Compute Modules
Subcategory covering compute modules responsible for data processing, control, and execution of robotic software.
Compute Modules is a subcategory of hardware components that provide processing power for robotic systems. It encompasses onboard computers, single-board computers (SBCs), AI accelerators, embedded processors, GPU/NPU compute modules, and other units responsible for processing sensor data and executing control logic. These modules form the foundation of modern autonomous, humanoid, and perception-capable robots.
1X NEO Cortex
Onboard AI compute module (the "brain") of the 1X NEO humanoid, based on the NVIDIA Jetson Thor SoC. Delivers up to 2070 FP4 TFLOPS for on-device model inference.
Architect Labs Redwood
The Architect Labs Redwood compute module — the base variant of the vendor's onboard AI computer for robots. Specification not publicly verified.
Architect Labs Redwood Nano
The compact Architect Labs Redwood Nano compute module — a smaller, power-efficient variant of the Redwood family for the vendor's robots. Specification not publicly verified.
D-Robotics RDK
D-Robotics' family of developer kits/boards (RDK X3, X5, S100) for robotics and edge AI, featuring proprietary AI acceleration engines (Sunrise) and 200+ open-source algorithms.
Intel NUC 13 Pro
Compact mini-PC Intel Core i7-1360P (12C/16T), 2× TB4, 2× 2.5 GbE, up to 64 GB DDR4. Intel-discontinued, continued by ASUS.
NVIDIA Blackwell Ultra
NVIDIA Blackwell Ultra (GB300) GPU optimized for AI reasoning: 1.5x larger HBM3E memory and 1.5x more dense FP4 FLOPS than standard Blackwell; the core of the GB300 NVL72 system.
NVIDIA DGX Spark
Compact NVIDIA desktop AI supercomputer (1.2 kg) with the GB10 Grace Blackwell Superchip; 128 GB unified memory and up to 1 PFLOP FP4; runs models up to 200B parameters.
NVIDIA GB200 NVL4
NVIDIA compute module (superchip) combining 2 Grace CPUs and 4 Blackwell GPUs on a single server board, with up to 1.3 TB coherent memory, for converged HPC and AI workloads.
NVIDIA GB200 NVL72
Rack-scale, liquid-cooled computing system built around 72 NVIDIA Blackwell (B200) GPUs and 36 NVIDIA Grace CPUs, designed for training and inference of large AI models.
NVIDIA GB300 NVL72
Rack-scale, liquid-cooled computing system built around 72 NVIDIA Blackwell Ultra (B300) GPUs and 36 NVIDIA Grace CPUs, designed for large AI model inference and training.
NVIDIA H100
A data-center AI accelerator based on the NVIDIA Hopper architecture (2022): 80 GB HBM3, 700 W, 4th-generation Tensor Cores with FP8 and Transformer Engine. The standard GPU for LLM training and hyperscale AI inference.
NVIDIA H200
A data-center AI accelerator based on the NVIDIA Hopper architecture (2023): the first GPU with HBM3e memory — 141 GB at 4.8 TB/s. Same compute as the H100 but nearly double the memory capacity and bandwidth for training and inference of large language models.
NVIDIA IGX Thor
NVIDIA industrial edge AI platform (launched October 28, 2025, GTC Washington) — Blackwell GPU, up to 5,581 FP4 TFLOPS, iGPU + optional dGPU, 400 GbE, NVIDIA Halos safety, 10-year support. For robotics and medical.
NVIDIA Jetson
NVIDIA's leading embedded AI platform and module family for edge AI and robotics: compact edge computers with GPUs and the JetPack SDK; Thor, Orin, Xavier and Nano tiers.
NVIDIA Jetson AGX Thor
NVIDIA flagship compute module for humanoids and physical AI - 2070 TFLOPS FP4, Blackwell GPU, 128 GB LPDDR5X.
NVIDIA Jetson AGX Xavier
Edge AI module, 32 TOPS with Volta GPU and 8-core Carmel ARM. End-of-sale — superseded by AGX Orin.
NVIDIA Jetson Orin NX 16GB
NVIDIA Jetson Orin NX 16GB compute module based on the Ampere architecture, used as the processing unit in the Unitree G1 robot for perception, AI, and edge computing tasks.
NVIDIA Jetson Orin Nano 2
Entry-level edge AI SO-DIMM module: 78 TOPS, 8-core Arm Cortex-A78, 8 GB LPDDR5X. 2x performance vs Orin Nano Super. Available 1H 2027.
NVIDIA Jetson Orin Nano 8GB
Compact SO-DIMM module, 40 TOPS (67 TOPS Super mode) with Ampere GPU, for small robots and drones.
NVIDIA Jetson Orin Nano Super
Edge-AI module/dev kit, 67 TOPS (INT8): Ampere GPU 1024 CUDA, 6-core Cortex-A78AE, 8 GB LPDDR5 102 GB/s, 7–25 W, from $249.
NVIDIA RTX PRO 4500 Blackwell Server Edition
NVIDIA professional Blackwell GPU in a server variant: 32 GB GDDR7 memory with ECC, 896 GB/s, 5th-gen Tensor Cores and 4th-gen RT Cores, FP4 support.
NVIDIA Rubin
NVIDIA's upcoming GPU (the generation after Blackwell), announced at Computex 2024 and expected in Q3 2026: TSMC 3 nm, HBM4 memory, 50 sparse PFLOPS in FP4; paired with the Vera CPU.
NVIDIA Rubin Ultra
An improved Rubin architecture expected in 2027: 100 sparse PFLOPS in FP4 (double Rubin), functionally two Rubin cores connected together.
OpenAI Jalapeño
OpenAI's first in-house AI accelerator (2026), designed from scratch as an ASIC for large-language-model inference. Co-developed with Broadcom, with production planned from late 2026 as part of a multi-generation compute platform.
Qualcomm Robotics RB5 Platform
5G-ready robotics platform on QRB5165 SoC (Snapdragon 865), 15 TOPS DSP, 8 GB LPDDR5, Wi-Fi 6.
Raspberry Pi 5
BCM2712 Cortex-A76 single-board computer, PCIe 2.0, dual 4K HDMI, RTC, Wi-Fi 5 — Raspberry Pi flagship SBC.
Raspberry Pi Compute Module 4
BCM2711 SO-DIMM compute module, 1–8 GB LPDDR4, 0–32 GB eMMC, PCIe Gen2, optional Wi-Fi/BT.
Rockchip RK3566
Power-efficient Rockchip SoC (RK35xx series) with a quad-core Arm Cortex-A55 CPU, Mali-G52 2EE GPU and a built-in NPU; used in edge, IoT and robotics (including the Microduck robot).
Rockchip RK3588
Rockchip's flagship 8 nm SoC: octa-core CPU (4× Cortex-A76 + 4× Cortex-A55), Mali-G610 MP4 GPU and a 6 TOPS NPU. Used as onboard compute in robots.
Tesla FSD Computer
Tesla-designed AI inference computer originally used in Tesla vehicles and adapted as the onboard compute module for the Optimus humanoid robot.
comma 4
comma.ai ADAS device mounted behind the mirror — an integrated onboard computer with cameras running openpilot (lane centering and adaptive cruise control). Successor to the comma 3X.