边缘设备Edge devices
NVIDIA Jetson AGX ThorNVIDIA Jetson AGX Thor
机器人/边缘平台:128 GB 统一内存但只有 273 GB/s 带宽,容量大、带宽窄,适合 7B–30B 级模型 4-bit 常驻。40–130 W 的功耗窗口意味着持续推理会降频;2070 TFLOPS 是 FP4 稀疏口径,不能与数据中心卡的 FP16 数字直接比。A robotics and edge platform: 128 GB unified memory but only 273 GB/s bandwidth, so capacity is large while bandwidth is narrow. It suits persistent 4-bit 7B–30B models. The 40–130 W window means sustained inference throttles, and 2070 TFLOPS is FP4 sparse, not comparable with data-centre FP16 figures.
关键参数Key specs
- AI 算力 / AI performance
- 2070 TFLOPS FP4 (sparse)
- 内存 / memory
- 128 GB 256-bit LPDDR5X
- 带宽 / bandwidth
- 273 GB/s
- CPU
- 14-core Arm Neoverse-V3AE
- 功耗 / power
- 40–130 W
- 网络 / networking
- 1x 5GbE RJ45 + 1x QSFP28 (4x 25GbE)
- 存储 / storage
- 1 TB NVMe (developer kit)
官方原文怎么说What the source says
Jetson Thor series modules deliver up to 2070 FP4 TFLOPS of AI compute and 128 GB of memory
Memory | 128 GB 256-bit LPDDR5X | 273 GB/s
Power | 40 W–130 W
资料与许可Source and licence
- 资料时间Source checked
- 2026-11-17
- 许可 / 类型Licence / kind
- official-spec-page
- 许可说明Licence evidence
- https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/
- 类目Category
- 边缘设备Edge devices