英文原文(节选)

If the target market is real time robotics, the Jetson Thor system … seems over-provisioned with expensive memory.

中文翻译

如果目标市场是实时机器人,Jetson Thor 配备 128 GB 内存、但带宽只有 273 GB/s,这样的昂贵内存配置显得偏高。模型需要以每秒数十帧运行,因此最多只能使用大约 10 GB 的权重。

如果模型采用较宽的混合专家架构(MoE),或者大型战略规划模型以较小时间片在更长时间尺度上运行,更多内存仍然可能被利用;更多内存也总能让开发更轻松。不过,对成本优化的系统而言,应当可以用更少的内存完成任务。

原始 X:John Carmack on X: "If the target market is real time robotics, the Jetson Thor system, with 128 GB of memory, but only 273 GB/s of bandwidth, seems over-provisioned with expensive memory. You want the model evaluating at tens of fps, so it can't use more than 10 GB of weights at most. More memory… / X