Reply to thread

Message

[QUOTE="Munyaradzi Mafaro, post: 79515, member: 636"]
Apple's tactic of linking Macs together for AI work is facing a big price problem. They can pool the memory from several Mac mini or Apple Studio computers using Thunderbolt 5, creating a huge shared resource that beats out setups using NVIDIA's hardware in cost for some machine learning jobs. This works because of Apple's unified memory design, where a single M4 Pro system already offers more RAM than a high-end graphics card.

The company is pushing this advantage hard, with new software support in its latest macOS to better handle these linked clusters. The fast connection lets machines directly access each other's memory, making the whole group act like one powerful computer. A demo using four loaned Mac Studios showed it could be cheaper than building a similar system with NVIDIA's competing DGX Spark units.

But that cost edge is about to get slammed. Apple got good prices on memory chips through long-term deals with suppliers. Those contracts are expiring soon, with manufacturers ready to charge them way more. This means the upcoming M5-based machines and their cluster potential will likely see a serious price jump. The money they currently save against competitors could shrink to almost nothing once their component costs surge.
[/QUOTE]

Name