AI News AI资讯 2h ago Updated 1h ago 更新于 1小时前 49

[AINews] Hot Chips: OpenAI's Jalapeño, Cerebras CS-5, Groq 3 LPX, Apple M6 【AI新闻】Hot Chips大会:OpenAI的Jalapeño、Cerebras CS-5、Groq 3 LPX、Apple M6

OpenAI unveiled Jalapeño, a custom inference chip that outperforms NVIDIA GB200/GB300 systems with 1.5–1.9× better work per watt and 1.7–3.6× lower end-to-end latency at a 700W TDP (running at ≤550W in tests) Jalapeño achieves these gains without relying on aggressive prefill/decode disaggregation or speculative decoding, suggesting a more balanced inference architecture that reduces the traditional throughput/latency tradeoff OpenAI used GPT-Astra + Codex to write and optimize low-level kernels OpenAI在Hot Chips 37发布自研推理芯片Jalapeño,性能超越NVIDIA Blackwell/Rubin系列,每瓦性能提升1.5-1.9倍,端到端延迟降低1.7-3.6倍 Jalapeño为700W芯片,实际测试功耗≤550W,年底前开始部署,Gen 2已开发中,Gen 3已启动 GPT-Astra + Codex辅助优化底层内核,注意力机制和MoE块实现比人类专家代码快1.5-1.8倍 前沿实验室推理经济正从依赖NVIDIA转向自研芯片,但TSMC/CoWoS封装产能仍是瓶颈 Agent Harness质量与模型选择同等重要,Microsoft AutoSaddler在G

75
Hot 热度
65
Quality 质量
70
Impact 影响力

Analysis 深度分析

TL;DR

  • OpenAI unveiled Jalapeño, a custom inference chip that outperforms NVIDIA GB200/GB300 systems with 1.5–1.9× better work per watt and 1.7–3.6× lower end-to-end latency at a 700W TDP (running at ≤550W in tests)
  • Jalapeño achieves these gains without relying on aggressive prefill/decode disaggregation or speculative decoding, suggesting a more balanced inference architecture that reduces the traditional throughput/latency tradeoff
  • OpenAI used GPT-Astra + Codex to write and optimize low-level kernels, achieving 1.5–1.8× faster attention and MoE block implementations than human-expert-written code, signaling a new model-assisted compiler optimization paradigm
  • Microsoft's AutoSaddler framework treats agent harnesses as code, patching prompts, tool configs, and control logic offline to deliver gains of +9.0 on GAIA2, +9.6 on SWE-Bench Pro, and +10.0 on Terminal-Bench 2.0
  • Long-horizon software engineering remains critically unsolved: SWE Refactor Bench showed only a 5.4% survival rate across 520 runs of whole-repository migration tasks (C→Rust, Maven→Gradle, POSIX→WebAssembly)

Why It Matters

OpenAI's move to custom silicon marks a strategic inflection point where frontier labs are no longer strictly downstream of NVIDIA for inference economics, potentially reshaping the AI hardware supply chain. Simultaneously, the emerging evidence that harness quality can outweigh model choice in agent benchmarks forces practitioners to invest equally in evaluation infrastructure and system design rather than chasing marginal model improvements.

Technical Details

  • Jalapeño chip specs: Custom inference ASIC rated at 700W TDP (operating at ≤550W), targeting Blackwell/Rubin-class systems; deployment in OpenAI's infrastructure begins by year-end with Gen 2 in development and Gen 3 underway
  • Benchmark performance: 1.5–1.9× more work per watt at peak throughput, 1.7–3.6× lower end-to-end latency, and 2.1–4.1× higher performance on highly interactive workloads compared to NVIDIA GB200/GB300 systems
  • Model-assisted kernel optimization: GPT-Astra + Codex were used to generate and optimize low-level kernels for three open-weight models, producing attention and MoE block implementations 1.5–1.8× faster than existing human-expert code within approximately two months
  • AutoSaddler harness optimization: Microsoft-led approach that treats agent harnesses as patchable code, using failure traces to offline-optimize prompts, tool configurations, and control logic
  • SWE Refactor Bench methodology: Measures whole-repository migration tasks across real projects (SQLite, zlib, libsodium) with a three-stage evaluation pipeline, revealing a 5.4% survival rate across 520 runs

Industry Insight

Frontier labs will increasingly pursue vertical integration into custom silicon to control inference economics, but foundry and packaging capacity (particularly TSMC CoWoS) remain hard bottlenecks that could delay widespread adoption beyond OpenAI's infrastructure. The AutoSaddler and harness variance findings suggest evaluation engineering should be treated as a first-class research discipline—poor harness design can invalidate benchmark comparisons and mislead model selection decisions. Practitioners should prioritize investing in robust agent scaffolding and standardized "Harness Card" disclosures before allocating additional resources to model upgrades.

TL;DR

  • OpenAI在Hot Chips 37发布自研推理芯片Jalapeño,性能超越NVIDIA Blackwell/Rubin系列,每瓦性能提升1.5-1.9倍,端到端延迟降低1.7-3.6倍
  • Jalapeño为700W芯片,实际测试功耗≤550W,年底前开始部署,Gen 2已开发中,Gen 3已启动
  • GPT-Astra + Codex辅助优化底层内核,注意力机制和MoE块实现比人类专家代码快1.5-1.8倍
  • 前沿实验室推理经济正从依赖NVIDIA转向自研芯片,但TSMC/CoWoS封装产能仍是瓶颈
  • Agent Harness质量与模型选择同等重要,Microsoft AutoSaddler在GAIA2/SWE-Bench Pro/Terminal-Bench 2.0上分别提升9.0/9.6/10.0分
  • 长期软件工程仍极困难:SWE Refactor Bench在SQLite/zlib/libsodium等真实项目迁移任务中仅5.4%存活率

为什么值得看

本文揭示了AI基础设施从"依赖GPU供应商"向"自研专用芯片"转型的关键拐点,对云厂商和AI实验室的硬件战略具有直接参考价值。同时,模型辅助编译器/内核优化的实践表明AI工程化正在形成新的技术范式,值得关注。

技术解析

  • Jalapeño芯片架构:非ASIC,而是完整推理芯片,针对吞吐量和延迟的平衡设计,无需依赖激进prefill/decode分离或speculative decoding即可击败使用这些技术的NVIDIA系统
  • 性能基准:在真实模型工作负载下,Jalapeño相比GB200/GB300实现1.5-1.9×每瓦性能、1.7-3.6×更低端到端延迟、2.1-4.1×更高交互性能
  • 模型辅助开发:GPT-Astra + Codex在约两个月内将三个未计划支持的开源模型适配到Jalapeño,attention和MoE块实现比人类专家代码快1.5-1.8×
  • AutoSaddler框架:Microsoft-led项目将harness视为代码,通过失败trace离线修补prompt、tool配置和控制逻辑
  • SWE Refactor Bench:测量C→Rust、Maven→Gradle、POSIX→WebAssembly等全仓库迁移任务,520次运行仅28次存活(5.4%),13/20任务无人完成

行业启示

  • 硬件自主化趋势加速:OpenAI、Google等前沿实验室正通过自研芯片降低对NVIDIA的依赖,推理成本结构将发生根本性变化,建议关注TSMC产能分配和封装技术进展
  • AI辅助系统优化成为新范式:模型辅助编写/优化底层代码(编译器、内核)正在从实验走向生产,未来AI工程团队需同时掌握模型训练和系统编程能力
  • 评估体系需要重构:Harness variance研究显示换harness比换模型更能改变排名,行业需建立标准化披露(如Harness Card),同时长期软件工程能力仍是当前AI系统的明显短板

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Chip 芯片 GPU GPU Training 训练 Inference 推理 Product Launch 产品发布