BestHub
Sign in
Architects' Tech Alliance
Architects' Tech Alliance
Dec 28, 2025 · Artificial Intelligence

Google’s TPU v7: How 1.5 & 2.6 Optical Modules per Chip Power AI Supercomputers

The article explains how Google’s TPU v7 supercomputer uses a simple yet powerful networking scheme—1.5 optical modules per TPU for intra‑rack communication and an additional 2.6 modules per TPU for inter‑rack high‑speed links—enabling massive AI model training with balanced cost and performance.

AI supercomputingGoogleLarge-Scale Training
0 likes · 13 min read
Google’s TPU v7: How 1.5 & 2.6 Optical Modules per Chip Power AI Supercomputers
BirdNest Tech Talk
BirdNest Tech Talk
Nov 20, 2024 · Industry Insights

Inside xAI’s 100k‑GPU Colossus: Supermicro Liquid‑Cooled Racks Explained

The article provides a detailed, step‑by‑step tour of xAI’s Colossus supercomputer— a $‑billion AI cluster built in 122 days with 100,000 NVIDIA H100 GPUs—covering Supermicro liquid‑cooled 4U racks, cooling distribution units, power and water infrastructure, storage nodes, CPU servers, 400 GbE networking, and the operational challenges of scaling such a massive system.

AI supercomputingColossusData center architecture
0 likes · 16 min read
Inside xAI’s 100k‑GPU Colossus: Supermicro Liquid‑Cooled Racks Explained
Architects' Tech Alliance
Architects' Tech Alliance
May 1, 2024 · Industry Insights

How NVIDIA’s Blackwell Platform Redefines AI Supercomputing Networks

The article examines NVIDIA’s Blackwell platform network architecture, detailing the fifth‑generation NVLink, sixth‑generation PCIe, 800 Gb/s InfiniBand and Ethernet adapters, the DGX B200 and GB200 configurations, new IB and Ethernet switches, and the implications of increased optical module demands for large‑scale AI clusters.

AI supercomputingBlackwellDGX
0 likes · 10 min read
How NVIDIA’s Blackwell Platform Redefines AI Supercomputing Networks