delivering exceptionally detailed and smooth ultra-high-definition visuals.

36MP Triple Camera

Equipped with the Qualcomm Spectra Cognitive ISP featuring triple 18-bit ISPs and 2x DPHY 1.2/CPHY 2.0 interfaces, it supports flexible camera configurations up to a triple 36MP, dual 64MP+36MP, or a single 108MP setup for multi-stream vision systems.

4nm Low-Power Design

Built on a 4nm process for high efficiency and stable performance, it is ideal for always-on devices. With hardware interface compatibility to NVIDIA Jetson Orin NX/Nano, it enables rapid hardware platform iteration.

Gold-Finger Package

Features a compact 69.6mm × 45mm form factor with a standard 260-pin gold finger SODIMM interface, weighing just 16g for easy installation and broad compatibility across diverse device scenarios.

Development Toolchain

A comprehensive suite is provided—including an AI model optimization tool, video processing SDK, and reference design with full documentation—to enable efficient secondary development and accelerate the creation of proprietary products.
48 TOPS NPU Computing Power

Integrated 48 TOPS NPU supports mixed-precision computation (INT4 to FP16), enabling powerful edge AI capabilities such as intelligent data processing, voice recognition, and image analysis for most terminal devices.

Interfaces
Specifications

CORE-8550JD4

Basic Specifications

SOC

Qualcomm QCS8550

CPU

Qualcomm Kryo® CPU: Octa-core 64-bit (1×GoldPlus@3.2GHz + (2+2)Gold@2.8GHz + 3×Silver@2.0GHz), 4nm advanced process, maximum frequency up to 3.36GHz

GPU

Adreno 740 GPU: Supports ray tracing technology, OpenGL ES 3.2, Vulkan 1.2, full-profile OpenCL 3.0, and Adreno NN Direct

ISP

Equipped with Qualcomm Spectra Cognitive ISP (Image Signal Processor), featuring three 18-bit 36MP ISPs

NPU

Dual eNPU V3: Equipped with 4 HVX (Hexagon Vector Extensions), 1 HMX (Hexagon Matrix Extension); Computing power up to 48 TOPS (INT8), 12 TOPS (FP16)

Codec

Video decoding: 8K@60fps/4K@240fps H.265/H.264/VP9/AV1 Video encoding: 8K@30fps/4K@120fps H.265/H.264 Supports concurrent 4K@60fps decoding and 4K@60fps encoding for wireless display scenarios

RAM

16GB LPDDR5x

Storage

256GB UFS4.0

Power

5V (voltage tolerance ±5%)

Power consumption

Normal: 3W(5V/600mA), Max: 10W(5V/2000mA), Min(Sleep): 0.4W(5V/80mA)

OS

Ububtu

Software support

Supports on-premises deployment of large-scale parameter models based on the Transformer architecture, such as large language models (LLMs) including the Deepseek-R1 series, Gemma series, Llama series, Qwen series, Phi series, etc. Supports the QNN AI inference framework, as well as various deep learning frameworks including TensorFlow, TensorFlow Lite, PyTorch, ONNX, etc.

Interface type

SODIMM (260PIN, 0.5mm pitch)

Size

69.6mm × 45.0mm × 4.3mm

Weight

≈16g

Environment

Operating Temperature: -20℃～60℃, Storage Temperature: -20℃～70℃, Operating Humidity: 10%～90%RH (No condensation)

Interface Specifications

Network

Supports 2×Gigabit Ethernet (via PCIe expansion; onboard Ethernet PHY chip on core board); 5G/4G expandable

Video input

2 × MIPI D/C PHY (4 lanes DPHY or 3 trios CPHY) + 2 × MIPI D/C PHY (2 lanes DPHY or 1 trio CPHY) Supports 36MP triple camera / 64MP + 36MP dual camera / 108MP single camera

Video output

1 × HDMI2.0 (4K@60Hz), 1 × DP1.4 (via Type-C output)

Audio

2 × I2S (2 data channels per channel; supports 4-channel TX/RX, TDM/PCM modes, 48kHz sampling rate)

PCIe

1 × PCIe3.0 (2lanes), 1 × PCIe4.0 (2lanes)

USB

1 × USB3.1 Gen2 DRD with DP, 2 × USB3.0 HOST (via PCIe expansion)

Others

3 × CCI I2C, 2 × I3C/I2C, 2 × SPI, 4 × UART, 1 × ADC, 1 × SDMMC, GPIOs
Customization

Firefly team, with over 20 years of experience in product design, research and development, and production, provides you with services such as hardware, software, complete machine customization, and OEM server.

Hardware Solution

Software Solution

Turnkey Solution