DOCSLIB.ORG

GPU Virtualization on Vmware's Hosted I/O Architecture

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
GPU Virtualization on Vmware's Hosted I/O Architecture GPU Virtualization on VMware’s Hosted I/O Architecture Micah Dowty, Jeremy Sugerman VMware, Inc. 3401 Hillview Ave, Palo Alto, CA 94304 [email protected], [email protected] Abstract more computational performance than CPUs. At the Modern graphics co-processors (GPUs) can produce same time, GPU acceleration has extended beyond en- high fidelity images several orders of magnitude faster tertainment (e.g., games and video) into the basic win- than general purpose CPUs, and this performance expec- dowing systems of recent operating systems and is start- tation is rapidly becoming ubiquitous in personal com- ing to be applied to non-graphical high-performance ap- puters. Despite this, GPU virtualization is a nascent field plications including protein folding, financial modeling, of research. This paper introduces a taxonomy of strate- and medical image processing. The rise in applications gies for GPU virtualization and describes in detail the that exploit, or even assume, GPU acceleration makes specific GPU virtualization architecture developed for it increasingly important to expose the physical graph- VMware’s hosted products (VMware Workstation and ics hardware in virtualized environments. Additionally, VMware Fusion). virtual desktop infrastructure (VDI) initiatives have led We analyze the performance of our GPU virtualiza- many enterprises to try to simplify their desktop man- tion with a combination of applications and microbench- agement by delivering VMs to their users. Graphics vir- marks. We also compare against software rendering, the tualization is extremely important to a user whose pri- GPU virtualization in Parallels Desktop 3.0, and the na- mary desktop runs inside a VM. tive GPU. We find that taking advantage of hardware GPUs pose a unique challenge in the field of virtu- acceleration significantly closes the gap between pure alization. Machine virtualization multiplexes physical emulation and native, but that different implementations hardware by presenting each VM with a virtual device and host graphics stacks show distinct variation. The mi- and combining their respective operations in the hyper- crobenchmarks show that our architecture amplifies the visor platform in a way that utilizes native hardware overheads in the traditional graphics API bottlenecks: while preserving the illusion that each guest has a com- draw calls, downloading buffers, and batch sizes. plete stand-alone device. Graphics processors are ex- Our virtual GPU architecture runs modern graphics- tremely complicated devices. In addition, unlike CPUs, intensive games and applications at interactive frame chipsets, and popular storage and network controllers, rates while preserving virtual machine portability. The GPU designers are highly secretive about the specifi- applications we tested achieve from 86% to 12% of na- cations for their hardware. Finally, GPU architectures tive rates and 43 to 18 frames per second with VMware change dramatically across generations and their gener- Fusion 2.0. ational cycle is short compared to CPUs and other de- vices. Thus, it is nearly intractable to provide a virtual 1 Introduction device corresponding to a real modern GPU. Even start- Over the past decade, virtual machines (VMs) have be- ing with a complete implementation, updating it for each come increasingly popular as a technology for multi- new GPU generation would be prohibitively laborious. plexing both desktop and server commodity x86 com- Thus, rather than modeling a complete modern GPU, puters. Over that time, several critical challenges in our primary approach paravirtualizes: it delivers an ide- CPU virtualization were solved and there are now both alized software-only GPU and our own custom graphics software and hardware techniques for virtualizing CPUs driver for interfacing with the guest operating system. with very low overheads [1]. I/O virtualization, how- The main technical contributions of this paper are (1) ever, is still very much an open problem and a wide a taxonomy of GPU virtualization strategies—both emu- variety of strategies are used. Graphics co-processors lated and passthrough-based, (2) an overview of the vir- (GPUs) in particular present a challenging mixture of tual graphics stack in VMware’s hosted architecture, and broad complexity, high performance, rapid change, and (3) an evaluation and comparison of VMware Fusion’s limited documentation. 3D acceleration with other approaches. We find that a Modern high-end GPUs have more transistors, draw hosted model [2] is a good fit for handling complicated, more power, and offer at least an order of magnitude rapidly changing GPUs while the largely asynchronous Published in the USENIX Workshop on I/O Virtualization 2008 1 graphics programming model is still able efficiently to ware, formerly fixed-function transformation and shad- utilize GPU hardware acceleration. ing has become generally programmable. Graphics ap- The rest of this paper is organized as follows. Sec- plications use high-level Application Programming In- tion 2 provides background and some terminology. Sec- terfaces (APIs) to configure the pipeline, and provide tion 3 describes a taxonomy of strategies for exposing shader programs which perform application specific GPU acceleration to VMs. Section 4 describes the de- per-vertex and per-pixel processing on the GPU [13]. vice emulation and rendering thread of the graphics vir- Future GPUs are expected to continue providing in- tualization in VMware products. Section 5 evaluates the creased programmability. Intel recently announced its 3D acceleration in VMware Fusion. Section 6 summa- Larrabee [14] architecture, a potentially disruptive tech- rizes our findings and describes potential future work. nology which follows this trend to its extreme. With the recent exception of many AMD GPUs, for 2 Background which open documentation is now available [15], GPU While CPU virtualization has a rich research and com- hardware is proprietary. NVIDIA’s hardware documen- mercial history, graphics hardware virtualization is a rel- tation, for example, is a closely guarded trade secret. atively new area. VMware’s virtual hardware has al- Nearly all graphics applications interact with the GPU ways included a display adapter, but it initially included via a standardized API such as Microsoft’s DirectX or only basic 2D support [3]. Experimental 3D support the vendor-independent OpenGL standard. did not appear until VMware Workstation 5.0 (April 2005). Both Blink [4] and VMGL [5] used a user-level 3 GPU Virtualization Taxonomy Chromium-like approach [6] to accelerate fixed function This section explores the GPU virtualization approaches OpenGL in Linux and other UNIX-like guests. Parallels we have considered at VMware. We use four primary Desktop 3.0 [7] accelerates some OpenGL and Direct3D criteria for judging them: performance, fidelity, multi- guest applications with a combination of Wine and pro- plexing, and interposition. The former two emphasize prietary code [8], but loses its interposition while those minimizing the cost of virtualization: users desire native applications are running. Finally, at the most recent Intel performance and full access to the native hardware fea- Developer Forum, Parallels presented a demo that ded- tures. The latter two emphasize the added value of virtu- icates an entire native GPU to a single virtual machine alization: virtualization is fundamentally about enabling using Intel’s VT-d [9, 10]. many virtual instances of one physical entity and then The most immediate application for GPU virtualiza- hopefully using that abstraction to deliver secure isola- tion is to desktop virtualization. While server workloads tion, resource management, virtual machine portability, still form the core use case for virtualization, desktop and many other features enabled by insulating the guest virtualization is now the strongest growth market [11]. from physical hardware dependencies. Desktop users run a diverse array of applications, in- We observe that different use cases weight the crite- cluding entertainment, CAD, and visualization software. ria differently—for example a VDI deployment values Windows Vista, Mac OS X, and recent Linux distribu- high VM-to-GPU consolidation ratios (e.g., multiplex- tions all include GPU-accelerated windowing systems. ing) while a consumer running a VM to access a game or Furthermore, an increasing number of ubiquitous appli- CAD application unavailable on his host values perfor- cations are adopting GPU acceleration. Adobe Flash mance and likely fidelity. A tech support person main- Player 10, the next version of a product which currently taining a library of different configurations and an IT ad- reaches 99.0% of Internet viewers [12], will include ministrator running server VMs are both likely to value GPU acceleration. There is a user expectation that vir- portability and secure isolation (interposition). tualized applications will “just work”, and this increas- Since these criteria are often in opposition (e.g., per- ingly includes having access to their graphics card. formance at the expense of interposition), we describe several possible designs. Rather than give an exhaustive 2.1 GPU Hardware list, we describe points in the design space which high- This section will briefly introduce GPU hardware. It is light interesting trade-offs and capabilities. At a high not within the scope of this paper to provide a full dis- level, we group them into two categories: front-end (ap- cussion of GPU architecture and programming models. plication
Load more

Recommended publications
  • A Virtual Machine Environment for Real Time Systems Laboratories
    AC 2007-904: A VIRTUAL MACHINE ENVIRONMENT FOR REAL-TIME SYSTEMS LABORATORIES Mukul Shirvaikar, University of Texas-Tyler MUKUL SHIRVAIKAR received the Ph.D. degree in Electrical and Computer Engineering from the University of Tennessee in 1993. He is currently an Associate Professor of Electrical Engineering at the University of Texas at Tyler. He has also held positions at Texas Instruments and the University of West Florida. His research interests include real-time imaging, embedded systems, pattern recognition, and dual-core processor architectures. At the University of Texas he has started a new real-time systems lab using dual-core processor technology. He is also the principal investigator for the “Back-To-Basics” project aimed at engineering student retention. Nikhil Satyala, University of Texas-Tyler NIKHIL SATYALA received the Bachelors degree in Electronics and Communication Engineering from the Jawaharlal Nehru Technological University (JNTU), India in 2004. He is currently pursuing his Masters degree at the University of Texas at Tyler, while working as a research assistant. His research interests include embedded systems, dual-core processor architectures and microprocessors. Page 12.152.1 Page © American Society for Engineering Education, 2007 A Virtual Machine Environment for Real Time Systems Laboratories Abstract The goal of this project was to build a superior environment for a real time system laboratory that would allow users to run Windows and Linux embedded application development tools concurrently on a single computer. These requirements were dictated by real-time system applications which are increasingly being implemented on asymmetric dual-core processors running different operating systems. A real time systems laboratory curriculum based on dual- core architectures has been presented in this forum in the past.2 It was designed for a senior elective course in real time systems at the University of Texas at Tyler that combines lectures along with an integrated lab.
    [Show full text]
  • KVM Based Virtualization and Remote Management Srinath Reddy Pasunuru St
    St. Cloud State University theRepository at St. Cloud State Culminating Projects in Information Assurance Department of Information Systems 5-2018 KVM Based Virtualization and Remote Management Srinath Reddy Pasunuru St. Cloud State University, [email protected] Follow this and additional works at: https://repository.stcloudstate.edu/msia_etds Recommended Citation Pasunuru, Srinath Reddy, "KVM Based Virtualization and Remote Management" (2018). Culminating Projects in Information Assurance. 53. https://repository.stcloudstate.edu/msia_etds/53 This Starred Paper is brought to you for free and open access by the Department of Information Systems at theRepository at St. Cloud State. It has been accepted for inclusion in Culminating Projects in Information Assurance by an authorized administrator of theRepository at St. Cloud State. For more information, please contact [email protected]. 1 KVM Based Virtualization and Remote Management by Srinath Reddy Pasunuru A Starred Paper Submitted to the Graduate Faculty of St. Cloud State University in Partial Fulfillment of the Requirements for the Degree Master of Science in Information Assurance May, 2018 Starred Paper Committee Susantha Herath, Chairperson Ezzat Kirmani Sneh Kalia 2 Abstract In the recent past, cloud computing is the most significant shifts and Kernel Virtual Machine (KVM) is the most commonly deployed hypervisor which are used in the IaaS layer of the cloud computing systems. The Hypervisor is the one which provides the complete virtualization environment which will intend to virtualize as much as hardware and systems which will include the CPUs, Memory, network interfaces and so on. Because of the virtualization technologies such as the KVM and others such as ESXi, there has been a significant decrease in the usage if the resources and decrease in the costs involved.
    [Show full text]
  • Demystifying Desktop Virtualization
    Proceedings of the 9th WSEAS International Conference on APPLIED COMPUTER SCIENCE Demystifying desktop virtualization TOMISLAV PETROVIĆ Vetropack Straža d.d. Hum na Sutli Hum na Sutli 203, 49231 Hum na Sutli CROATIA [email protected] http://www.vetropack.hr KREŠIMIR FERTALJ University of Zagreb Faculty of Electrical Engineering and Computing, Unska 3 CROATIA [email protected] http://www.fer.hr Abstract: - This paper presents concepts of virtualization techniques. Both server virtualization and desktop virtualization are briefly presented, but desktop virtualization is discussed in more details. As the size of computer systems grows every day with new applications, and with new client and market requirements, it is obvious that something must be done to ease the administration and effectively secure data. Additionally, disaster recovery plans must be improved too. Virtualization has solved many of these requirements and has become more and more popular so that many companies decided to implement a virtual server, virtual storage or/and virtual desktops. The paper lists the reasons why to implement virtual desktops and servers and what are the main benefits of such implementations. Virtual environments the also have some downsides, which are presented, explained and compared to the advantages. Key-Words: - desktop virtualization, virtual machine, infrastructure client, connection manager, thin client 1 Introduction access, etc. Manual administration of many separate computers is In this paper we describe the concept of virtualization, hard, tedious and time consuming job. A variety of use of desktop virtualization within leading companies, choices exists for installing and reinstalling of different discuss reasons for its usage and give main advantages, software and hardware by a user at a local computer.
    [Show full text]
  • Gscale: Scaling up GPU Virtualization with Dynamic Sharing of Graphics
    gScale: Scaling up GPU Virtualization with Dynamic Sharing of Graphics Memory Space Mochi Xue, Shanghai Jiao Tong University and Intel Corporation; Kun Tian, Intel Corporation; Yaozu Dong, Shanghai Jiao Tong University and Intel Corporation; Jiacheng Ma, Jiajun Wang, and Zhengwei Qi, Shanghai Jiao Tong University; Bingsheng He, National University of Singapore; Haibing Guan, Shanghai Jiao Tong University https://www.usenix.org/conference/atc16/technical-sessions/presentation/xue This paper is included in the Proceedings of the 2016 USENIX Annual Technical Conference (USENIX ATC ’16). June 22–24, 2016 • Denver, CO, USA 978-1-931971-30-0 Open access to the Proceedings of the 2016 USENIX Annual Technical Conference (USENIX ATC ’16) is sponsored by USENIX. gScale: Scaling up GPU Virtualization with Dynamic Sharing of Graphics Memory Space Mochi Xue1,2, Kun Tian2, Yaozu Dong1,2, Jiacheng Ma1, Jiajun Wang1, Zhengwei Qi1, Bingsheng He3, Haibing Guan1 {xuemochi, mjc0608, jiajunwang, qizhenwei, hbguan}@sjtu.edu.cn {kevin.tian, eddie.dong}@intel.com [email protected] 1Shanghai Jiao Tong University, 2Intel Corporation, 3National University of Singapore Abstract As one of the key enabling technologies of GPU cloud, GPU virtualization is intended to provide flexible and With increasing GPU-intensive workloads deployed on scalable GPU resources for multiple instances with high cloud, the cloud service providers are seeking for practi- performance. To achieve such a challenging goal, sev- cal and efficient GPU virtualization solutions. However, eral GPU virtualization solutions were introduced, i.e., the cutting-edge GPU virtualization techniques such as GPUvm [28] and gVirt [30]. gVirt, also known as GVT- gVirt still suffer from the restriction of scalability, which g, is a full virtualization solution with mediated pass- constrains the number of guest virtual GPU instances.
    [Show full text]
  • Crane: Fast and Migratable GPU Passthrough for Opencl Applications
    Crane: Fast and Migratable GPU Passthrough for OpenCL Applications James Gleeson, Daniel Kats, Charlie Mei, Eyal de Lara University of Toronto Toronto, Canada {jgleeson,dbkats,cmei,delara}@cs.toronto.edu ABSTRACT (IaaS) providers now have virtualization offerings with ded- General purpose GPU (GPGPU) computing in virtualized icated GPUs that customers can leverage in their guest vir- environments leverages PCI passthrough to achieve GPU tual machines (VMs). For example, Amazon Web Services performance comparable to bare-metal execution. However, allows VMs to take advantage of NVIDIA’s high-end GPUs GPU passthrough prevents service administrators from per- in EC2 [6]. Microsoft and Google recently introduced simi- forming virtual machine migration between physical hosts. lar offerings on Azure and GCP platforms respectively [34, Crane is a new technique for virtualizing OpenCL-based 21]. GPGPU computing that achieves within 5:25% of pass- Virtualized GPGPU computing leverages modern through GPU performance while supporting VM migra- IOMMU features to give the VM direct use of the GPU tion. Crane interposes a virtualization-aware OpenCL li- through DMA and interrupt remapping – a technique which brary that makes it possible to reclaim and subsequently is usually called PCI passthrough [17, 50]. By running reassign physical GPUs to a VM without terminating the the GPU driver in the context of the guest operating guest or its applications. Crane also enables continued GPU system, PCI passthrough bypasses the hypervisor and operation while the VM is undergoing live migration by achieves performance that is comparable to bare-metal OS transparently switching between GPU passthrough opera- execution [47]. tion and API remoting.
    [Show full text]
  • A Full GPU Virtualization Solution with Mediated Pass-Through
    A Full GPU Virtualization Solution with Mediated Pass-Through Kun Tian, Yaozu Dong, David Cowperthwaite Intel Corporation Abstract shows the spectrum of GPU virtualization solutions Graphics Processing Unit (GPU) virtualization is an (with hardware acceleration increasing from left to enabling technology in emerging virtualization right). Device emulation [7] has great complexity and scenarios. Unfortunately, existing GPU virtualization extremely low performance, so it does not meet today’s approaches are still suboptimal in performance and full needs. API forwarding [3][9][22][31] employs a feature support. frontend driver, to forward the high level API calls inside a VM, to the host for acceleration. However, API This paper introduces gVirt, a product level GPU forwarding faces the challenge of supporting full virtualization implementation with: 1) full GPU features, due to the complexity of intrusive virtualization running native graphics driver in guest, modification in the guest graphics software stack, and and 2) mediated pass-through that achieves both good incompatibility between the guest and host graphics performance and scalability, and also secure isolation software stacks. Direct pass-through [5][37] dedicates among guests. gVirt presents a virtual full-fledged GPU the GPU to a single VM, providing full features and the to each VM. VMs can directly access best performance, but at the cost of device sharing performance-critical resources, without intervention capability among VMs. Mediated pass-through [19], from the hypervisor in most cases, while privileged passes through performance-critical resources, while operations from guest are trap-and-emulated at minimal mediating privileged operations on the device, with cost. Experiments demonstrate that gVirt can achieve good performance, full features, and sharing capability.
    [Show full text]
  • VIRTUAL DESKTOP INFRASTRUCTURE an Rcpsolution Spotlight on a Joint Effort by Microsoft and Citrix to Bring a VDI Solution to the Market
    0309rcp_Supp.v5 2/10/09 4:53 PM Page C1 SPECIAL PULLOUT SECTION Partner’s Guide to VIRTUAL DESKTOP INFRASTRUCTURE An RCPSolution Spotlight on a joint effort by Microsoft and Citrix to bring a VDI solution to the market. By Scott Bekker ChannelRedmond Partner Project15 2/4/09 3:07 PM Page 1 Project15 2/4/09 3:08 PM Page 1 0309rcp_Supp.v5 2/10/09 4:53 PM Page 2 VIRTUAL DESKTOP INFRASTRUCTURE icrosoft and Citrix Systems Inc. are working together on a joint go-to-market strategy for partners to take Virtual MDesktop Infrastructure (VDI) solutions to the market. VDI is a special subset of desktop virtualization, itself a subset of virtualization in general. While virtualization separates hardware from workers, for users who need access to their work software, Microsoft’s overall virtualization environment from anywhere, including from a non- approaches fit into four broad buckets—server virtu- company-owned PC, as well as enterprise customers alization, desktop virtualization, presentation virtu- with a centralized desktop strategy for office work- alization and application virtualization. ers,” the Microsoft materials explain. That’s not to The biggest part of the virtualization market to date say everyone is appropriate for VDI, even when a fat has been consolidating multiple software servers client isn’t the answer for some reason. onto comparatively fewer hardware servers. “Terminal Services, which has been widely Such server consolidation is still a growth adopted for virtualizing the presentation of entire industry, especially in a down economy, in which desktops or individual applications, is an alternative there are savings to be gained by reducing hardware centralized desktop delivery solution from and power expenses.
    [Show full text]
  • GPU Virtualization on Vmware's Hosted I/O Architecture
    GPU Virtualization on VMware's Hosted I/O Architecture Micah Dowty Jeremy Sugerman VMware, Inc. VMware, Inc. 3401 Hillview Ave, Palo Alto, CA 94304 3401 Hillview Ave, Palo Alto, CA 94304 [email protected] [email protected] ABSTRACT 1. INTRODUCTION Modern graphics co-processors (GPUs) can produce high fi- Over the past decade, virtual machines (VMs) have be- delity images several orders of magnitude faster than general come increasingly popular as a technology for multiplexing purpose CPUs, and this performance expectation is rapidly both desktop and server commodity x86 computers. Over becoming ubiquitous in personal computers. Despite this, that time, several critical challenges in CPU virtualization GPU virtualization is a nascent field of research. This paper have been overcome, and there are now both software and introduces a taxonomy of strategies for GPU virtualization hardware techniques for virtualizing CPUs with very low and describes in detail the specific GPU virtualization archi- overheads [2]. I/O virtualization, however, is still very tecture developed for VMware’s hosted products (VMware much an open problem and a wide variety of strategies are Workstation and VMware Fusion). used. Graphics co-processors (GPUs) in particular present a challenging mixture of broad complexity, high performance, We analyze the performance of our GPU virtualization with rapid change, and limited documentation. a combination of applications and microbenchmarks. We also compare against software rendering, the GPU virtual- Modern high-end GPUs have more transistors, draw more ization in Parallels Desktop 3.0, and the native GPU. We power, and offer at least an order of magnitude more com- find that taking advantage of hardware acceleration signif- putational performance than CPUs.
    [Show full text]
  • Enabling VDI for Engineers and Designers Positioning Information
    Enabling VDI for Engineers and Designers Positioning Information The ability to work remotely has been a critical business need for many years. But with the recent pandemic, the focus has shifted from supporting the occasional business trip or conference attendee, to supporting all non-essential employees. Virtual Desktop Infrastructure (VDI) provides a robust solution for businesses, allowing their employees to access business data on any mobile device – be it a laptop, tablet, or smartphone – without risk of losing that business data if the device is lost or stolen. With VDI the data resides in the data center, behind company firewalls. VDI allows companies to implement flexible work schedules, where employees have the freedom to pick up the kids from school and log on later in the evening to finish up. VDI also provides for centralized management of security fixes and application upgrades, ensuring all workers are using the latest version of applications. Although a great solution for your typical office workers, the benefits of VDI have largely been unavailable to another set of employees. Power users – engineers and designers using 3D or other graphics visualization applications – have generally been left out of these flexible arrangements. The graphics visualization applications they use are processor intensive, and the files they create and modify can be very large. The workstations that run the applications and the high-resolution monitors that display the designs are heavy and thus difficult to move. In order to have decent response times, power users have needed to be connected to a high-speed local area network. Work has been underway for several years to expand the benefits of VDI to the power user, and today solutions exist to support the majority of visualization applications.
    [Show full text]
  • Desktop Virtualization: Not for Every User – but It Is for Every Company
    Desktop Virtualization: Not for every user – but it is for every company Industry researchers are forecasting steady growth in the desktop virtualization market. According to Gartner, “HVD (hosted virtual 1 Source: Gartner, “Emerging desktop) deployments are forecast to reach 74 million users by Technology Analysis: Servers 1 Deployed to Support Hosted Virtual 2014 or 15% of professional desktop users.” in addition, Gartner Desktops,” 18 March 2010. also states, “while the installed base is relatively small today, 2 Source: Gartner, “Hosted Virtual Gartner has spoken to clients with deployment plans that include Desktops Are the Catalyst Behind Changing How Organizations 2 tens of thousands of employees using HVDs in an organization.” Manage Storage,” October 2009. Tech Dossier If you think desktop virtualization should only be discussed in theoretical terms, or put on the back burner because it’s not ready for today’s environ- ment, think again. Desktop virtualization products and services are viable and available; it behooves IT and business executives to begin exploring how the technology can transform the way people work. Desktop virtualization may well turn out to be the most disruptive technology innovation since server virtualization in the 1990’s and even perhaps since the advent of the personal computer itself. Decision makers should look at it as a new way of computing that leads to enhanced IT controls, improved intellectual property management and substantial employee productivity improvements. Industry researchers are forecasting broad economic impact on organiza- Desktop solution with 15 desktops us- steady growth in the desktop virtual- tions. As with most organizations the ing Dell™ OptiPlex™ flexible comput- ization market.
    [Show full text]
  • Virtual GPU Software User Guide Is Organized As Follows: ‣ This Chapter Introduces the Capabilities and Features of NVIDIA Vgpu Software
    Virtual GPU Software User Guide DU-06920-001 _v13.0 Revision 02 | August 2021 Table of Contents Chapter 1. Introduction to NVIDIA vGPU Software..............................................................1 1.1. How NVIDIA vGPU Software Is Used....................................................................................... 1 1.1.2. GPU Pass-Through.............................................................................................................1 1.1.3. Bare-Metal Deployment.....................................................................................................1 1.2. Primary Display Adapter Requirements for NVIDIA vGPU Software Deployments................2 1.3. NVIDIA vGPU Software Features............................................................................................. 3 1.3.1. GPU Instance Support on NVIDIA vGPU Software............................................................3 1.3.2. API Support on NVIDIA vGPU............................................................................................ 5 1.3.3. NVIDIA CUDA Toolkit and OpenCL Support on NVIDIA vGPU Software...........................5 1.3.4. Additional vWS Features....................................................................................................8 1.3.5. NVIDIA GPU Cloud (NGC) Containers Support on NVIDIA vGPU Software...................... 9 1.3.6. NVIDIA GPU Operator Support.......................................................................................... 9 1.4. How this Guide Is Organized..................................................................................................10
    [Show full text]
  • With NVIDIA Virtual GPU
    Virtualizing AI/DL platform: NVIDIA Virtual GPU Solution Doyoung Kim, Sr. Solution Architect NVIDIA Virtual GPU Solution Support, Updates & Maintenance NVIDIA Virtual GPU Software Tesla Datacenter GPUs How it works? CPU Only VDI With NVIDIA Virtual GPU Apps and VMs NVIDIA Graphics Drivers Apps and VMs NVIDIA Virtual GPU Hypervisor NVIDIA virtualization software Hypervisor Server NVIDIA Tesla GPU Server Evolution of Virtual GPU Live Migration Ultra High End Simulation, NGC Support Photo Realism, 3D Rendering, AL/DL Business Designers, User Architects, Engineers 2013 2016 2017 Today GPU Compute on Virtual GPU Optimized DL Container from NGC Virtual Machine Virtual Quadro GPU CUDA & OpenCL enabled NVIDIA Virtualization Software Hypervisor NVIDIA Tesla GPU Server Virtual GPU can…. ✓ Run GPU compute workload - Any CUDA / OpenCL applications - Requires NVIDIA Virtual GPU 5.0 or higher - Requires Pascal or later GPU ✓ Be fully integrated with Virtualization solution - Live migration support from Virtual GPU 6.0 or higher - Support Cluster / Host / VM / Application level performance monitoring - Enabled for all major solutions such as vSphere / XenServer / KVM ✓ Provide every level of AI / DL Platforms - From desktop level to multi-GPU server - Fully support NVIDIA GPU Cloud (NGC) - Support up to 4 multi-vGPU per Virtual Machine* * Requires Virtual GPU 7.0 or higher / RHEL KVM only Why Virtualization? Which is best for AI/DL research system? PC with Consumer GPU GPU Server w/o Virtualization NVIDIA GPU Virtualization Manageability Require MGMT
    [Show full text]