Deploy VTA on Intel FPGA## DEPLOY VTA ON INTEL FPGA ## HARMAN A SAMSUNG COMPANY LIANGFU CHEN 11/16/2019 ## Moore's Law is Slowing Down 42 Years of Microprocessor Trend Data . More...| ## DEPLOY VTA ON INTEL FPGA ## Software - Driver ## Cyclone V & Arria V SoC HPS Physical Memory Map  ## DEPLOY VTA ON INTEL FPGA ## Hardware     ## 当Python遇上FPGA PYNQ开源项目的实践与体会 陆佳华 joshual@Xilinx.com 目录 CONTENTS >> FPGA 35th >> Computer Architecture Golden Age >> PYNQ 589e3488f94aabc1b7868/p3_2.jpg) ## FPGA 35th ## National Inventors Hall of Fame Integrated Circuit Jack Kilby, 1958 Moore's Law Gordon Moore, 1968 FPGA Ross Freeman, 1984  ## Field Programmable Gate Array ## FPGA ## Look-Up Tables (LUT) • Look-up table with N-inputs can be used to implement any combinatorial0 码力 | 9 页 | 3.42 MB | 2 年前3
XDNN TVM - Nov 2019## FPGA CNN Accelerator and TVM Elliott Delaye EXILINX ## TVM Target devices and models HW Platforms  , name=name) return out ## Example of FPGA node in TVM graph { "nodes": [ { "op": "null", "name": "data" (resize) FPGA Acceleration Post-Process (fc/softmax/nms) >> Streamlined multi-process pipeline using shared memory >> Usually need >4 Pre-Process cores running to keep up with FPGA >0 码力 | 16 页 | 3.35 MB | 1 年前3
OctoML OSS 2019 11 88820/p7_8.jpg)  FPGA ASIC ## AutoTVM on μTVM Optimize TVM operators on microcontrollers by making use of AutoTVM https://github Shen, UW and AWS| |12:30|Lunch (boxed lunches will be provided), contributors meetup| |13:30|Building FPGA-Targeted Accelerators with HeteroCL - Zhiru Zhang, Cornell| |14:00|TVM @ Microsoft - Jon Soifer and0 码力 | 16 页 | 1.77 MB | 1 年前3
Bring Your Own Codegen to TVMRuntime (VM, Graph Runtime, Interpreter) Your Dispatcher Target Device General Devices (CPU/GPU/FPGA) ## Mark supported operators or subgraphs 1. Implement an operator-level annotator, OR 2. Implement Runtime (VM, Graph Runtime, Interpreter) Your Dispatcher Target Device General Devices (CPU/GPU/FPGA) ## Mark supported operators or subgraphs 1. Implement extern operator functions, OR 2. Implement Runtime (VM, Graph Runtime, Interpreter) Your Dispatcher Target Device General Devices (CPU/GPU/FPGA) ## Mark supported operators or subgraphs 1. Implement extern operator functions, OR 2. Implement0 码力 | 19 页 | 504.69 KB | 1 年前3
从高并发到极端并发:百度 Feed 与春晚红包的高可用实践-吴永巍jpg) 万一预加载比例低, 高压缩版本兜底 ## 语音+搜索 • 庞大成熟系统,可用性和效果要求极高 - 如何抵御突增的并发? • 语音:专项定制高速模型 + 动态调度 + GPU/FPGA硬件优化 - 搜索:热点,多级Cache+漏斗控制,状态分级+相应策略集合 ✓Best Effort模式,非核心系统做限流 ✓L0准备状态,L1戒备状态,L2活动状态,L2+救急状态 活动准备 L1戒备状态 L2活动状态 春晚活动结束 L2状态准备 动态调度 L2+救急准备 L1状态准备 专项定制 高速模型 完整超大模型 海量算力池 GPU 00:30 CPU FPGA L1戒备状态 恢复常态 L2状态准备 - 抢红包强依赖项,海量用户瞬间涌入 - 多管齐下,多种登录手段,不能让用户等 与百度云、运营商、供应商共建: ✓短信验证码海量能力 √一键登录海量能力 ✓统一管理:流量漏斗,集群间负载 ✓超高性能+精细估算+全局共享+充足buffer ✓部分业务插件前置,为极端并发减少overhead Powered by Golang Powered by FPGA GOLANG  - 安全 ✓全面的防攻击:四层,七层,业务0 码力 | 28 页 | 58.98 MB | 2 年前3
Kubernetes全栈容器技术剖析测序需求多样,测序流程难以灵活自定义  - 结合FPGA加速计算可进一步压缩成本 ## 基于容器的生物信息分析平台  数据库 SFS/OBS 计算 BMS/ECS/FPGA HUAWEI 基于容器的应用流 ## 案例:企业级云容器服务,助力上海蓝鲸传媒容器化上云,提高SLA,降低人力成本 蓝鲸传媒是证券时报旗下,国内首家针对科技媒体人打造的工具型SaaS服务,包含新闻线索平台和记者编辑工作平台 ,参与容器格式及运行规范的定义与实现、积极贡献联邦集群、亲和反亲和等重要特性。 华为CCE在裸金属容器集群、windows容器、集群高可用、自动化运维、容器网络/存储、异构计算(ARM、GPU、FPGA)能力方面具有差 国内首发裸金属容器应对游戏高性能场景;独家提供ARM容器服务支撑低成本APP测试场景 全球首发云容器实例服务CCI:更快的弹性,更高的资源利用率;国内首发windows容器、0 码力 | 26 页 | 3.29 MB | 2 年前3
共 60 条
- 1
- 2
- 3
- 4
- 5
- 6













