<p><em><span>为下一代汽车应用场景提供响应最快和能效最高的神经网络加速功能</span></em></p>
<p><strong><span>英国</span></strong><strong><span lang="EN-US">伦敦</span></strong><strong><span>，</span></strong><strong><span lang="EN-GB">2020</span></strong><strong><span lang="EN-US">年</span></strong><strong><span lang="EN-GB">11</span></strong><strong><span lang="EN-US">月</span></strong><strong><span lang="EN-GB">12</span></strong><strong><span lang="EN-US">日</span></strong><span lang="EN-GB">– </span><span lang="EN-GB"><a href="http://www.imaginationtech.com/"><span>Imagination Technologies</span></a></span><span lang="EN-US">宣布推出</span><span>面向先进驾驶辅助系统（</span><span lang="EN-GB">ADAS</span><span>）和自动驾驶应用的新一代神经网络加速器（</span><span lang="EN-GB">NNA</span><span>）产品</span><span lang="EN-GB"><a href="https://www.imgtec.com/vision-ai/img-series4-nna/"><span>IMG Series4</span></a></span><span>。</span><span lang="EN-GB">Series4</span><span>可为领先的汽车行业颠覆者、一级供应商、整车厂（</span><span lang="EN-GB">OEM</span><span>）和汽车系统级芯片（</span><span lang="EN-GB">SoC</span><span>）厂商提供强大助力。</span></p>
<p><span lang="EN-GB">Series4</span><span>拥有全新的多核架构，可提供</span><span lang="EN-GB">600TOPS</span><span>（每秒万亿次操作）甚至更高的超高性能，并且可为大型神经网络工作负载提供低带宽和极低的延迟。</span></p>
<p><span>汽车行业正处于变革的风口浪尖，自动驾驶汽车和自动驾驶出租车等新的应用场景对人工智能（</span><span lang="EN-GB">AI</span><span>）的性能水平提出了更高的要求。为此，</span><span lang="EN-GB">Imagination</span><span>已经与汽车行业和其他重视功能安全的行业中的领先企业和创新者展开了合作。</span></p>
<p><span lang="EN-GB">Series4</span><span>已经开始提供授权，并将于</span><span lang="EN-GB">2020</span><span>年</span><span lang="EN-GB">12</span><span>月在市场上全面供应。</span></p>
<p><span lang="EN-GB">Imagination</span><span>的低功耗</span><span lang="EN-GB">NNA</span><span>架构旨在执行完整的网络推理，同时满足功能安全要求。它可以一次执行多个操作，从而最大化每瓦性能并提供行业领先的能效。</span></p>
<p><span lang="EN-GB">Series4</span><span>具有以下特性：</span></p>
<ul>
<li><strong><span>多核扩展性和灵活性：</span></strong><span>多核架构支持在多个内核之间对工作负载进行灵活的分配和同步。</span><span lang="EN-GB">Imagination</span><span>的软件提供了精细的控制能力，并通过对多个工作负载进行批处理、拆分和调度而提高了灵活性，现在可以在任意数量的内核上使用。</span><span lang="EN-GB">Series4</span><span>可为每个集群配置</span><span lang="EN-GB"> 2</span><span>个、</span><span lang="EN-GB">4</span><span>个、</span><span lang="EN-GB">6</span><span>个或者</span><span lang="EN-GB">8</span><span>个内核。</span></li>
<li><strong><span>超高性能：</span></strong><span lang="EN-GB">Series4</span><span>的每个单核能够以不到一瓦的功耗提供</span><span lang="EN-GB">12.5 TOPS</span><span>的性能。举例来说，一个</span><span lang="EN-GB">8</span><span>内核集群可以提供</span><span lang="EN-GB">100 TOPS</span><span>的算力，那么，配有</span><span lang="EN-GB">6</span><span>个</span><span lang="EN-GB">8</span><span>核集群的解决方案就可以提供</span><span lang="EN-GB">600 TOPS</span><span>的算力。在</span><span lang="EN-GB">AI</span><span>推理方面，</span><span lang="EN-GB">Series4 NNA</span><span>的性能比嵌入式</span><span lang="EN-GB">GPU</span><span>快</span><span lang="EN-GB">20</span><span>倍以上，比嵌入式</span><span lang="EN-GB">CPU</span><span>快</span><span lang="EN-GB">1000</span><span>倍。</span></li>
<li><strong><span>超低延迟：</span></strong><span>通过将多个单核组成</span><span lang="EN-GB">2</span><span>核、</span><span lang="EN-GB">4</span><span>核、</span><span lang="EN-GB">6</span><span>核或</span><span lang="EN-GB">8</span><span>核的多核集群，所有内核可以相互协作，并行处理一个任务，降低处理延迟，缩短响应时间。例如，对于一个</span><span lang="EN-GB">8</span><span>核集群，理想情况下延迟会减少为单核独立执行时的</span><span lang="EN-GB">1/8</span><span>。</span></li>
<li><strong><span>节省大量带宽：</span></strong><span lang="EN-GB">Imagination</span><span>的</span><span lang="EN-GB">TensorTiling</span><span>技术（</span><span lang="EN-GB">Imagination’s </span><span lang="EN-GB">Tensor Tiling</span><span>，</span><span lang="EN-GB">ITT</span><span>）是一项正在申请专利的技术，也是</span><span lang="EN-GB">Series4</span><span>中新增的功能，它可以通过对计算任务进行</span><span lang="EN-GB">tiling</span><span>，充分利用片上存储，提升数据处理效率，并节省访问外部存储的带宽。</span><span lang="EN-GB">ITT</span><span>利用本地数据的依赖性将中间数据保存在片上存储器中，这样可以最大限度地减少将数据传输至外部存储器，从而将带宽降低多达</span><span lang="EN-GB">90</span><span>％。</span><span lang="EN-GB">ITT</span><span>是一种可扩展的算法，在拥有大量输入数据的网络上具有显著优势。</span></li>
<li><strong><span>车规级安全性：</span></strong><span lang="EN-GB">Series4</span><span>包含</span><span lang="EN-GB">IP</span><span>级别的安全功能且设计流程符合</span><span lang="EN-GB">ISO 26262</span><span>标准，可以帮助客户获得</span><span lang="EN-GB">ISO 26262</span><span>认证。</span><span lang="EN-GB">ISO 26262</span><span>是旨在解决汽车电子产品风险的行业安全标准。</span><span lang="EN-GB">Series4</span><span>可以在不影响性能的情况下，安全地进行神经网络推理。硬件安全机制可以保护编译后的网络、网络的执行和数据处理管道。</span></li>
</ul>
<p><span lang="EN-GB">ABI Research</span><span>智慧出行和汽车首席分析师</span><span lang="EN-GB">James Hodgson</span><span>说道：“虽然我们还在预计到</span><span lang="EN-GB">2027</span><span>年左右，对</span><span lang="EN-GB">ADAS</span><span>的需求将增长两倍，但汽车行业已然将目光投向了更远的全自动驾驶汽车和自动驾驶出租车。在从</span><span lang="EN-GB">L2</span><span>和</span><span lang="EN-GB">L3</span><span>级</span><span lang="EN-GB">ADAS</span><span>向</span><span lang="EN-GB">L4</span><span>和</span><span lang="EN-GB">L5</span><span>级</span><span>全自动驾驶演进的过程中，神经网络的广泛应用将是至关重要的因素。这些系统将要处理数以百计的复杂场景，从多个摄像头和激光雷达等大量传感器中提取数据，从而实现自动代客泊车、十字路口管理和复杂城市环境安全导航等解决方案。高性能、低延迟和高能效的结合将是实现高度自动驾驶的关键所在。</span><span lang="EN-US">”</span></p>
<p><span lang="EN-GB">Imagination Technologies</span><span>人工智能业务高级总监</span><span lang="EN-GB">Andrew Grant</span><span>表示：“我们相信</span><span lang="EN-GB">Series4 NNA</span><span>将成为开发先进驾驶辅助系统和自动驾驶汽车的行业标准平台。一些创新者已经在着手打造支持下一代</span><span lang="EN-GB">ADAS</span><span>功能和自动驾驶汽车的芯片。任何希望在汽车领域扮演重要角色的公司或研发团队是时候将这项技术集成至他们的平台中了。</span><span lang="EN-US">”</span></p>
<p><span>有关</span><span lang="EN-GB">Series4 NNA</span><span>的更多信息，请点击<span>下方链接</span>观看我们的主题会议。</span></p>
<p><span lang="EN-GB"><a href="http://imgtec.eetrend.com/content/2020/100058657.html"><span>http://img…;
<p><strong><span>关于</span></strong><strong><span lang="EN-GB">Imagination Technologies</span></strong></p>
<p><span lang="EN-GB">Imagination</span><span>是一家总部位于英国的公司，致力于打造半导体和软件知识产权（</span><span lang="EN-GB">IP</span><span>），使客户在竞争激烈的全球技术市场中获得足够优势。公司的图形、计算、视觉和人工智能以及连接技术可以实现出众的</span><span lang="EN-GB">PPA</span><span>（功耗、性能和面积）指标、强大的安全性、快速的上市时间和更低的总体拥有成本（</span><span lang="EN-GB">TCO</span><span>）。基于</span><span lang="EN-GB">Imagination IP</span><span>的产品被全球数十亿人用于他们的手机、汽车、住宅和工作场所。</span><span lang="EN-GB">Imagination Technologies</span><span>于</span><span lang="EN-GB">2017</span><span>年被全球私募股权投资基金</span><span lang="EN-GB">Canyon Bridge</span><span>收购。更多信息，请访问</span><span><span lang="EN-GB"><a href="http://www.imaginationtech.com/"><span>www.imaginationtech.com</span></…;。</span></p>
<p><span>在</span><span lang="EN-GB"><a href="https://twitter.com/ImaginationTech"><span>Twitter</span></a></span><sp… lang="EN-US">、</span><span lang="EN-GB"><a href="http://www.youtube.com/imgtec"><span>YouTube</span></a></span><span lang="EN-US">、</span><span lang="EN-GB"><a href="http://www.linkedin.com/company/imagination-technologies"><span>LinkedI… lang="EN-US">、</span><span lang="EN-GB"><a href="http://www.imgtec.com/pressrelease-rss.asp"><span>RSS</span></a></span>… lang="EN-US">、</span><span lang="EN-GB"><a href="https://www.facebook.com/imgtec"><span>Facebook</span></a></span><span lang="EN-US">和</span><span lang="EN-GB"><a href="http://imgtec.com/blog"><span>Blog</span></a></span><span lang="EN-US">上关注</span><span lang="EN-GB">Imagination</span><span lang="EN-US">。</span></p>