Event ·Official first-hand

AI与HPC交叉的模板计算研究

TinyTC和OpenAI Triton等张量编译器最初为AI工作负载开发，但相同的平铺和内存抽象可应用于实现科学和工业应用的高效高阶模板。研究团队实现了8阶25点声学模板，在GPGPUs上比较TinyTC与PyTorch/Triton实现。在Battlemage B580上，TinyTC达到15.6 Gpts/s，而PyTorch/Triton为13.5 Gpts/s。零初始化运行因硬件内存压缩可达35.8 Gpts/s。

Current read

TinyTC团队展示了AI编译器如何优化HPC模板计算，在Intel GPU上性能领先，PyTorch/Triton则提供跨厂商开发基准。

8 sources73° AI heatLast updated 2026/9/9 15:59:22

Evidence

8 sources

Primary sourcearXiv: OpenAI

2026/9/9 15:59:22View source

Related source 1OpenAI

2026/9/8 17:23:27View source

Related source 2elvis

2026/9/8 17:31:10View source

Related source 3宝玉

2026/9/8 17:54:23View source

Related source 4Simon Willison

2026/9/8 23:58:33View source

Related source 5SuperTechFans

2026/9/9 22:57:08View source

Related source 6Decoder

2026/9/10 10:31:20View source

Related source 7Thomas Wolf

2026/9/10 10:38:45View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story