苹果自研集群跑千亿模型
推荐指数 82.0 NO. 010 · 2026.10.17
发布2026/10/16Score217Comments101
为什么值得看
Apple 用 MLX 框架在四台 Mac Studio 集群上本地运行近万亿参数的 Muse Spark Open 模型，通过 RDMA 实现分布式推理，生成速度超 36 tokens/s。这对需要数据隐私合规、不想上云的企业 AI 团队有直接参考价值。
编辑判断
之前本地跑大模型的主流方案是 llama.cpp 配合消费级 GPU，但跨卡通信一直是瓶颈，多机几乎不可行。Apple 这次用 RDMA over Thunderbolt 绕过了传统 PCIe 限制，让 Mac Studio 的共享内存架构真正发挥集群效应。
这对 NVIDIA 的 DGX 工作站产品线是个微妙信号：如果 Apple 能把 RDMA 生态做成熟，AI 实验室的采购预算可能会分流。正在做私有化部署的团队可以关注 MLX 的分布式 API 稳定性，目前还是 0.x 版本，生产环境建议等两个 release 周期再跟进。
社区反馈
意见分歧 71 条评论
核心争论：本地隐私合规 vs 云API成本效益，苹果硬件溢价是否值得
Not exactly your average "desk", but still neat. In this post, we show how to run the latest frontier-class model — Muse Spark Open, with nearly a trillion parameters — right from your desk, using the latest version of MLX, released today. On a cluster of four Mac Studios, Muse Spark Open generate
Yes, I'm very glad they're doing this, but as of 2026, it's a bit academic. Pricing for the 512GB Mac Studio hasn't even been announced yet, but the 256GB version starts at $9499. So when you start talking about four of those puppies, it's substantial. Someone else will do the math on how many years
There was the M3 Ultra 512GB Mac Studio, though.