高通推150W推理卡叫板NVIDIA
推荐指数 52.0 NO. 023 · 2026.11.05
发布2026/11/04Score72Comments35
简介
高通Cloud AI 100 Ultra是一款150W功耗、128GB HBM的推理加速卡，带宽548GB/s，主打大模型推理场景。这是高通罕见直接对标NVIDIA L4/L40的云端AI芯片，对成本敏感的推理服务商可能是新选项。
编辑判断
高通这张卡的关键数字是150W对350W——NVIDIA L40S的功耗。对于推理即服务的创业公司，电费占TCO的30%以上，低功耗意味着更高的毛利空间。
但128GB显存配548GB/s带宽是个奇怪组合，H100的带宽是3TB/s级别，这意味着高通可能在牺牲单卡吞吐换显存容量，更适合批处理大模型而非实时高并发场景。
已经在用AWS Inferentia或Google TPU做推理的团队可以对比测试，但新入场的AI infra创业者建议观望软件生态成熟度，高通的历史短板不在硬件在驱动。
社区反馈
意见分歧 35 条评论
核心争论：高算力低带宽设计是否适合大模型推理，还是内存带宽才是瓶颈
Curious how the memory bandwidth actually holds up under sustained inference loads compared to the spec sheet numbers.
Aren't these specs already outdated? Seems meh. I would think players would like to bring something new to the table instead of bringing what was okay two years ago or so. Give me 1TB RAM and 5TB/s or something that exciting.
5 year old GPUs are still at crazy prices. Plenty of businesses want in local inference and are priced out.