Why GLM-5 (744B MoE) pressures system RAM
GLM-5 (744B MoE) is Mixture-of-Experts: inference activates 40B active/token, but VRAM/RAM must usually hold the full ~744B expert set for fast routing. At Q4 the weight slab is ~418.5GB before KV (~0.1GB at 8K) and ~12GB OS/runtime overhead — totaling ~430.6GB raw, rounded to a 512GB 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 Zhipu GLM MoE pages.
What RAM kit to buy
Shop 512GB-class capacity for GLM-5 (744B MoE): workstation DDR5 RDIMM/LRDIMM or multi-kit desktop builds, not a single gamer 2×16GB stick. Use our 128GB+ price hubs and RAM Finder; confirm ECC needs for your board. GPU path: Apple Mac Studio (192GB Unified Memory) or Institutional Node (8x H100 / A100) (430.5GB VRAM class) if you want weights on-device instead of system-RAM offload.
Workload notes
For Zhipu GLM's GLM-5 (744B MoE), treat published parameter counts as the weight floor and add OS + KV + runtime overhead before shopping kits. At 744B total parameters this is frontier-scale — expect multi-GPU or heavy CPU offload even in Q4; the 512GB kit is a host-memory floor, not a promise of interactive tokens/s. Release window noted as February 2026; always re-check the official source before buying hardware for a specific checkpoint.