Built for On-Device LLMs
Runs LLMs up to 14B parameters locally, unleashing full Physical AI performance.
Rhino Boost Q86
An embedded AI accelerator card powered by Qualcomm QCS8625 (8 Gen 3), built for edge AI and smart IoT with high compute and low power.
QCS862560 TOPS Dense AI16GB LPDDR5XNPU power draw 7WLinux EmbeddedUbuntu22.04
Highlights
4nm Architecture
Massive AI compute at just 7W.
Flagship Video Codec
Full-frame-rate 8K@60 FPS decoding and efficient 8K@30 FPS encoding.
Zero-Code Model Adaptation
450+ AI models, each with complete code, ready to run out of the box.
Specifications
Chip process
Qualcomm QCS8625 4nm
CPU
1*3.4Ghz+5*3.2Ghz+2*2.3GHz 64-bit
GPU
Adreno750 5.2TFLOPS-FP16 OpenGL Es 3.2，Vulkan 1.3 openCL 3.0
NPU
15TOPS（FP16)）、60TOPS（int8）Dense AI、120TOPS（int8）Sparse AI
VPU
Video decode up to 4K@240 fps/8K@60 fps、Video encode up to 4k@120 fps/8K@30 fps
Memory
12GB LPDDR5X（Std）、24GB LPDDR5X（Opt）、16GB LPDDR5X（Opt）
Storage
128GB UFS3.1
OS
Linux （Linux Embedded）
Business Data Interface
1*2.5Gpbs RJ45、1*PCle/2 lane
Device Management
1*USB3.1、1*UART、1*USB2.0、1*12C
Environment
Working temperature：-20-45 ℃ Storage temperature：-40-75 ℃
Device power consumption
Typical AI Inference Power（NPU）：7W
Radiator
Fin Radiator*1（Opt.Acc）、Fan *1（5V Power supply，Opt.Acc）
Size
47.0*64.0*3.25mm
Certification
CE & FCC
Product Life Cycle
2032
Build with Rhino Boost Q86
Get Rhino Boost Q86 documentation, evaluation kits, pricing and support for your Physical AI project.