Google unveils two AI chip configurations aimed at privacy and memory workloads

AI · First reported 9 Nov, 15:20 · Updated 15:25

Image: Wikimedia Commons · NASA Johnson Space Center · Public domain

Google announced TPU 8tx and TPU 8i Max for its eighth-generation accelerator lineup, targeting confidential processing and memory-intensive AI tasks. Google also described performance claims and software support, but did not provide pricing or availability dates.

Google says TPU 8tx uses hardware-level confidential computing to isolate customer data and model parameters during processing. It combines TPU 8t silicon, fifth-generation AMD EPYC processors and Google’s Boardfly network.
Google claims TPU 8tx can cut latency by up to half compared with TPU 8t, with almost no performance penalty from the added protection.
TPU 8i Max combines two TPU 8i chips in one package; Google says it doubles computing power, on-chip SRAM and high-bandwidth memory throughput versus TPU 8i.
Google says its open-source TPU Sync software can reduce some data-transfer delays and response startup times by more than 90%.
Google did not disclose prices or availability dates. The reported performance comparisons are against earlier Google hardware, not Nvidia’s latest products.

Summary written by Lux AI from 4 articles

Key facts

New configurations: TPU 8tx and TPU 8i Max
TPU 8tx latency claim: Up to half the latency of TPU 8t
TPU Sync claim: More than 90% reduction in certain data-transfer delays and response startup times
Customer count: Google says it serves nearly 100 TPU customers
Pricing and availability: Not disclosed