Google Adds Confidential TPU 8tx and Higher-Bandwidth TPU 8i Max Google推出支持机密计算的TPU 8tx与更高带宽的TPU 8i Max
Google's TPU 8tx pairs AMD EPYC Turin hosts with confidential computing, while TPU 8i Max doubles memory bandwidth for inference, ServeTheHome reports. The variants address two deployment constraints: protecting sensitive workloads and moving data quickly enough to feed accelerator compute. The bandwidth increase is particularly relevant to memory-bound inference, although application throughput also depends on the workload. 据ServeTheHome报道,Google的TPU 8tx搭配AMD EPYC Turin主机并支持机密计算,TPU 8i Max则将推理内存带宽翻倍。两款产品分别针对敏感工作负载保护,以及数据传输能否跟上加速器计算速度这两项部署约束。带宽提升对受内存限制的推理尤其有意义,实际应用吞吐量仍取决于工作负载。