Skip to content
Hot storiesContinuously updated

研究者提出EdgeDAE异构系统加速具身智能推理

1 reports1 reporting sourcesUpdated 9 hours ago

Start here

AI overview

2026年11月,研究人员提出面向实时具身智能的FPGA-GPU异构加速系统EdgeDAE。该系统采用异构协同架构,将视觉Transformer计算卸载至GPU,并在FPGA上加速扩散动作专家(DAE)推理。性能测试显示,相比边缘GPU基线,EdgeDAE将Octo-Small与Octo-Base的端到端推理延迟分别降低了52.5%和38.7%,吞吐量提升最高达2.10倍;此外,与RTX 4090相比,该系统实现了约17倍的能效提升。

AI-generated from reports · Updated 8 hours ago

Report timeline

Follow the reports to see different sides of the story.

Nov 4
  1. Hacker News · AI
    EdgeDAE:面向实时具身智能的扩散动作专家加速系统

    研究人员提出 FPGA-GPU 异构加速系统 EdgeDAE,将视觉 Transformer 卸载至 GPU,并在 FPGA 上加速扩散动作专家(DAE)推理。相比边缘 GPU 基线,该系统将 Octo-Small 与 Octo-Base 的端到端推理延迟分别降低 52.5% 和 38.7%,吞吐量提升高达 2.10x,且相比 RTX 4090 能效提升约 17x。

Story heat trend

Current heat 8·Comparable peak 10(Nov 4 22:00)·Comparable change, last 24 hours–

02.557.510Nov4Nov5Nov5Nov5

Trends compare only the same parties observed continuously, so the scope can be narrower than current heat counts. Move the pointer or tap the chart to see hourly heat; use the left and right arrow keys on a keyboard.