Desktop virtualization with vGPU support

Vasmax vPod is an integrated hardware and software suite designed to deploy and manage virtual infrastructure using vGPU. It allows customers to quickly and efficiently implement virtual desktops, reduce infrastructure costs, and ensure system stability. Vasmax specialists provide support at all stages - from planning and creating an architecture to setting up and scaling.

In the context of modern requirements for import substitution, Vasmax vPod is designed to provide an easier transition to Russian solutions with the help of: Rosa Virtualization, Brest, Space VDI, zVirt, Termidesk. These solutions provide a reliable and secure alternative to foreign VDI systems, fully meeting the requirements of Russian customers.

Consultation needed

Schematic diagram of the solution

Hardware implementation:

The hardware complex consists of Vasmax EGX computing servers, infrastructure servers, network equipment and data storage systems.

Software implementation:

Russian software for virtualization management provides control over software hypervisors, virtual machines and data center infrastructure in a network environment.

Integration work -
Vasmax vGPU Professional Services:

At this stage, a full range of integration work is carried out, starting with development of the system architecture, which, during preliminary testing, is supplemented by an accurate definition of the required computing power and vGPU profiles (sizing). After delivery and installation of the equipment in the customer's infrastructure, final optimization and configuration of the system is performed to ensure its maximum performance and efficiency, providing a comprehensive and functional solution for the client's tasks.

Description

What is vGPU technology?

There are three main options for running a virtual machine with a GPU:

GPU pass-through: 1:1 dedicated GPU per VM

The most productive mode of operation. NVIDIA CUDA, DirectX 9,10,11, OpenGL 4.4 are fully operational. All other components (processors, memory, storage, network adapters) are virtualized and divided between hypervisor instances, but one GPU remains one GPU. Each virtual machine gets its own GPU with virtually no performance loss.

An obvious limitation is that the number of such virtual machines is limited by the number of available graphics adapters in the system.

Shared GPU: Software virtualization of GPU

Implementation via Microsoft RemoteFX: This option relies on the capabilities of VDI software, the virtual machine works as if with a dedicated adapter and the server GPU also believes that it works with one host, although in fact this is an abstraction layer. The hypervisor intercepts API calls and translates all commands, rendering requests, etc. before passing them to the graphics driver, and the host machine works with the virtual card driver.

Shared GPU is a reasonable solution for many cases. If the applications are not too complex, then a significant number of users can work simultaneously.

Virtual GPU: Hardware virtualization (HW&SW)

The most advanced option for sharing GPUs between users, when the resources of a physical GPU are shared between virtual machines and can be configured with specific profiles, while each VM works with an isolated vGPU, unlike the Shared GPU mode

Supported profiles

Virtual PC

This profile is designed to provide efficient virtualization of PCs and applications, providing high performance and graphics quality. Suitable for office applications, web browsers and multimedia content.

Virtual Workstation

The Virtual Workstation profile provides powerful graphics performance for professionals working with 3D modeling, animation, video editing and design. The solution allows you to use the most demanding graphics applications, such as Autodesk, Siemens NX, SOLIDWORKS and others.

Server Application Virtualization (vCompute)

This profile is designed to optimize server application virtualization, providing fast data processing and graphics tasks. Suitable for big data processing, machine learning and artificial intelligence.

Depending on the profile and performance requirements, different GPU models can be used, such as NVIDIA A10, A16, L40. The choice of the optimal profile and hardware components depends on the needs of your project, the applications and tasks used.

VM configurations and user density per server

User type Office workers Design engineers* Senior designers Designers
Applications MS Office, WEB browsers,
Trueconf, Zoom
Kompas 3D, AutoCad,
NanoCad, Siemens NX
+ SolidWorks
ProEngineer, Catia
3DS Max, Cinema 4D,
Maya, Photoshop
Number of
users on
server
32-64 12-24 9-12 3-6
Configurations
VM (min)
vCPU - 2 cores
vRAM - 6GB
vHDD - 100GB
vGPU - NVIDIA A16-1Q
vCPU - 4 cores
vRAM - 16GB
vHDD - 200GB
vGPU - NVIDIA A40-4Q
vCPU - 8 cores
vRAM - 24GB
vHDD - 200GB
vGPU - NVIDIA L40-12Q
vCPU - 12 cores
vRAM - 32GB
vHDD - 200GB
vGPU - NVIDIA L40-24Q
Memory Bandwidth 4x 200 GB/s 864 GB/s 2039 GB/s 960 GB/s
Cuda Cores 4x 1280 Cores 18176  Cores 14592 Cores 18176  Cores
Maximum power consumption 250W 300W 350W 300W
NVEnc/NVDec 4 NVENC/8 NVDEC (includes AV1 decode) 3 NVENC/3 NVDEC (includes AV1 decode) 7 NVENC/7 NVDEC (includes AV1 decode) 3 NVENC/3 NVDEC (includes AV1 decode)
vGPU support vPC,vWS,vCS vPC,vWS,vCS vCS (MIG) vPC,vWS,vCS
MIG Support NO NO YES NO
FP32 Performance 4x 4.5 TFLOPS 90.5 TFLOPS 51 TFLOPS 91.1 TFLOPS
INT8 Performance 4x 35.9 TFLPOS 362 TFLOPS 3026 TFLOPS 365 TFLOPS
INT4 Performance no data 724 TFLOPS no data 729 TFLOPS

* Example of a configuration of a computing node providing an organization with 24 workstations of the profile "Design Engineer": Vasmax EGX Server Dual CPU AMD EPYC 9374F 32core @3.85GHz /512GB RAM/2 GPU Nvidia A40 48GB RAM/512GB NVME/10G Network.

Advantages

Exceptional graphics experience

Exceptional user experience with support for both graphics and compute workloads on vGPU

Predictable performance
 

Consistent performance and guaranteed quality of service both on the local network and in the cloud

Maximum user density

The maximum user density solution supports up to 24 virtual desktops on a single physical GPU.

Implementation

Software implementation

deep-learning.jpg

ROSA VIRTUALIZATION

ROSA Virtualization is a full-featured platform based on a Type 1 hypervisor, allowing you to deploy a corporate virtualized data center out of the box in just a few hours. The Russian-language graphical interface allows you to manage hypervisors, virtual machines, storage, clusters and other objects of the virtualization environment from a single center.

The system supports up to several thousand VMs and can be used not only in "private" data centers, but also in data centers of government agencies or enterprises. The presence of developed discretionary and role-based access models distinguishes ROSA Virtualization from similar solutions, for example, based on OpenStack. In 2018, ROSA Virtualization received a certificate from the FSTEC of Russia, confirming the possibility of its use in government information systems, including those working with personal data.

Features of the ROSA VIRTUALIZATION platform

VERSATILE

  • Unified solution for server virtualization and VDI
  • Support for network storage iSCSI, FC, NFS, GlusterFS
  • FreeIPA directory service, AD integration

MIGRATION AND COMPATIBILITY

  • Optimal replacement for VMware vSphere, Microsoft Hyper-V
  • Built-in automated migration service
  • VM conversion from Hyper-V and VMware vCenter
  • Support for Windows 2016/2019 and Linux 5.x guest OS

MANAGEABILITY

  • Live migration of VMs and their hard drives
  • Dynamic resource management
  • Rest API
  • Management portal, reporting

Technical Specifications

  • Type 1 hypervisor
  • Support for x86 architecture up to 384 VP and up to 4 TB RAM

«ALL IN ONE»

  • Virtualization Environment Management System
  • Distributed Storage
  • Directory Service

RELIABILITY

  • Authentication and delimitation of rights
  • Deploying a VM system in fail-safe mode

Virtual Desktop Delivery Protocol:
Loudplay

A software client application that allows users to connect to their virtual machine from any device of their choice, including Windows, macOS, iOS, Linux, Chrome, and Android. The protocol provides stable and high-quality desktop transmission in 4K 60 FPS resolution

Loudplay Features

  • Works with video data redundancy;
  • Has fault-tolerant management;
  • Auto bitrate;
  • Hardware decoding of the video stream;

Hardware implementation

Network switch
Management server
Vasmax EGX Compute Nodes
Data storage system

Vasmax Compute Server(s):

GPU server(s) that host compute resources for virtual desktops.

Vasmax Management Server:

Performs the role of Active Directory (AD), VDI broker, system monitoring and other management functions related to user authorization and distribution of compute resources.

Data Storage Systems (DSS):

Provides safe and reliable data storage, quick access to it and the ability to recover in case of failures.

Network components:

Provide fast interconnection between all system nodes, including the management server, computing servers and storage systems.

Vasmax EGX Servers for Virtualization

The Vasmax EGX platform delivers accelerated computing in data centers, at the edge, and beyond with optimized hardware, an easy-to-deploy software stack, and management software.

Vasmax VGX 2040A-16

Vasmax EGX 2040A-16 Server

A server optimized for deploying virtual desktops Virtual Desktop (VDI) for office workers who need modern applications & multimedia tools for their work.

Technical specifications:

  • Processor: 2x AMD EPYC™ 7763 64 Core 2.45GHz up to 3.5GHz
  • RAM: 512GB DDR4 RDIMM (16 x 32GB ECC REG)
  • Coprocessor: Nvidia A16 (4x 16GB GDDR6)
  • GPU memory: 64GB GDDR6

Virtual machine feature options:

  • 64 user workstations with   configuration

    2 vCPU 6GB vRAM Nvidia vGPU 1GB (Nvidia A16-1Q profile)

    OR
  • 32 user workstations with   configuration

    4 vCPU 12GB vRAM Nvidia vGPU 2GB (Nvidia A16-2Q profile)

Vasmax VGX 2040A-L40

Vasmax EGX 2040A-L40 Server

A server optimized for deploying virtual workstations (NVIDIA RTX Virtual Workstation (vWS)) for creative and technical professionals using professional CAD applications.

Technical specifications:

  • Processor: 2x AMD EPYC™ 75F3 32 Core 2.95GHz up to 4.0GHz
  • RAM: 512GB DDR4 RDIMM (16 x 32GB ECC REG)
  • Coprocessor: 2 x Nvidia L40 48GB GDDR6
  • GPU memory: 96GB GDDR6

Virtual machine specification options:

  • 24 designer workstations (small assemblies and projects)

    4 vCPU 20GB vRAM Nvidia vGPU 4GB

    OR
  • 12 designer workstations (medium assemblies and projects)

    6 vCPU 32GB vRAM Nvidia vGPU 8GB

    OR
  • 8 designer workstations (large assemblies and projects)

    8 vCPU 60GB vRAM Nvidia vGPU 12GB

    OR
  • 6 3D designer workstations

    10 vCPU 80GB vRAM Nvidia vGPU 16GB

Nvidia GPUs for Virtualization

Tech. parameters Nvidia Tesla A16 Nvidia Tesla L40 Nvidia Tesla A10 Nvidia Quadro A6000 Ada Lovelace
GPU NVIDIA Ampere architecture NVIDIA Ada Lovelace architecture NVIDIA Ampere architecture NVIDIA Ada Lovelace architecture
Tire type PCIe Gen4 (x16) PCIe Gen4 (x16) PCIe Gen4 (x16) PCIe Gen4 (x16)
Memory Capacity 4x 16 GB GDDR6 48 GB GDDR6 384 GDDR6 48 GB GDDR6
Memory Bandwidth 4x 200 GB/s 864 GB/s 600 GB/s 960 GB/s
Cuda kernels 4x 1280 Cores 18176  Cores 9216 Cores 18176  Cores
Maximum power consumption 250W 300W 150W 300W
NVEnc/NVDec 4 NVENC/8 NVDEC (includes AV1 decode) 3 NVENC/3 NVDEC (includes AV1 decode) 1 NVENC/2 NVDEC (includes AV1 decode) 3 NVENC/3 NVDEC (includes AV1 decode)
vGPU support vPC,vWS,vCS vPC,vWS,vCS vPC, vWS vPC,vWS,vCS
MIG support NO NO NO NO
FP32 performance 4x 4.5 TFLOPS 90.5 TFLOPS 31.24 TFLOPS 91.1 TFLOPS
INT8 Performance 4x 35.9 TFLPOS 362 TFLOPS 208 TFLOPS 365 TFLOPS
INT4 Performance no data 724 TFLOPS 500 729 TFLOPS

Vasmax vGPU Professional Services (FPS)

This is a set of services that reduces the time and effort for deploying and operating VDI with vGPU support.

It is focused on deep integration of specialists at the planning stage, which allows for a detailed assessment and configuration of the environment for optimal vGPU operation. FPS includes comprehensive support in selecting configurations, testing and recommendations for hardware and software components, as well as a comprehensive assessment and configuration of the infrastructure, detailed planning, selection of configurations, local and remote testing, equipment installation and commissioning, as well as support for the transition to domestic solutions.

Request a test


If you have doubts about which virtual desktop solution (VDI) is better to choose and whether it is right for you, preliminary testing will help you make the right decision. We offer free remote access to the Vasmax EGX server, where you can determine the optimal settings and test various options based on domestic products.

Any questions?
Contact our technical specialist

If you have any questions about the specification or delivery of our servers, you can use the request form or call us at: 8 800 500-69-93.

By clicking the "Submit" button, you confirm your consent to the processing of personal data