Tag
Artificial Analysis 的 AA-AgentPerf 基准通过回放真实编码 agent 轨迹测试推理硬件能效，结果显示 NVIDIA Vera Rubin 在几乎所有速度档位上均居前沿首位，领先于 DeepSeek V4 Flash/Pro、Kimi K3 和 GLM-5.2/5.3 等模型的测试配置。
微软宣布将于2027年起将Azure Maia系统扩展到多吉瓦级部署，Maia 200已投入生产并为Copilot等服务提供算力，Maia 300将于2027年推出，每兆瓦代币吞吐量比领先GPU系统高出40%。
The author argues that Tesla's iterative scaling of in-car inference computers was unprecedented, and that the current AI compute shortage is just the beginning — a far bigger shortage will emerge once autonomy becomes indispensable.
NVIDIA is integrating Groq technology into rack-scale products to enhance agentic AI performance by splitting workloads across specialized processors, improving token generation latency and overall responsiveness.
At Startup School 2026, Google's Chief Scientist Jeff Dean recounts the napkin math that led to Google's search index fitting in RAM and the TPU development, discussing inference hardware specialization and how startups can still compete.
Etched, an AI inference hardware startup, exited stealth after raising $800M and securing over $1B in customer contracts. Their first racks ship this summer, claiming state-of-the-art throughput, latency, and power efficiency.