Entity · Technology
AI Accelerator
A purpose-built computing architecture designed for machine learning workloads.
- Number of articles
- 2 articles
- First published
- First published
- Latest update
- Latest update
Latest coverage
Infrastructure
算力堆到一起之后，瓶颈变成了网络
机柜内每颗 GPU 有 1.8 TB/s 的互联带宽，跨出机柜的单端口速率是 800 Gb/s。从 scale-up、scale-out 到 scale-across，三层网络的落差决定了并行策略承受什么压力、以太网在 UEC 1.0 之后解决什么问题，以及跨数据中心训练能扩展到多远。
Infrastructure
What Actually Makes AI Expensive?
From final training runs to HBM, advanced packaging, networking, depreciation, utilisation, inference scheduling and customer total cost of ownership, this article maps where AI spending goes and why a lower unit price does not guarantee lower total expenditure.