Edge work comes down to a balance between power draw, compute, and heat. This collection covers edge inference chips, development boards, and complete units, from NVIDIA to domestic vendors. Work out your model memory footprint and latency budget on paper before choosing a board; it saves a lot of trial and error.
Tool list
7 toolsNVIDIA Jetson HotNew
Edge computing platform for physical AI and robotics sharing the Blackwell GPU architecture and JetPack software stack; the T5000 module delivers 2,070 FP4 sparse TFLOPS with 128GB of memory
Hailo-10H HotNew
Among the first edge accelerators with generative AI support, rated at 40 TOPS INT4 with 2.5W typical power; the Raspberry Pi AI HAT+ 2 is built on this chip
Rockchip RK3588 HotNew
Flagship embedded SoC from Rockchip, an 8nm octa-core part with a 6 TOPS in-house NPU, 8K encode and decode, and four independent display outputs
Doubao Phone Assistant New
System-level phone agent from ByteDance's Doubao team, debuting on the nubia NaviX Ultra with a fingerprint AI key, Seed full-duplex voice that can be interrupted, and both GUI and interface paths across apps
Sophgo BM1688 New
High-throughput edge TPU from Sophgo, rated at 32 TOPS INT8 and up to 64 TOPS INT4, able to run 7B-class models locally within an 18W envelope
NVIDIA DGX Spark New
NVIDIA personal AI supercomputer built on the GB10 Grace Blackwell Superchip with unified memory; the 64GB model runs up to 100-billion-parameter models on device and two units pool memory to 128GB
Intel GPU D270 New
First product in Intel data centre inference GPU line named at the 2026 OCP Global Summit. Xe3P architecture, 32 Xe cores, up to 480GB LPDDR5X memory, and a single card serving a 300B parameter model.
No matching tools found. Try another keyword.