AMD MI450X
AMD's 288 GB CDNA 5 GPU for AI inference and training in Helios rack systems.
- Arch
- CDNA 5
- Memory
- 288 GB HBM4
- Bandwidth
- 23.30 TB/s
- Released
- Q4 2026
At a glance
We don't currently track any MI450X offers. See alternative GPUs.
Technical Specifications
AMD MI450X · Per GPU
| Compute · dense | |
|---|---|
| FP4 | 40,265 TFLOPS |
| FP8 | 20,133 TFLOPS |
| FP16 / BF16 | 5,033 TFLOPS 10,066 with sparsity |
| INT8 | 5,033 TOPS 10,066 with sparsity |
| FP32 | 315 TFLOPS |
| Precision support | FP4FP6FP8FP16BF16FP32INT8 |
| Memory | |
|---|---|
| Capacity | 288 GB HBM4 |
| Bandwidth | 23,300 GB/s |
| ECC | Yes |
| Silicon | |
|---|---|
| Architecture | CDNA 5 |
| Process | TSMC 2nm / 3nm |
| Transistors | 320 billion |
| Compute units | 256 |
| Fabric and host | |
|---|---|
| GPU interconnect | Infinity Fabric 3,600 GB/s |
| Power | |
|---|---|
| Cooling | Liquid |
Source: official AMD MI450X datasheet.
Frequently Asked Questions
Why choose the MI450X?
288 GB of HBM4 per GPU at the same 23.3 TB/s of bandwidth and the same compute as the MI455X, in the same 72-GPU Helios rack with UALink scale-up networking. A fit for inference and training where the model, context and batch fit in 288 GB per GPU.
When is the MI450X not a good fit?
Launched in November 2026 and shipping only in Helios racks, so cloud availability is minimal. If a model's weights plus KV cache need more than 288 GB per GPU, the MI455X carries 432 GB on the same platform. For AMD capacity in clouds today, the MI355X offers the same 288 GB on CDNA 4. ROCm is narrower than CUDA, so check software compatibility before committing.
Alternatives to AMD MI450X
Last updated