Virtualizing GPUs In VMware VSphere Using NVIDIA Virtual Compute Server .
Reference ArchitectureVirtualizing GPUs in VMwarevSphere using NVIDIA VirtualCompute Serveron Dell EMC infrastructureAbstractDell Technologies delivers an on-premises enterprise solution stack to enableorganizations to virtualize GPUs and offer GPUs as a Service (GPUaaS) on DellEMC infrastructure.Michael Bennett, Senior Principal EngineerRamesh Radhakrishnan, Distinguished EngineerJune 2020Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
ContentsAudience . 3Dell EMC Reference Architecture for NVIDIA Virtual Compute Server . 3GPU Virtualization with Virtual Compute Server . 5Use cases. 6Data Science Workbench . 6Increase GPU Utilization . 6Live Migration with GPUs . 7Understanding the Performance of Virtual GPUs. 7Virtual GPU Resource Scheduling . 7Test Setup . 8Comparing Virtual GPU to Bare Metal. 9Fractional Virtual GPU . 10Multiple Virtual GPUs . 11Architecture Overview . 12NVIDIA Virtual Compute Server . 12PowerEdge NGC-Ready Servers . 13Dell EMC PowerEdge R740. 13Dell EMC PowerEdge C4140 . 13VMware vSphere . 13NVIDIA GPUs . 14Networking . 17Storage . 18Get started virtualizing GPUs with Virtual Compute Server. 19Enterprise Virtual GPU with VSAN . 19Scale-out GPU Virtualization . 19Summary . 20Learn more . 20Appendix . 21VMware Documentation. 21NVIDIA Documentation . 21Dell EMC Documentation. 212 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
AudienceThis guide to the Dell EMC Reference Architecture for NVIDIA Virtual ComputeServer is intended for data center operators, IT administrators and systemsengineers who support GPU infrastructure or will in the future. This guide alsocontains information for data science leaders looking for information onaccelerating data science efforts by gaining access to GPU resources ondemand.Dell EMC Reference Architecture for NVIDIA Virtual ComputeServerThe Dell EMC Reference Architecture for NVIDIA Virtual Compute Servercontains the foundational elements to build a solution that enables GPUvirtualization inside the datacenter. With this reference architecture, ITadministrators unlock the ability to allocate partitions of GPU resources withinVMware vSphere , as well as support for live migration of virtual machinesrunning NVIDIA CUDA accelerated workloads. Some of the key benefitscustomers can achieve with this solution are: Democratize GPU access by providing partitions of GPUs on demand Ability to scale up and scale down GPU resource assignments asneeded Support for live migration of GPU memory.3 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
GPU on DemandDeploy and Scale AIStack ID4 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
The purpose of this guide is to describe the technologies used to enable GPUvirtualization, how vGPUs can be a tool in digital transformation and theperformance of NVIDIA T4 and V100 Tensor Core GPUs when used in severaldifferent configurations. With this knowledge, you will come out well equipped toradically change how your enterprise operates by enabling artificial intelligence(AI) to be deployed at scale and enabling on demand access to NVIDIA GPUs toaccelerate data centric workloads such as analytics and data science.GPU Virtualization with Virtual Compute ServerNVIDIA Virtual Compute Server enables the benefits of VMware virtualization forGPU-accelerated PowerEdge servers. With Virtual Compute Server, data centeradmins are able to power compute-intensive workloads with GPUs in a virtualmachine (VM). Virtual Compute Server software virtualizes NVIDIA GPUs toaccelerate large workloads, including more than 600 GPU acceleratedapplications for AI, deep learning, and HPC. With GPU sharing, multiple VMs canbe powered by a single GPU, maximizing utilization and affordability, or a singleVM can be powered by multiple virtual GPUs, making even the most intensiveworkloads possible. And with support for all major hypervisor virtualizationplatforms, data center admins can use the same management tools for theirGPU-accelerated servers as they do for the rest of their data center.Features of NVIDIA Virtual Compute Server include: GPU sharing (fractional) enables multiple VMs to share a GPU,maximizing utilization for lighter workloads that require GPUacceleration. With GPU aggregation, a VM can access more than one GPU, which isoften required for compute-intensive workloads. Virtual Compute Serversupports using multiple virtual GPUs and where NVLink is present,Virtual Compute Server will take advantage of the peer-to-peer data pathbetween GPUs. Proactive management features provide the ability to do live migration,suspend and resume, and create thresholds that expose consumptiontrends impacting user experiences. Administrators can use the samehypervisor virtualization tools to manage GPU servers, with visibility atthe host, virtual machine and app level. Advanced compute: Error correcting code and dynamic page retirementprevent against data corruption for high-accuracy workloads. Increased security: Enterprises can extend security benefits of servervirtualization to GPU clusters. Multi-tenant isolation: Workloads can be isolated to securely supportmultiple users on a single infrastructure. Broad range of supported GPUs: Virtual Compute Server is supported onNVIDIA T4 or V100 Tensor Core GPUs, as well as Quadro RTX 8000and 6000 GPUs, and prior generations of Pascal microarchitecture.5 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Use casesThis Virtual Compute Server Reference Architecture offers flexible ways for youto you consume GPU resources by providing the ability to orchestrate GPUresources at a granular level, aggregate multiple GPUs to achieve results fasterand live migrate CUDA-accelerated applications from one GPU node to anotherin your VMware vSphere environment. This helps IT administrators bettersupport AI workloads because it brings lifecycle management of the NVIDIA GPUaccelerator resources inside the familiar vSphere environment. Youradministrators will have the freedom to move GPU workloads around thedatacenter as needed for lifecycle management of the VMware ESXi hosts too.Data Science WorkbenchWhy does every data scientist ask for a GPU? When developing a deep learningmodel, your data science team will go through an iterative development process.To enable each member of a team to work independently, their data scienceworkbench will need to have a GPU accelerator available for model training andmany types of data transformation operations. Providing each team member theirown GPU would be expensive as well as bring power and cooling issues.With NVIDIA Virtual Compute Server, IT administrators can allocate a fraction ofthe memory and compute power of an NVIDIA GPU when attaching acceleratorsto a virtual machine, enabling up to eight data scientists to use a single NVIDIAV100 GPU with 32GB memory at the same time for training model candidatesand hyper parameter tuning.Dell EMC offers several AI reference architectures that deliver Data ScienceWorkbench functionality. When used in conjunction with the Virtual ComputeServer reference architecture, you will achieve higher consolidation ratios so thatmore data scientists can enjoy the benefit of access to GPU accelerators.Increase GPU UtilizationWhen running AI in production environments, it is challenging for ITadministrators size the workload, that is to determine how many GPU I/Oresources it needs across CUDA Cores, Tensor Cores and memory. With GPUvirtualization provided by NVIDIA Virtual Compute Server, multiple virtualizedinstances can be combined, such as three instances of an NVIDIA T4 GPU,effectively equivalent to 3x the performance of a single GPU. The ability tocombine up to 16 virtualized GPUs allows granular scalability and scheduling tohelp right-size each application and achieve higher utilization throughcomposable consolidation and expansion of resources.6 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Live Migration with GPUsVirtual Compute Server enables VMware vSphere vMotion capabilities for virtualmachines with virtual GPUs attached. This enables IT operations to balanceresources effectively across the cluster without any disruption to the user orworkload. This migration capability can also be used to enable automatic rollingupdates on clusters that have GPU-accelerated workloads running on them.Understanding the Performance of Virtual GPUsTo understand how GPU virtualization with NVIDIA changes the performance ofthe GPU, we put the reference architecture through several tests. First, wecompared an NVIDIA V100 16GB GPU running on bare metal Linux to avirtualized GPU using the 16c profile.After establishing that baseline of performance, additional testing was doneinvolving multiple virtual GPUs and fractional virtual GPUs.Virtual GPU Resource SchedulingWhen using GPU profiles that assign a fraction of the physical GPU, you canexpect some difference, for example two V100-8C profiles will not always offerthe same aggregate performance as a single V100-16C profile. This is due totime-shared scheduling and context switching that enables multiple independentvirtual machines to run CUDA workloads to simultaneously on a single GPU.There are scenarios where the performance difference is favorable. Whentraining two neural networks, you can sometimes see faster time to result usingmultiple, fractional GPUs in parallel than you can using a full GPU andscheduling the training tasks in a serially. There are two primary reasons for this:1. The GPU scheduling policy “Best Effort” allows idle time slices to beused by virtual machines with workloads that can make use of theprocessing power.2. Parallel processing of GPU-accelerated computation steps is only mostof the work in training a neural network. Parallel execution of jobs onfractional GPU profiles engages more CPU resources to help in thetraining effort.In most cases, users can expect there to be a small difference in performance ofbetween 2%-5% compared to bare metal when using virtual GPU profiles formachine learning and deep learning workloads.7 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Test SetupFor our testing of Virtual Compute Server, we captured the performance for deeplearning training workloads. These tests generate synthetic data rather thanreading data from disk or over the network. By using synthetic data, we cancapture the performance of the GPU with little to no impact from other variablessuch as drive type inside the server or what network protocol the VM uses tocommunicate.Three virtual GPU configurations supported by Virtual Compute ServerTesting these configurations was performed with three Dell EMC PowerEdgeR740 servers, each equipped with two Intel Xeon Gold 6132 processors and384GB of DDR4-2666 memory. We used V100 and T4 GPUs from NVIDIA forthe training and selected several similar neural networks to observe howchanges in model architecture for similar use cases change performance.In total we trained a few neural networks such as, Resnet-50, ResNet-152,Inception V2 and VGG networks, at various batch sizes. Testing on two differentGPU models results in us collecting data from 48 unique experiment designs. Inthis section we will present some of the results.8 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Comparing Virtual GPU to Bare MetalIn a server equipped with a 16GB NVIDIA V100 GPU, any virtual machineassigned a V100-16c virtual GPU profile will consume the entire GPU, but thereis a small amount of overhead from virtualization that can lower performance by2%-5% when compared to a server running the same operating system and GPUinstalled on bare metal and 0%-2% when compared to a virtual machine usingVMware vSphere DirectPath I/O .Comparing vGPU performance to bare metal with NVIDIA V100 16GBComparing vGPU performance to bare metal with NVIDIA T4 16GB9 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Fractional Virtual GPUFractional virtual GPU profiles are those that do not consume all of the memoryavailable in the GPU. When using fractional virtual GPU profiles, it is important toconsider the scheduling policy used on the GPU because other workloads andtheir access patterns can cause performance variations. Let’s assume we have avirtual machine as part of an image recognition service. This virtual machine isresponsible for inference and is always operating using a virtual GPU profile thatconsumes half the available memory of a T4. If the other half of the T4 isallocated to a virtual machine used for data science research and we choose“Best Effort” then the image recognition service will have higher, but lesspredictable performance. This is because the Best Effort scheduler allows avGPU to use GPU processing cycles that are not being used by other vGPUs aslong as they remain idle.When a researcher decides to do GPU-accelerated data transformation or train amodel candidate, the image recognition service will suddenly see delays as othercommands are scheduled through timesharing. If we instead used Fixed Shareor Equal Share scheduling, then the inference service would have morepredictable performance but less aggregate.To test the effect of scheduling policy changes, we ran our neural networktraining jobs as follows. First, we assigned a V100-8C profile to a virtual machine(half vGPU). Then, we trained the neural network again but this time with asecond virtual machine, also assigned a v100-8C profile, training a copy of thesame neural network also. As you can see from the graphs below Best Effortscheduling gives the best performance when other virtual machines are idle butduring times where a second VM is co-scheduled the performance drops by alittle more than half.Best Effort scheduling policy tested10 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
With the Fixed Share policy, the virtual machine is never allocated more resources evenwhen other virtual machines are idle. When other virtual machines begin using their coscheduled GPUs, there is no impact to the other users.Fixed Share scheduling policy testedMultiple Virtual GPUsLeveraging multiple GPU devices can be beneficial in achieving results faster. Wetested this by training the neural networks again with two virtual GPU resourcesassigned to it. Important to note is that Virtual Compute Server does not support usingmultiple virtual GPUs if fractional GPU profiles are assigned to the virtual machine.Speed-up achieved with multiple virtual GPUs11 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Architecture OverviewThe Dell EMC Reference Architecture for NVIDIA Virtual Compute Server usesthe latest generation of PowerEdge servers. PowerEdge GPU-capable serverscome in a diverse set of form factors and there are options for NVLink or PCIe forGPU to GPU communication.Reference Architecture for Virtual Compute ServerNVIDIA Virtual Compute ServerThe NVIDIA Virtual Compute Server technology is enabled with two components– a .vib file that gets installed on the ESXi host and a license server that runs asa virtual machine inside your environment.Virtual GPU ESXi PackageUsing Virtual Compute Server with vSphere requires installation of the vGPUManager VIB on the ESXi hosts. When installing this package, take caution – theversions for the NVIDIA Virtual GPU Manager and the Guest VM driver must bethe same version.License ServerThe license server is a virtual machine running Java Runtime Environment andApache Tomcat . This virtual machine is responsible for communication with theVirtual GPU Manager installed on the ESXi host regarding the status of VirtualCompute Server licenses.12 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
PowerEdge NGC-Ready ServersIt is recommended that you use PowerEdge NGC-Ready servers with NVIDIAVirtual Compute Server. These systems include Dell EMC PowerEdge R740,R740XD and C4140 servers that have passed an extensive suite of tests thatvalidate the ability to deliver workloads on NGC containers with highperformance. NVIDIA NGC-Ready system validation includes tests of: Single and multi-GPU Deep Learning training using TensorFlow,PyTorch and NVIDIA DeepStream Transfer Learning Toolkit High volume, low latency inference using NVIDIA TensorRT, TensorRTInference Server, and DeepStream Data Science using RAPIDS and XGBoost Application development using the CUDA Toolkit.Dell EMC PowerEdge R740The PowerEdge R740 platform is a 2U dual socket server equipped with thelatest Intel Xeon Scalable processors. It can be configured with up to 3xNVIDIA Quadro RTX6000, Quadro RTX8000 or V100 GPUs or 6x NVIDIA T4GPUs, but also offers the flexibility to support additional configurations such as24x 2.5” NVMe drives and two NVIDIA GPUs. The PowerEdge R740 is alsocertified for VMware vSAN making it the ideal building block for a hyperconverged, GPU-accelerated cluster to power AI workloads in the datacenter andremote office deployments that require GPU acceleration locally for inference orreal-time model training/tuning close to the source of data.Dell EMC PowerEdge C4140With room for 4x full-length, full-width PCIe GPUs or 4x SXM2 form factor V100GPUs in 1U of rack space, the PowerEdge C4140 is ideal for scale-outenterprise AI workloads. The PowerEdge C4140 used in our testing wasequipped with NVIDIA NVlink, enabling peer to peer communication of the 4xSXM2 V100s. The PowerEdge C4140 storage configurations cannot supportVMware vSAN but can be configured to use external storage. Manyconfigurations, including the NVlink configuration, have available expansion for adedicated storage adapter such as a Fiber Channel HBA or RDMA-capablenetwork interface card (NIC).VMware vSphereVMware vSphere is a virtualization platform that allows data center operators tomanage infrastructure as a unified operating environment and provides the toolsfor end-to-end lifecycle management. The two core components of vSphere areESXi and vCenter Server . ESXi is the virtualization platform where you createand run virtual machines and virtual appliances. vCenter Server is the servicethrough which you manage multiple hosts connected in a network and pool hostresources.13 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
There are several features delivered by ESXi that are necessary to ensureamazing end user experience when delivering virtualized GPU resources.Support for RoCE (RDMA over Converged Ethernet) with PVRDMA adaptersprovides a low-latent, highly performant connection. Through integration withvSphere, GPU clusters are managed within vCenter to help customers maximizeutilization and protection. The vSphere management framework also brings toolsfor automating your GPU infrastructure:VMware vRealize Suite is a multi-cloud management solution for infrastructureautomation, consistent operations, and governance based on DevOps and MLprinciples. vRealize Suite includes vRealize Automation , vRealizeOperations , vRealize Log Insight , and vRealize Suite Lifecycle Manager .VMware Cloud Foundation is a future-proof hybrid cloud platform formanaging VMs, deploying modern apps and orchestrating containers, all built onfull-stack hyper-converged technology. With a single architecture, VMware CloudFoundation enables consistent, secure infrastructure and operations acrossprivate and public cloud. Cloud Foundation delivers enterprise agility, reliability,and provides consistent infrastructure and operations from the data center to thecloud to the edge, making it an ideal platform for hybrid cloud deployments.vSphere resources can also be created and managed with Ansible , Chef,Puppet , Hashicorp Terraform and other popular CI/CD tools to enableInfrastructure as Code.NVIDIA GPUsThe focus in this paper is on the use of NVIDIA GPUs for accelerating computeworkloads using virtualized GPUs using Virtual Compute Server. Virtual ComputeServer is designed to complement existing GPU virtualization capabilities forgraphics and virtual desktop infrastructure (VDI), by addressing the needs of thedata centers to virtualize compute-intensive workloads such as artificialintelligence (AI), deep learning (DL) and high performance computing (HPC).PowerEdge servers support various NVIDIA GPU models. The GPU modelssupported on Dell EMC PowerEdge servers are available atdelltechnologies.com/accelerators.The GPUs have different compute, memory and power characteristics and aredesigned to accelerate compute-intensive AI and HPC applications. GPU andPowerEdge combinations enable accelerated performance for an intendedworkload and meet the requirements for rack density, cost and power.14 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
GPU CharacteristicsGPU ModelCUDACoresPeak FP32(AI/HPC)RTX6000460815 TFLOPSPeak Mixed-Precision(AI)120 TFLOPSRTX8000460815 TFLOPS120 TFLOPSN/AT425608 TFLOPS65 TFLOPSN/AV100 (PCIe)512014 TFLOPS112 TFLOPS7 TFLOPSV100(SXM2)V100S512015.7TFLOPS16.4TFLOPS125 TFLOPS7.8 TFLOPS130 TFLOPS8.2 TFLOPS5120Peak DR632GBHBM232GBHBM232GBHBM2MemoryBandwidthPower624 GB/s250W624 GB/s250W300 GB/s70W900 GB/s250W900 GB/s300W1134 GB/s250WWe recommend NVIDIA V100, T4 and RTX 6000/8000 GPUs for use with VirtualCompute Server. These graphics accelerators offer a range of features that canmeet the application and workload demands of AI and HPC compute-intensiveapplications that will be deployed in virtualized AI and HPC deployments.NVIDIA V100The NVIDIA V100 GPU will best accelerate high performance computing (HPC)and dedicated AI training workloads. The V100 is equipped with the doubleprecision performance required by various HPC applications such as engineeringsimulation, weather prediction and molecular dynamics. The V100 is alsoequipped with 32GB of memory that can run at 900GB/s to support the memorybandwidth requirements of HPC workloads. The V100S is the latest addition tothe V100 family and can speed up HPC applications with its increased memorybandwidth capability. AI training workloads leverage the processing capability ofmulti-GPUs using scale-out distributed training techniques to improveperformance. Using the V100 SXM2 GPU with the NVLink capabilities enablesdirect communication between GPUs with bandwidth of up to 300GB/s; furtherincreasing performance of AI training workloads.The NVIDIA V100 GPU powered by NVIDIA Volta architecture is the most widelyused accelerator for scientific computing and artificial intelligence. HPC andscientific computing workloads are recommended to use the V100/V100S PCIein R740 (1-3GPUs), R7425(1-3GPUs) and PowerEdge C4140 (4 GPUs). DeepLearning training workloads can leverage NVLink capability of the V100 SXM2GPUs on the C4140 with NVLink capabilities or DSS8440 that support up to 10V100 PCIe GPUs.15 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
Using Virtual Compute Server, the V100 GPUs can be configured as up to 16virtual GPUs and allocated to different VMs. The different virtual GPU profilessupported are show in the table. For VMs requiring compute power of multipleV100 GPUs or multiple V100 virtual GPUs with NVLink, Virtual Compute ServerGPU aggregation capabilities and support for peer-to-peer computing providesnear bare metal performance in a virtualized environment.Virtual Compute Server Supported GPU A Cores640640Tensor Coresn/an/aRT Cores32GB/16GB HBM232GB HBM2Memory4GB,8GB,16GB,4GB,8GB, 16GB,Profiles32GB32GBForm FactorPowerSupported DellEMC ServersPCIe dual slot/SXM2250W/300WR740, R740xd,T640, C4140,R940xa, R7525,DSS8440PCIe dual slot250WR740, R740xd,C4140, R940xa,R7525, DSS8440RTX 6000/8000Turing46085767224GB/48GB GDDR64 GB, 6 GB, 8 GB, 12GB, 24 GB, 48GB(RTX 8000)PCIe dual slot250WR740, R740xd,R7425, R7525,DSS8440T4Turing25603204016GB GDDR64GB, 8GB, 16GBPCIe single slot70WR640, R7515, R740,R740xd, C4140,R7525, DSS8440RTX8000Quadro RTX 6000/8000 will best accelerate performance graphics, render farmsand edge computing workloads. In addition to having high CUDA core counts,memory speeds and floating-point performances, these GPUs have uniquefeatures that make them ideal for graphics, such ray tracing cores and NVLINKcapability for supporting large memory capacities.It is important to remember that the workload dictates which server to choose forbest results. The RTX 6000/8000 supports high-performance graphics workloadsand optimizing this type of workload will require sourcing as many GPUs aspossible into datacenter racks. For this reason, we recommend the DSS8440 asa first option, as it can support up to 10 full-length/height GPUs, followed by theR740 and R7525 as second options, which are commonly used compute nodesin render farms.In a virtualized environment, the RTX 6000/8000 can also be leveraged forcompute-intensive AI and HPC workloads using Virtual Compute Server profiles.One of the advantages of virtualization is that you can run mixed workloads onthe same platform and get better utilization and efficiencies for you computeinfrastructure.16 Virtualizing GPUs in VMware vSphere using NVIDIA Virtual Compute Server on Dell EMC infrastructure 2020 Dell Inc. or its subsidiaries.
NVIDIA T4The NVIDIA T4 GPU will best accelerate AI inference, training, general-purposecompute applications and graphics. The T4 introduced the Turing architecturewith multi-precision computing ranging from FP32/FP16 for floating pointarithmetic to INT8/INT4 integer precision capability to handle diverse workloads.With low power consumption, modest pricing and a low-profile, single-width formfactor, the T4 is both versatile in functionality and easy to integrate into mostPowerEdge servers, making it ideal for accelerating general-purpose workloads.It is an optimized solution for workloads that don’t need high precision (FP64)capabilities.The servers that we recommend populating with T4s are the R640, R740,R740xd and DSS8440. Users can add 1-2x T4 GPUs for inference on R640, 1-6xT4 GPUs on the R740(xd) for more demanding applications and up to 16x T4GPUs on the DSS8440 for applications requiring highly dense GPU computecapability.With Virtual Compute Server, each T4 GPU can be shared between differentVMs and supports three virtual GPU profiles. For VMs requiring compute powerof multiple T4
GPU Virtualization with Virtual Compute Server NVIDIA Virtual Compute Server enables the benefits of VMware virtualization for GPU-accelerated PowerEdge servers. With Virtual Compute Server, data center admins are able to power compute-intensive workloads with GPUs in a virtual machine (VM). Virtual Compute Server software virtualizes NVIDIA .
o VMware vSphere Web Client o DR to the Cloud services Optional Features: o VMware vSphereSDKs o vSphere Virtual Machine File System (VMFS) o vSphere Virtual SMP o vSphere vMotion o vSphere Storage vMotion o vSphere High Availability (HA) o vSphere Distributed Resource Scheduler (DRS) o vSphere Storage DRS o vSphere Fault Tolerance o vSphere .
1 VMware vSphere and the vSphere Web Services SDK 15 . Introduction to vSphere Clusters 219 VMware DRS 219 VMware HA 220 VMware HCI 220 Creating and Configuring Clusters 221 . 17 vSphere Performance 263 vSphere Performance Data Collection 263 PerformanceManager Objects and Methods 265
1 VMware vSphere and the vSphere Web Services SDK 15 . Introduction to vSphere Clusters 220 VMware DRS 220 VMware HA 221 VMware HCI 221 Creating and Configuring Clusters 222 . 17 vSphere Performance 264 vSphere Performance Data Collection 264 PerformanceManager Objects and Methods 266
15. Create and manage a vSphere cluster that is enabled with VMware vSphere High Availability and VMware vSphere 16. Distributed Resource Scheduler 17. Discuss solutions for managing the vSphere life cycle 18. Use VMware vSphere Lifecycle Manager to perform upgrades to ESXi hosts and virtual machines 備註事項 1.
15. Create and manage a vSphere cluster that is enabled with VMware vSphere High Availability and VMware vSphere 16. Distributed Resource Scheduler 17. Discuss solutions for managing the vSphere life cycle 18. Use VMware vSphere Lifecycle Manager to perform upgrades to ESXi hosts and virtual machines 備註事項 1.
CHEAT SHEET 1 / 9 VMware vSphere 4 What is VMware vSphere 4? VMware vSphere 4, the industry’s rst cloud OS Internal Cloud External Cloud VMware vCenter Suite VMware vSphere 4 Application Services VMotion Storage VMotion HA Fault Tolerance Data Recovery vShield Zones VM afe DRS Hot Add Availability Security Scalablity ESX ESXi DRS .
VMware vSphere Basics guide vSphere Installation and Setup guide vSphere Upgrade guide VMware vSphere Examples and Scenarios guide Installing and Administering VMware vSphere Update Manager . Objective 1.4 – Secure vCenter Server and ESXi . Knowledge Identify common vCenter Server privileges and roles
VMware vSphere Basics guide vSphere Installation and Setup guide vSphere Upgrade guide VMware vSphere Examples and Scenarios guide Installing and Administering VMware vSphere Update Manager . Objective 1.4 – Secure vCenter Server and ESXi . Knowledge Identify common vCenter Server privileges and roles
