Shulou(Shulou.com)12/24 Report--

Thanks to CTOnews.com netizens Xiao Zhan cut, OC_Formula clue delivery! CTOnews.com, December 7 (Xinhua)-- in addition to launching a new large language model Gemini 1.0, Google also announced an enhanced version of the Tensor Processing Unit (TPU) chip, the latest model is TPU v5p.

Google already uses TPU chips in products and services such as YouTube, Gmail, Google Maps, Google Play and Android, and the latest version is Google's most powerful, scalable and flexible AI accelerator chip to date.

CTOnews.com note: in terms of performance, Google TPU v5p can achieve 459 teraFLOPS; with bfloat16 precision and 918 teraOPS with Int8 precision.

Google TPU v5p is equipped with 95GB HBM3 memory, memory bandwidth is 2.76TB / sec, each Pod has up to 8960 acceleration cores, and self-developed 600GB/sec chip interconnection master can train AI models faster and more accurately.

Compared to TPU v4, the newly released TPU v5p has twice the FLOPS (floating point operations per second) and three times the high memory bandwidth.

Google said that with BF16 precision, GPT3, which trains OpenAI with 175 billion parameters, is 90% faster than the previous TPU v4 AI accelerator chip; it can reach 180% with Int8 precision.

The operating cost of each TPU v5p accelerator is $4.20 per hour, which is a little more expensive than TPU v4 ($3.22 per hour) and TPU v5e ($1.20 per hour).

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.