Ampere GPU, Arm Cortex-A78AE CPU, NVDLA and PVA accelerators on a production SoM
Compute & Control · Embedded AI system-on-module
Jetson AGX Orin 64GB
A high-end Jetson module whose headline TOPS value is an INT8 sparse throughput claim and must be paired with its power mode and workload precision.
NVIDIA Jetson AGX Orin 64GB · Also: P3701-0005
Reported values, conditions attached.
Evidence state describes the claim—not the prestige of the company or document.
| Claim | Reported value | Condition / boundary | State | Source |
|---|---|---|---|---|
| AI performanceai_performance | 275 INT8 sparse TOPS | Vendor theoretical peak with sparsity; not an application benchmark. | M · Manufacturer claimed | NVIDIA Jetson AGX Orin Developer Kit Reviewer’s Guide ↗Jetson AGX Orin specifications, table 4 Checked 2026-08-09 |
| CPUcpu | 12-core Arm Cortex-A78AE | No additional condition stated in the selected source. | M · Manufacturer claimed | NVIDIA Jetson AGX Orin Developer Kit Reviewer’s Guide ↗Jetson AGX Orin specifications, table 4 Checked 2026-08-09 |
| Memorymemory | 64 GB LPDDR5, 204.8 GB/s | Production 64 GB module. | M · Manufacturer claimed | NVIDIA Jetson AGX Orin Developer Kit Reviewer’s Guide ↗Jetson AGX Orin specifications, table 4 Checked 2026-08-09 |
| Module power modespower | 15–60 W | Configured module power modes; carrier and peripherals add system power. | M · Manufacturer claimed | NVIDIA Jetson AGX Orin Developer Kit Reviewer’s Guide ↗Jetson AGX Orin specifications, table 4 Checked 2026-08-09 |
| Camera I/Ocamera_io | 16 MIPI CSI-2 lanes / 16 virtual channels | Module interface capability; carrier routing determines exposed ports. | M · Manufacturer claimed | NVIDIA Jetson AGX Orin Developer Kit Reviewer’s Guide ↗Jetson AGX Orin specifications, table 4 Checked 2026-08-09 |
| Primary software stacksoftware | NVIDIA JetPack / CUDA / Isaac | Exact compatibility depends on JetPack and Jetson Linux release. | M · Manufacturer claimed | NVIDIA Jetson modules lineup ↗Jetson AGX Orin Series Checked 2026-08-09 |
Known limits
What this record does not prove.
- TOPS is precision- and sparsity-specific.
- Module power is not total robot-compute-system power.
- Carrier-board I/O exposure varies.