利用优化的基础设施加速人工智能推理.pdf

《利用优化的基础设施加速人工智能推理.pdf》由会员分享，可在线阅读，更多相关《利用优化的基础设施加速人工智能推理.pdf（17页珍藏版）》请在三个皮匠报告上搜索。

1、Accelerating AI Inference with Optimized InfrastructureNVIDIAAccelerating AI Inference with Optimized InfrastructureKevin ChenSolution Architect/NVIDIAOPTICAL COMMUNICATION NETWORKSAI Scaling Laws Drive Exponential Demand for Compute New“long thinking”supercharges inference scalingCost increasing wi

2、th compute requires full stack innovation to lower cost/tokenReasoning Drives 100X More Tokens and RevenueKNOWLEDGEKNOWLEDGEPre-trained models simulate internet knowledge100 tokens|$0.50/M TokenREASONINGREASONINGReasoning models think through problems10K tokens 100X costNVIDIA is a multinational tec

3、hnology company that specializes in designing and manufacturing graphics processing units(GPUs),high-performance computing hardware,and artificial intelligence(AI)technologies.The company was founded in 1993 by Jensen Huang,Chris Malachowsky,and Curtis Priem.NVIDIA is headquartered in Santa Clara,Ca

4、lifornia,and has become one of the leading companies in the field of computer graphics,gaming,and AI.What is NVIDIA?Okay,so I need to figure out seating for a family dinner with 8 people.Lets list out the requirements first to make sure Idont miss anything.First,the family has 8 members.Let me confi

5、rm who they are.The user mentions:in-laws(which I assume are the wifes parents),parents(users own parents),sister,little brother,wife,and the user themselves.That adds up to 8:user,wife,users mom,users dad,wifes mom,wifes dad,sister,and little brother.Wait,thats 7.Hmm,maybe I miscounted.Let me check

6、 again.The user says my family of 8 which includes the user,their spouse(wife),their parents(mom and dad),their in-laws(wifes parents:mom and dad),sister,and little brother.Thats 8:user,wife,users mom,users dad,wifes mom,wifes dad,sister,little brother.Yes,thats 8 people.Now,the constraints:1.In-law