🔌
New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
// RUN — FITS
Can the Instinct MI455X run Phi-3.5-MoE?
Yes — here's how fast and up to what context, with real measured numbers.
Phi-3.5-MoE on the Instinct MI455X
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 837k | 101.5-124.1 t/s | ~0.2-0.3 (slow) |
| Q8 | yes | 1023k | 203.1-248.2 t/s | ~0.4-0.5 (slow) |
| Q4 | yes | 1104k | 362.7-443.3 t/s | ~0.7-0.9 (slow) |
Estimate (memory-bound). Beyond "max context" the KV cache spills to system RAM offload (depends on your system RAM) and speed drops to the "beyond" figure.
FAQ
Can the Instinct MI455X run Phi-3.5-MoE?▶
Yes. At Q4 it generates ~362.7-443.3 tok/s and fits up to 1104k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~0.7-0.9 tok/s.
How much context fits?▶
At Q4, up to about 1104k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.