Why Mistral Small 4 (119B MoE) pressures system RAM
Mistral Small 4 (119B MoE) is Mixture-of-Experts: inference activates 6.5B active/token, but VRAM/RAM must usually hold the full ~119B expert set for fast routing. At Q4 the weight slab is ~66.9GB before KV (~0.02GB at 8K) and ~8GB OS/runtime overhead — totaling ~74.9GB raw, rounded to a 96GB kit. Stretching toward the full 256K-token window multiplies KV far faster than weights; that is the usual “I bought enough RAM for the model but still OOM” failure on Mistral AI MoE pages.
What RAM kit to buy
Buy a matched dual-channel DDR5 kit at 96GB for Mistral Small 4 (119B MoE) (EXPO/XMP only if stable). Avoid single-stick installs — local inference is bandwidth-sensitive when layers spill to host memory. Pair with 4x RTX 3090 / 4090 (96GB VRAM) or Apple Mac Studio (128GB Unified Memory) when staying in the Multi-GPU Workstation / Mac Studio tier, and keep 20–30% RAM free for the OS + browser.
Workload notes
Mistral releases like Mistral Small 4 (119B MoE) are common in production vLLM; dual-channel bandwidth helps prompt throughput when CPU offload is in play. At 119B, Mistral Small 4 (119B MoE) sits in the large local-LLM band: Q4 on a strong GPU is realistic, FP16 usually is not on consumer cards. Release window noted as March 2026; always re-check the official source before buying hardware for a specific checkpoint.