AMD · CDNA 3 · 2024
AMD Instinct MI325X
256 GB of HBM3e, the largest memory pool on any single accelerator you can rent.
256 GB
HBM3e
6 TB/s
Mem bandwidth
1,307 TF
FP16 dense
2,614 TF
FP8 dense
1000 W
TDP
$2.40 /hr
From · Vast.ai
Same CDNA 3 compute as MI300X with a memory upgrade. One MI325X holds a 4-bit ~400B-class model or a 70B in FP16 with enormous KV cache for long-context serving.
Where to rent a MI325X
On-demand prices per GPU-hour; marketplace rates refresh daily, list prices reviewed 2026-09. Reserved and spot run 30-70% lower. Full breakdown with the used market and rent-vs-buy math: MI325X pricing.
Specifications in context
Bars scaled against the best value in the whole catalog (B200 / MI325X era).
| Architecture | CDNA 3 (2024) |
|---|---|
| Memory | 256 GB HBM3e, 6 TB/s |
| Interconnect | Infinity Fabric · 896 GB/s |
| Form factor | OAM |
| Partitioning | No MIG (time-slicing / vGPU only) |
What fits on one MI325X
Weights + ~2 GB runtime overhead against 230 GB usable VRAM. Longer contexts and bigger batches need more, check the VRAM calculator.
| Model | Params | Highest precision that fits |
|---|---|---|
| Llama 3.2 1B | 1.24B | FP16 |
| Llama 3.2 3B | 3.21B | FP16 |
| Llama 3.1 8B | 8.03B | FP16 |
| Llama 3.3 70B | 70.6B | FP16 |
| Qwen2.5 7B | 7.62B | FP16 |
| Qwen2.5 14B | 14.8B | FP16 |
| Qwen2.5 32B | 32.8B | FP16 |
| Qwen2.5 Coder 32B | 32.8B | FP16 |
| Qwen2.5 72B | 72.7B | FP16 |
| QwQ 32B (reasoning) | 32.8B | FP16 |
| Mistral 7B | 7.25B | FP16 |
| Mixtral 8x7B | 46.7B | FP16 |
| Mixtral 8x22B | 140.6B | 8-bit |
| Gemma 2 9B | 9.24B | FP16 |
| Gemma 2 27B | 27.2B | FP16 |
| Phi-4 14B | 14.7B | FP16 |
| gpt-oss-20b | 20.9B | FP16 |
| gpt-oss-120b | 116.8B | 8-bit |
Best for
- Very long context serving
- Largest single-GPU models
- KV-cache-heavy multi-tenant inference