WR 20251017

Webto Wang -

题图：首都机场 T3 航站楼从天上看还是挺不错的（里面真的走断腿）

【AI】NVIDIA DGX Spark + Apple Mac Studio = 4x Faster LLM Inference with EXO 1.0

EXO Labs wired a 256GB M3 Ultra Mac Studio up to an NVIDIA DGX Spark and got a 2.8x performance boost serving Llama-3.1 8B (FP16) with an 8,192 token prompt. Their detailed explanation taught me a lot about LLM performance ... EXO noted that the Spark has 100 TFLOPS but only 273GB/s of memory bandwidth, making it a better fit for prefill. The M3 Ultra has 26 TFLOPS but 819GB/s of memory bandwidth, making it ideal for the decode phase.

因为 LLM 的运行逻辑，[英伟达新出的 DGX](# NVIDIA DGX Spark 搭载与 MediaTek 共同设计的 GB10 超级芯片) 和 Mac Studio 形成了奇妙的互补关系，建议直接读原文，很简单好读。

【苹果】The Just Plain M5 Chip Launches in Three Updated Products: 14-Inch MacBook Pro, iPad Pro (Both Sizes), and Some Sort of Headset Thingamajig Called Vision Pro

The new Dual Knit Band ($99 on its own) looks like a hybrid of the more attractive Solo Knit Band (which did not have a strap that went over the top of your head) and the Dual Loop Band (which did have an over-the-head strap, but which looked somewhat orthopedic). It’s a tacit acknowledgement that physical comfort has been a real problem for many people who’ve tried Vision Pro. (Me, personally, I find using it with the Solo Knit Band comfortable for as long as I care to use it — which is typically just 2–3 hours, tops.)

苹果突然（好吧也不太突然，毕竟 Mark Gurman 早就爆料了）发布了 M5 系列新品，对我来说最有兴趣的是新 Vision Pro 的这个头带，等能买了我去买来试试。另外，新的 iPad Pro 因为用了 N1 芯片，也有 Thread 支持了。

【书】父亲的解放日志

出门的路上读完了这本书，不是很长，但是真的很好看，强烈推荐。

其他一些乱七八糟的