[CDNA 3]
AMD Instinct MI300X
The AMD challenger — 192 GB HBM3 in a single GPU.
AMD Instinct MI300X offers 192 GB HBM3 per GPU with strong FP8 throughput. Competitive with H200 on inference workloads where memory capacity matters; ROCm software maturity is improving rapidly.
VRAM
192 GB HBM3
FP8 TFLOPS
2,615
Mem BW
5.3 TB/s
TDP
750 W
Best for
- Memory-bound inference
- Long-context serving
- ROCm/PyTorch workloads
Benchmarks
| WORKLOAD | METRIC | VALUE |
|---|---|---|
| Llama-3 70B inference (FP8) | tokens/sec | ~2,200 |
| FP8 dense matmul | TFLOPS | 2,615 |
AMD Instinct MI300X availability
No SKUs currently indexed for this GPU.
Definition
What is Servers.Computer?
Servers.Computer is an AI compute routing and procurement layer that benchmarks, compares, and deploys GPU clusters (NVIDIA H100, H200, B200 and AMD MI300) across global cloud providers in real time.