🤖 AI Summary
Google has launched its new TPU 8tx, which combines TPU 8t silicon with 5th Gen AMD Epyc "Turin" processors to provide confidential AI inference capabilities. This move marks a strategic pivot for Google, which has opted for AMD’s Epyc CPUs over its own Axion chip due to the latter's limitations in supporting audited confidential environments, which currently only operate on x86 architecture. The TPU 8tx positions itself as a significant advancement in secure computing, as it serves regulated enterprises and governments needing robust data protection during model training and inference while maintaining performance stability.
The TPU 8tx boasts 12.6 petaflops of FP4 compute and 216GB of high-bandwidth memory, with improvements in latency owing to the utilization of a lower-latency networking solution. AMD’s Secure Encrypted Virtualization (SEV) extends protection to the TPU environment by ensuring that sensitive data remains encrypted and isolated, even during active use. This collaboration not only broadens Google's TPU offerings to external data centers, potentially disrupting the market landscape, but it also emphasizes the importance of trusted hardware in confidential computing. This partnership additionally signals Google’s reliance on AMD for critical components in their AI infrastructure, suggesting deeper future collaborations between the two tech giants as they navigate the growing demand for secure AI solutions.
Loading comments...
login to comment
loading comments...
no comments yet