gpt-oss infrastructure

gpt-oss Hardware Requirements: RAM and VRAM for 20B and 120B

OpenAI's gpt-oss family has two MoE models branded 20B and 120B. The official cards list about 21B total / 3.6B active parameters for gpt-oss-20b and 117B total / 5.1B active for gpt-oss-120b. Hardware sizing must account for the resident total-weight footprint, not only the active experts.

Computed RAM / VRAM floors by model

Weight memory is parameter count × bits per weight ÷ 8. The final column is deliberately more conservative: it runs the same 4-bit estimate through AIG's default 20% runtime reserve and 2 GB context/cache reserve. Neither value includes workload-specific concurrency, unusually long context, offload buffers, or multimodal activations.

CALCULATED. The planning target is weight floor + 20% runtime reserve + 2 GB context reserve. Validate the actual artifact and runtime before buying hardware.

What changes for gpt-oss

These are mixture-of-experts models. Active parameter count describes compute per token, not the full resident model footprint. Use total parameters or the actual distributed artifact when planning memory, then validate with the target runtime's MXFP4 implementation.

Weight fit is not the same as deployment fit

• Check the exact GGUF, safetensors, AWQ, GPTQ or runtime-specific artifact you will load.
• Reserve memory for KV cache, runtime allocations, the operating system and concurrent requests.
• Unified-memory Macs share the same pool with the OS and applications; discrete-GPU systems still need system RAM.
• For MoE models, active parameters reduce compute per token but do not automatically remove inactive expert weights from memory.