AI推理算力AI inference compute

运行模型生成输出的算力

每条都摘自下方编号的来源原文;只写事实,不写预测或建议。

技术简介

  • AI推理是模型根据新数据生成预测或输出的过程。4
    Inferencing is the process of an artificial intelligence model generating educated predictions or outputs from new data.

已有应用

  • Cerebras高速推理将用于Bilt Neighborhood OS平台和Bilt Concierge AI助手,尚非已部署声明。2
    Cerebras high-speed inference will power the new Bilt Neighborhood OS platform and Bilt Concierge AI assistant to deliver faster, seamless interactions between the platform and the eight million consumers and 50,000 businesses it serves.
  • OpenAI 已在阿比林采用超过10万颗英伟达 Blackwell GPU 的 OCI Supercluster 训练并运行 GPT-6 Astra。3
    In Abilene, OpenAI has trained and served GPT-6 Astra on an OCI Supercluster with more than 100,000 NVIDIA Blackwell GPUs.

为何此时

  • 多步骤智能体工作流每增加一次模型调用都会增加延迟,可能拖慢工作流。1
    Each additional AI model call adds latency, which can slow agentic workflows that reason and act across multiple steps.

主要参与者

  • Cerebras计划将Wafer Scale Engine 5处理器用于CS-5;处理器已完成启动并进入系统测试。2
    Cerebras’ next-generation processor, the Wafer Scale Engine 5 to be deployed in CS-5, is back from the fab, completed bring up and is in system testing.
  • Cerebras 的 Wafer-Scale Engine 将计算与内存集成在单个晶圆级处理器上,用于高速AI推理。1
    Designed for high-speed AI inference, the Cerebras Wafer-Scale Engine integrates compute and memory on a single wafer-scale processor.

未定事项

  • Cerebras 推理服务计划在2027年登陆OCI,公告未给出具体日期。1
    Cerebras Inference on OCI is planned to be available in calendar year 2027, giving OCI customers another option for AI applications where speed is critical.

相关卡片

来源

  1. oracle.com · Cerebras and Oracle Team Up to Bring Cerebras AI Inference to Oracle Cloud Infrastructure · 2026-10-27
  2. SEC · Document
  3. 英伟达 · NVIDIA and Oracle Expand NVIDIA Vera Rubin to Multi-Gigawatt Scale and Bring NVIDIA Vera CPU to Oracle Cloud Infrastructure · 2026-10-27
  4. aws.amazon.com · What is Inferencing? - AI Inference Explained - AWS