Graphical use of Huawei 310 and design details

On October 10, Huawei was on the entire joint meeting 2018, and the Huawei AI strategy and full stack solution were announced. At the same time, Huawei released a self-study cloud AI chip "ASCEND" series, based on the Da Vinci architecture, the first batch launched the 7nm liter 910 and 12 nm slit 310. Among them, Zeng Teng 910 is the current single-chip computing density chip, calculating far super ceremonial and anti-Wenda. Limited 910 Semi-precision (FP16) Capability is 256TFLOPS, doubled than NVIDIA's Tesla V100, integer accuracy (INT8) 512TOPS, support 128 channel full HD video decoding (H.264 / 265), maximum power consumption 350W. The maximum power consumption of the slimming 310 chip is only 8W, and the main Ai chip is calculated on the extremely efficient calculation of the low power consumption. Half-precision (FP16) calculation capability 8TFLOPS, integer accuracy (INT8) 16tops, support 16 channel full HD video decoding (H.264 / 265). These two AI chips and large-scale distributed training systems will be launched in the second quarter of next year. A few days ago, Huawei terminal mobile phone product line president He Gang released a long picture, detailing the use of the rise 310 and design details. He Gang said that it is an important support for Huawei's comprehensive AI strategy for Huawei's full-time AI solution. In design, breaking through the power consumption, intelligent force of the artificial smart chip design, and realizes the significant increase in energy efficiency. The future will provide a new solution for safe, automatic driving, cloud business and IT intelligence, intelligent manufacturing, robot, portable machine, smartphone, smart watch and other application scenarios. Shengdu 310 uses Huawei Self-roller-Renfi architecture, using Huawei self-research efficient flexible CISC instruction set, each AI core can complete 4096 MAC computing within 1 cycle, integrated with tensor, vector, scalar, etc. The arithmetic unit supports a variety of mixed precision calculations, support training and reasoning data accuracy operations. The unified architecture can adapt a variety of scenarios, power consumption ranges from dozens millions to hundreds of tiles, elastic multi-core stacks, can provide optimal energy percentage in a variety of scenes. Read the full article, the original title: A picture detailed explanation of Huawei to rise 310 AI chip: Self-rustic Da Vinci architecture force and energy efficiency Article Source: [Micro Signal: EETOP-1, WeChat public number: EETOP] Welcome to add attention! Please indicate the source of the article.

Our other product:

Professional lightning protection system for Tower	Professional FM Radio Station Equipment Package	FMUSER FU-1000C 1000W FM Transmitter With USB Audio Input Interface
FMUSER 200W FM Radio Broadcasting Package	Hotel IPTV Solution	FMUSER FSN-1000T V5.0 1KW FM Radio Broadcast Transmitter