苹果发布M6与M5 Ultra芯片，推动Mac本地AI推理
Massive move by Apple for local inference. They just launched M6 and M5 Ultra to…
Massive move by Apple for local inference. They just launched M6 and M5 Ultra to move more AI inference onto Mac desktops.
苹果在本地推理上迈出巨大一步。他们刚刚发布了M6和M5 Ultra，旨在将更多AI推理任务转移到Mac桌面端。
The M5 Ultra Mac Studio is configurable upto 256GB of unified memory and 512 GB coming in Oct.
M5 Ultra Mac Studio可配置高达256GB的统一内存，10月份将推出512GB版本。
M6 debuts in Mac mini as Apple's first 2nm chip, with a 12-core CPU, 12-core GPU and Dual 16-core Neural Engine.
M6在Mac mini中首次亮相，是苹果首款2纳米芯片，配备12核CPU、12核GPU和双16核神经引擎。
The smaller process increases transistor density, while GPU Neural Accelerators and 170GB/s memory bandwidth give local models more compute and faster access to data.
更小的制程增加了晶体管密度，而GPU神经加速器和170GB/s的内存带宽为本地模型提供了更强的计算能力和更快的数据访问速度。
M5 Ultra takes the approach further by joining two dual-die M5 Max chips through UltraFusion into Apple's first quad-die SoC.
M5 Ultra通过UltraFusion将两个双芯片M5 Max连接成苹果首款四芯片SoC，进一步推进了这一方法。
It scales to 36 CPU cores, 80 GPU cores, 512GB unified memory and 1.2TB/s bandwidth, 50% above M3 Ultra.
它可扩展至36个CPU核心、80个GPU核心、512GB统一内存和1.2TB/s带宽，比M3 Ultra高出50%。
The memory capacity changes AI use because Apple says it can hold LLMs with hundreds of billions of parameters entirely on-device.
内存容量的提升改变了AI的使用方式，因为苹果表示，它可以在设备上完全容纳拥有数千亿参数的LLM。
Apple also supports Thunderbolt 5 clustering with RDMA, so it will let multiple Mac Studio systems pool memory for distributed inference.
苹果还支持带RDMA的Thunderbolt 5集群，这将允许多个Mac Studio系统汇集内存进行分布式推理。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力