GPU 优化虚拟机大小GPU optimized virtual machine sizes

GPU 优化 VM 大小是具有单个或多个 NVIDIA GPU 的专用虚拟机。GPU optimized VM sizes are specialized virtual machines available with single or multiple NVIDIA GPUs. 这些大小是针对计算密集型、图形密集型和可视化工作负荷设计的。These sizes are designed for compute-intensive, graphics-intensive, and visualization workloads. 本文介绍有关 GPU、vCPU、数据磁盘和 NIC 的数量和类型的信息。This article provides information about the number and type of GPUs, vCPUs, data disks, and NICs. 此分组中的每个大小还包括存储吞吐量及网络带宽。Storage throughput and network bandwidth are also included for each size in this grouping.

  • NCv3 大小针对计算密集型和网络密集型应用程序和算法进行了优化。NCv3 sizes are optimized for compute-intensive and network-intensive applications and algorithms. 一些示例包括基于 CUDA 和 OpenCL 的应用程序和模拟、AI 和深度学习。Some examples are CUDA- and OpenCL-based applications and simulations, AI, and Deep Learning. NCv3 系列专用于采用 NVIDIA Tesla V100 GPU 的高性能计算工作负载。The NCv3-series is focused on high-performance computing workloads featuring NVIDIA's Tesla V100 GPU. NCv3 系列 VM 使用 Intel Xeon E5-2690 v4 (Broadwell) 处理器。The NCv3-series VMs use the Intel Xeon E5-2690 v4 (Broadwell) processor.

NCv3 系列NCv3-series

高级存储:支持Premium Storage: Supported

高级存储缓存:支持Premium Storage caching: Supported

NCv3 系列 VM 采用 NVIDIA Tesla V100 GPU。NCv3-series VMs are powered by NVIDIA Tesla V100 GPUs. 这些 GPU 可提供 NCv2 系列的 1.5 倍计算性能。These GPUs can provide 1.5x the computational performance of the NCv2-series. 客户可将这些更新的 GPU 用于传统的 HPC 工作负荷,例如油藏模拟、DNA 测序、蛋白质分析、Monte Carlo 模拟和其他工作负荷。Customers can take advantage of these updated GPUs for traditional HPC workloads such as reservoir modeling, DNA sequencing, protein analysis, Monte Carlo simulations, and others. NC24rs v3 配置提供了针对紧密耦合的并行计算工作负荷优化的低延迟、高吞吐量网络接口。The NC24rs v3 configuration provides a low latency, high-throughput network interface optimized for tightly coupled parallel computing workloads. 除了 GPU 之外,NCv3 系列 VM 还采用 Intel Xeon E5-2690 v4 (Broadwell) CPU。In addition to the GPUs, the NCv3-series VMs are also powered by Intel Xeon E5-2690 v4 (Broadwell) CPUs.

Important

对于此大小系列,订阅中的 vCPU(核心)配额在每个区域中最初都设置为 0。For this size family, the vCPU (core) quota in your subscription is initially set to 0 in each region. 可以请求在某个可用区域提高此系列的 vCPU 配额Request a vCPU quota increase for this family in an available region.

大小Size vCPUvCPU 内存:GiBMemory: GiB 临时存储 (SSD):GiBTemp storage (SSD): GiB GPUGPU GPU 内存:GiBGPU memory: GiB 最大数据磁盘数Max data disks 最大非缓存磁盘吞吐量:IOPS/MbpsMax uncached disk throughput: IOPS / MBps 最大 NIC 数Max NICs
Standard_NC6s_v3Standard_NC6s_v3 66 112112 736736 11 1616 1212 20000 / 20020000 / 200 44
Standard_NC12s_v3Standard_NC12s_v3 1212 224224 14741474 22 3232 2424 40000 / 40040000 / 400 88
Standard_NC24s_v3Standard_NC24s_v3 2424 448448 29482948 44 6464 3232 80000 / 80080000 / 800 88
Standard_NC24rs_v3*Standard_NC24rs_v3* 2424 448448 29482948 44 6464 3232 80000 / 80080000 / 800 88

1 GPU = 一个 V100 卡。1 GPU = one V100 card.

*支持 RDMA*RDMA capable

大小表定义Size table definitions

  • 存储容量的单位为 GiB 或 1024^3 字节。Storage capacity is shown in units of GiB or 1024^3 bytes. 比较以 GB(1000^3 字节)为单位的磁盘和以 GiB(1024^3 字节)为单位的磁盘时,请记住以 GiB 为单位的容量数显得更小。When comparing disks measured in GB (1000^3 bytes) to disks measured in GiB (1024^3) remember that capacity numbers given in GiB may appear smaller. 例如,1023 GiB = 1098.4 GBFor example, 1023 GiB = 1098.4 GB
  • 磁盘吞吐量的单位为每秒输入/输出操作数 (IOPS) 和 Mbps,其中 Mbps = 10^6 字节/秒。Disk throughput is measured in input/output operations per second (IOPS) and MBps where MBps = 10^6 bytes/sec.
  • 数据磁盘可以在缓存或非缓存模式下运行。Data disks can operate in cached or uncached modes. 对于缓存数据磁盘操作,主机缓存模式设置为 ReadOnlyReadWriteFor cached data disk operation, the host cache mode is set to ReadOnly or ReadWrite. 对于非缓存数据磁盘操作,主机缓存模式设置为 NoneFor uncached data disk operation, the host cache mode is set to None.
  • 若要获得 VM 的最佳性能,应将数据磁盘数限制为每 vCPU 2 个磁盘。If you want to get the best performance for your VMs, you should limit the number of data disks to 2 disks per vCPU.
  • 预期的网络带宽是指跨所有 NIC 为每个 VM 类型分配的最大聚合带宽,适用于所有目标。Expected network bandwidth is the maximum aggregated bandwidth allocated per VM type across all NICs, for all destinations. 不保证能够达到上述预期上限,但目的是提供相关指导,以便为目标应用程序选择适当的 VM 类型。Upper limits are not guaranteed, but are intended to provide guidance for selecting the right VM type for the intended application. 实际的网络性能将取决于许多因素,包括网络拥塞、应用程序负载和网络设置。Actual network performance will depend on a variety of factors including network congestion, application loads, and network settings. 有关如何优化网络吞吐量的信息,请参阅为 Windows 和 Linux 优化网络吞吐量For information on optimizing network throughput, see Optimizing network throughput for Windows and Linux. 若要在 Linux 或 Windows 中达到预期的网络性能,可能需要选择特定版本,或者需要优化 VM。To achieve the expected network performance on Linux or Windows, it may be necessary to select a specific version or optimize your VM. 有关详细信息,请参阅如何针对虚拟机吞吐量进行可靠的测试For more information, see How to reliably test for virtual machine throughput.

支持的分发和驱动程序Supported distributions and drivers

若要利用运行 Linux 的 Azure N 系列 VM 的 GPU 功能,必须安装 NVIDIA GPU 驱动程序。To take advantage of the GPU capabilities of Azure N-series VMs running Linux, NVIDIA GPU drivers must be installed. NVIDIA GPU 驱动程序扩展可在 N 系列 VM 上安装适当的 NVIDIA CUDA 或 GRID 驱动程序。The NVIDIA GPU Driver Extension installs appropriate NVIDIA CUDA or GRID drivers on an N-series VM. 请使用 Azure 门户或工具(例如 Azure CLI 或 Azure 资源管理器模板)安装或管理该扩展。Install or manage the extension using the Azure portal or tools such as the Azure CLI or Azure Resource Manager templates. 有关受支持的分发版和部署步骤,请参阅 NVIDIA GPU 驱动程序扩展文档See the NVIDIA GPU Driver Extension documentation for supported distributions and deployment steps. 有关 VM 扩展的常规信息,请参阅 Azure 虚拟机扩展和功能For general information about VM extensions, see Azure virtual machine extensions and features.

如果选择手动安装 NVIDIA GPU 驱动程序,请参阅适用于 Linux 的 N 系列 GPU 驱动程序安装,了解受支持的分发版、驱动程序以及安装和验证步骤。If you choose to install NVIDIA GPU drivers manually, see N-series GPU driver setup for Linux for supported distributions, drivers, and installation and verification steps.

部署注意事项Deployment considerations

  • 有关 N 系列 VM 的可用性,请查看可用产品(按区域)For availability of N-series VMs, see Products available by region.

  • N 系列 VM 只能按 Resource Manager 部署模型部署。N-series VMs can only be deployed in the Resource Manager deployment model.

  • N 系列的 VM 在对其磁盘支持的 Azure 存储类型方面有所不同。N-series VMs differ in the type of Azure Storage they support for their disks. NCv3 VM 仅支持高级磁盘存储 (SSD) 所支持的 VM 磁盘。NCv3 VMs only support VM disks that are backed by Premium Disk Storage (SSD).

  • 如果需要部署的 N 系列 VM 较多,请考虑使用标准预付费套餐订阅或其他购买选项。If you want to deploy more than a few N-series VMs, consider a Standard Pay-in-Advance Offer subscription or other purchase options. 如果使用的是 Azure 试用帐户,则仅可使用有限数量的 Azure 计算核心。If you're using an Azure trial account, you can use only a limited number of Azure compute cores.

  • 可能需要提高 Azure 订阅中的核心配额(按区域)以及单独针对 NCv3 核心的配额。You might need to increase the cores quota (per region) in your Azure subscription, and increase the separate quota for NCv3 cores. 若要请求增加配额,可免费 建立联机客户支持请求To request a quota increase, open an online customer support request at no charge. 默认限制可能因订阅类别而异。Default limits may vary depending on your subscription category.

  • 不应在 Ubuntu NC VM 上安装 X server 或使用 Nouveau 驱动程序的其他系统。You shouldn't install X server or other systems that use the Nouveau driver on Ubuntu NC VMs. 在安装 NVIDIA GPU 驱动程序之前,需要禁用 Nouveau 驱动程序。Before installing NVIDIA GPU drivers, you need to disable the Nouveau driver.

其他大小Other sizes

后续步骤Next steps

了解有关 Azure 计算单元 (ACU) 如何帮助跨 Azure SKU 比较计算性能的详细信息。Learn more about how Azure compute units (ACU) can help you compare compute performance across Azure SKUs.