Primary-source skepticism done right — the benchmark chip has no SKU, wattage, or price in AMD's own 132-page deck — plus a concrete forecast that Nvidia splits its datacenter CPU line by workload (training headnode vs many-core agent sandbox).
Vera is designed to be a GPU headnode, RL training CPU due to its high single-thread performance. I expect Nvidia to bifurcate the lineup. 1 for GPU training/RL training (high single-thread performance) and 1 for agent sandbox (lots of cores, high DRAM/core) with Rosa
Primary-source skepticism done right — the benchmark chip has no SKU, wattage, or price in AMD's own 132-page deck — plus a concrete forecast that Nvidia splits its datacenter CPU line by workload (training headnode vs many-core agent sandbox).
Tweets, videos, filings, articles, and source pages tied to this read.