On all three numbers you asked about, the Jetson T4000 has more AI compute and memory bandwidth than the Jetson AGX Orin 64GB. In exchange, its power envelope starts and ends higher.
Jetson AGX Orin 64GB vs. Jetson T4000 nvidia.com +5nvidia.com ↗“up to 275 INT8 TOPS (sparse)”nvidia.com ↗“64GB LPDDR5 (204.8 GB/s)”nvidia.com ↗“15-60W”developer.nvidia.com ↗“performance.max_ai_tops=600 TOPS”developer.nvidia.com ↗“memory.memory_bandwidth_gb_per_s=273”developer.nvidia.com ↗“40-70W”
| Spec | Jetson AGX Orin 64GB | Jetson T4000 |
|---|---|---|
| Max AI performance (INT8, sparse) | 275 TOPS | 600 TOPS |
| Memory bandwidth | 204.8 GB/s (64 GB LPDDR5) | 273 GB/s (64 GB LPDDR5X) |
| Power envelope | 15–60 W | 40–70 W |
What this means for the move:
- AI TOPS: Both figures are INT8 with sparsity (Orin and Thor generations), so you can compare them directly.
- FP4: The T4000 also lists up to 1200 FP4 TFLOPS. That's a different precision from the INT8 TOPS above, so don't compare the two numbers. developer...developer.nvidia.com ↗“up to 1200 FP4 TFLOPS”
- Memory: Both modules have 64 GB. The T4000 uses faster LPDDR5X, which gives it the higher bandwidth. developer...developer.nvidia.com ↗“memory.memory_type=LPDDR5X”
- Power: The T4000's lowest power setting is 40 W. The AGX Orin 64GB can go as low as 15 W. If your platform has a tight power or thermal budget, check that it can supply at least 40 W and handle up to 70 W.
- Super mode: Neither module has a Super-mode rating, so the figures above are the only ones that apply.
For the full specs, see the Jetson AGX Orin technical brief and the Jetson Thor series modules datasheet.