HC2022.BirenTech.MikeHong.LingjieXu.v01.pdf

1、壁仞TMBR100 GPGPU:Accelerating Datacenter Scale AI ComputingMike Hong,Lingjie Xu&TeamHot Chips 34,Aug 20221Notice and DisclaimerCopyright 2020-2022 Biren Technology.All rights reserved.Confidentiality.This document contains confidential information of Biren Technology,which shall not be disclosed to a

2、ny third party by any means unless explicitly permitted by Biren Technology.Trademark.All trade names,trademarks,graphical marks and domain names in this document are the properties of Biren Technology.Without prior written consent,they may not be copied,reproduced,modified,published,uploaded,posted

3、,transmitted,or distributed in any way.Forward Looking Statements.Information in this document,other than statements or description of historical fact,may contain forward-looking statements.The forward-looking statements are made based on certain assumptions,projections and calculations made by us w

4、ith regards to the industry and managements expertise.These forward-looking statements are subject to significant risks and uncertainties and our actual results may differ materially.Your business decisions shall not be made solely based on the information.Disclaimer.This material is provided as is

5、without any express or implied warranty of any kind,including warranties of merchantability,title,non-infringement of intellectual property,or fitness for any particular purpose.23BIREN BR100Area1074mm27nmTransistor Count77BillionHost InterfacePCIe Gen 5 x16 w/CXLPeak Performance2048 TOPS INT81024 T

6、FLOPS BF16512 TFLOPS TF32+256 TFLOPS FP32Also supports FP16,INT32,INT16 and otherformatsMemory64GB HBM2EInterconnections8 BLinkTM2.3TB/s external I/O bandwidthForm FactorOAM with 550W Max TDP1.0 X1.0 X1.0 X1.0 X1.0 X2.5 X2.8 X2.7 X2.4 X2.6 X0.0 X0.5 X1.0 X1.5 X2.0 X2.5 X3.0 XResNet-50Bert-LargeTacot