IT之家 · 智能时代T2·68 pts
- Each node packs eight MI355X accelerators, dual EPYC 9575F CPUs, 3TB DDR5 and 63.36TB SSD, offered via Dell and Supermicro
- GLM-5.3 runs locally at 98% of frontier-model performance, defaulting to on-prem inference over cloud APIs
- Supports up to 85 developers per node with 50 concurrent, aimed at private enterprise coding assistants