For physics simulation running at the edge, which NVIDIA production module has the highest FP32 TFLOPS?

Question

Rapidflare · Accepted Answer

The **Jetson AGX Thor T5000** has the highest FP32 performance of any NVIDIA production module: **8.064 TFLOPS**. It's a Blackwell-generation module aimed at physical AI and humanoid robotics, with 128GB of LPDDR5X memory at 273 GB/s and a 40–130 W power range.

**Top 5 NVIDIA production modules by FP32 TFLOPS (default mode)**

| Rank | Module | FP32 TFLOPS | GPU | Memory | Power range |
|---|---|---|---|---|---|
| 1 | Jetson AGX Thor T5000 | 8.064 | Blackwell, 2560 CUDA cores | 128 GB LPDDR5X | 40–130 W |
| 2 | Jetson AGX Orin 64GB | 5.3 | Ampere, 2048 CUDA cores | 64 GB LPDDR5 | 15–60 W |
| 3 | Jetson AGX Orin Industrial | 4.8 | Ampere, 2048 CUDA cores | 64 GB LPDDR5 | 15–75 W |
| 4 | Jetson T4000 | 4.7 | Blackwell, 1536 CUDA cores | 64 GB LPDDR5X | 40–70 W |
| 5 | Jetson AGX Orin 32GB | 3.8 | Ampere, 1792 CUDA cores | 32 GB LPDDR5 | 15–40 W |

**Notes:**
- **Super mode doesn't change the result.** Only four production modules list a Super-mode FP32 figure. The highest are the Jetson Orin NX 16GB and Orin NX 8GB, both at 2.4 TFLOPS, well below the top five above.
- **Three of the 20 production modules have no FP32 figure listed in our catalog.** They're not part of this ranking.
- **Jetson AGX Orin Industrial** may be worth a look if your edge site has harsh conditions. It's described as an industrial-grade, extended-temperature module that's pin-compatible with AGX Orin.

NVIDIA production module with the highest FP32 TFLOPS

Sources

Rank	Module	FP32 TFLOPS	GPU	Memory	Power range
1	Jetson AGX Thor T5000	8.064	Blackwell, 2560 CUDA cores	128 GB LPDDR5X	40–130 W
2	Jetson AGX Orin 64GB	5.3	Ampere, 2048 CUDA cores	64 GB LPDDR5	15–60 W
3	Jetson AGX Orin Industrial	4.8	Ampere, 2048 CUDA cores	64 GB LPDDR5	15–75 W
4	Jetson T4000	4.7	Blackwell, 1536 CUDA cores	64 GB LPDDR5X	40–70 W
5	Jetson AGX Orin 32GB	3.8	Ampere, 1792 CUDA cores	32 GB LPDDR5	15–40 W