imec 发布 3D HBM-on-GPU 热学研究，破解 AI 算力架构散热难题。在2025 年 IEEE 国际电子器件会议（IEDM）上，比利时微电子研究中心（imec）发布了全球首项针对 3D 高带宽内存（HBM）与图形处理器（GPU）堆叠组件（HBM-on-GPU）的系统技术协同优化（STCO）热学研究成果。

An essential part of any good graphics card review is extensive benchmark testing, and TechRadar has always taken this process very seriously. But just because I know the ins-and-outs of my graphics ...

当英伟达 GB300 的峰值算力突破 15000TFlops 时，整个行业突然意识到：制约 AI 算力增长的不再是芯片本身，而是封装里的 “毛细血管”。 从 V100 到 Rubin，GPU 封装功耗从 300W 飙到 2300W，2.5D 与 3D 封装的混战、液冷技术的迭代、纳米级互联的突破，正在上演一场 ...

A little over three years ago, Intel lured AMD's former Radeon Technologies Group (RTG) boss Raja Koduri to come over and lead its own efforts in graphics, at which time Intel also announced it ...

When NVIDIA first announced that it would create a workstation-class remote virtualization system that enabled full GPU acceleration with latency-free streaming I was more than a little skeptical. Not ...