Side by side datacenter GPU specifications from the Flopper.io database.
| AWS Trainium3 | NVIDIA Groq 3 LPU | |
|---|---|---|
| FP32 TFLOPS | 183.0 | — |
| Memory | 144 GB | — |
| Memory Bandwidth | 4.9 TB/s | — |
| TDP | — | — |
| Precision | AWS Trainium3 peak TFLOPS (dense) | NVIDIA Groq 3 LPU peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | — |
| FP32 | 183.0 | — |
| TF32 | 671.0 | — |
| BF16 | 671.0 | — |
| FP16 | 671.0 | — |
| FP8 | 2517.0 | 1200.0 |
| FP6 | — | — |
| FP4 | 2517.0 | — |
| INT8 | — | — |
| AWS Trainium3 | NVIDIA Groq 3 LPU | |
|---|---|---|
| Architecture | NeuronCore-v4 | Groq LPU |
| Process Node | 3nm | — |
| Launch Year | 2025 | 2026 |
| Form Factor | — | — |
| Memory Type | HBM3e | On-die SRAM |
| Interface Width | — | — |
| Interconnect | NeuronLink-v4 | LPU C2C |
| TDP | — | — |
| FP32 Efficiency | — | — |