跳转到主要内容

AWS采用Habana Gaudi人工智能处理器

由 winniewei 提交于

<p><em><span>图中显示为</span><span lang="EN-US">Habana Labs</span><span>的</span><span lang="EN-US">HL-205 Gaudi</span><span>夹层卡。基于</span><span lang="EN-US">Gaudi</span><span>的</span><span lang="EN-US">EC2</span><span>实例可提供卓越的成本效益和高性能，且同时原生支持</span><span lang="EN-US">TensorFlow</span><span>、</span><span lang="EN-US">PyTorch</span><span>等通用框架。（图片来源：</span><span lang="EN-US">Habana Labs</span><span>）</span></em></p>

<p><strong><span>新闻：</span></strong><span>在近日举行的</span><span lang="EN-US">AWS re:Invent 2020</span><span>（亚马逊</span><span lang="EN-US"> re:Invent 2020</span><span>）大会上，</span><span lang="EN-US">AWS</span><span>首席执行官</span><span lang="EN-US">Andy Jassy</span><span lang="EN-US"><a href="https://aws.amazon.com/ec2/instance-types/habana-gaudi"><span lang="EN-US"><span lang="EN-US">宣布</span></span></a></span><span>了采用最多</span><span lang="EN-US">8</span><span>个</span><span lang="EN-US">Habana® Gaudi®</span><span>加速器的</span><span lang="EN-US">EC2</span><span>实例。对于机器学习工作负载，这些全新</span><span lang="EN-US">EC2</span><span>实例的性价比相较目前基于</span><span lang="EN-US">GPU</span><span>的</span><span lang="EN-US">EC2</span><span>实例提升高达</span><span lang="EN-US">40%1</span><span>。</span><span lang="EN-US">Gaudi</span><span>加速器专为训练深度学习模型而设计，适用于自然语言处理、对象检测和机器学习训练、分类、推荐以及个性化等工作负载。</span></p>

<p><em><span>“我们为</span></em><em><span lang="EN-US">AWS</span></em><em><span>选择</span></em><em><span lang="EN-US">Habana Gaudi</span></em><em><span>处理器用于其即将推出的</span></em><em><span lang="EN-US">EC2</span></em><em><span>训练实例感到自豪。</span></em><em><span lang="EN-US">Habana</span></em><em><span>团队期待继续与</span></em><em><span lang="EN-US">AWS</span></em><em><span>展开合作，为客户持续提供确保连续性及先进性的产品路线图。”</span></em></p>

<p><em><span lang="EN-US">– David Dahan</span></em><em><span>，英特尔公司旗下</span></em><em><span lang="EN-US">Habana Labs</span></em><em><span>首席执行官</span></em></p>

<p><strong><span>重要性：</span></strong><span>作为</span><span lang="EN-US"><a href="https://www.gartner.com/en/newsroom/press-releases/2020-08-10-gartner-s… lang="EN-US"><span lang="EN-US">全球领先</span></span></a></span><span>的云服务提供商，</span><span lang="EN-US">AWS</span><span>被全球各地的开发者用于训练其人工智能模型。然而，机器学习模型复杂程度的增加推升了训练所需的时间和成本，特别是在越来越多的数据可用、且开发者寻求改进其模型的情况下。基于</span><span lang="EN-US">Gaudi</span><span>的</span><span lang="EN-US">EC2</span><span>实例旨在通过交付高成本效益及高性能，并同时原生支持</span><span lang="EN-US">TensorFlow</span><span>和</span><span lang="EN-US">PyTorch</span><span>等通用框架，来满足这些需求。借助</span><span lang="EN-US">Habana</span><span>的</span><span lang="EN-US">SynapseAI</span><span>软件套件，开发者将能够轻松地构建新的训练模型或将现有训练模型从</span><span lang="EN-US">GPU</span><span>移植到</span><span lang="EN-US">Gaudi</span><span>加速器。</span></p>

<p><strong><span>英特尔的人工智能和</span></strong><strong><span lang="EN-US">XPU</span></strong><strong><span>愿景</span></strong><strong><span>：</span></strong><span>英特尔在</span><span lang="EN-US">2019</span><span>年收购了</span><span lang="EN-US">Habana</span><span>，并借此推进其人工智能战略，加强其面向云和数据中心的人工智能加速器产品组合。这包含一系列支持商业、社会及科研领域最具前景的人工智能应用场景的产品和技术。这也体现了公司面向提供全面的跨</span><span lang="EN-US">CPU</span><span>、</span><span lang="EN-US">GPU</span><span>、</span><span lang="EN-US">FPGA</span><span>等的</span><span lang="EN-US">XPU</span><span>架构并以此帮助客户和整个生态系统释放数据潜能的转型。</span></p>

<p><span>英特尔数据平台事业部首席战略官</span><span lang="EN-US">Remi El-Ouazzane</span><span>表示：“我们的产品组合反应了人工智能并非是能用一种解决方案来满足所有场景的计算挑战。云服务提供商目前已广泛采用了内置人工智能性能的英特尔</span><span lang="EN-US">®</span><span>至强</span><span lang="EN-US">®</span><span>处理器来应对人工智能推理工作负载。通过</span><span lang="EN-US">Habana</span><span>，我们现在得以帮助他们降低人工智能模型的大规模训练成本，在高增长的市场机遇中提供极富竞争力的替代选择。”</span></p>

<p><span lang="EN-US">1</span><span>该</span><span lang="EN-US"><a href="https://aws.amazon.com/cn/ec2/instance-types/habana-gaudi/?cid=em&amp;s… lang="EN-US"><span lang="EN-US">成本效益声明</span></span></a></span><span>由</span><span lang="EN-US">AWS</span><span>根据</span><span lang="EN-US">AWS</span><span>内部测试提出。</span><span lang="EN-US"> Habana Labs</span><span>不控制或审核第三方数据；您的费用和效果可能会有所不同。</span></p>

<p><strong><span>关于英特尔</span></strong></p>

<p><span>英特尔（</span><span lang="EN-US">NASDAQ: INTC</span><span>）作为行业引领者，创造改变世界的技术，推动全球进步并让生活丰富多彩。在摩尔定律的启迪下，我们不断致力于推进半导体设计与制造，帮助我们的客户应对最重大的挑战。通过将智能融入云、网络、边缘和各种计算设备，我们释放数据潜能，助力商业和社会变得更美好。<span>如需了解英特尔创新的更多信息，请访问英特尔中国新闻中心</span></span><span lang="EN-US"><a href="http://newsroom.intel.cn/"><span>newsroom.intel.cn</span></a></span><sp… lang="EN-US">&nbsp;</span><span>以及官方网站</span><span lang="EN-US">&nbsp;</span><span lang="EN-US"><a href="http://www.intel.cn/"><span>intel.cn</span></a></span><span&gt;。</span></p>