Virtualizing Computing for AI,
Deep Learning, and Data Science
NVIDIA Virtual Compute Server (vComputeServer) delivers accelerated virtualization of data center servers using GPUs, enabling the most demanding workloads, such as AI, deep learning, and data science, to run on a virtual machine (VM).
Possibilities
GPU Sharing
GPU Sharing (partial) is only possible with NVIDIA vGPU technology. GPU resources can be shared across multiple virtual machines, increasing utilization for lighter workloads that require GPU acceleration.
GPU Teaming
A virtual machine can access the power of multiple GPUs, which is often necessary for compute-intensive workloads. vComputeServer supports both multi-vGPU and peer-to-peer computing. Multi-vGPU does not directly team the GPUs. Peer-to-peer uses NVLink for higher throughput.
Management and Monitoring
vComputeServer provides monitoring support at the application, guest, and server levels. Proactive management enables live migration, pause and resume, and thresholds that reflect usage dynamics, all via the vGPU Management SDK.
NGC
NVIDIA GPU Cloud (NGC) is a registry of GPU-optimized software designed to simplify deep learning, machine learning, and HPC workloads. The registry now supports virtualized environments with NVIDIA vComputeServer.
Peer-to-Peer Computing
NVIDIA® NVLink™ is a high-speed, direct interconnect between GPUs. The technology delivers higher throughput, more connections, and improved scalability for multi-GPU configurations. The connection now supports NVIDIA Virtual GPU (vGPU) technology.
Error Correction Code and Page Removal
Error Correction Code and Page Removal provide higher reliability for computing applications that are sensitive to data corruption. This is especially important in large-scale cluster computing environments where GPUs process very large data sets and/or run applications for long periods.
Request for a preliminary
vCS calculation
Fill out a short questionnaire with questions and we will select the optimal configuration for you!
If your task is beyond the scope of standard solutions, please contact our technical specialists for help:
VASMAXORIENT@GMAIL.COM +852 97058258
NVIDIA vComputeServer
Recommended GPUs
| NVIDIA T4 | NVIDIA V100 (SXM2) | |
|---|---|---|
| RT Cores | 48 | - |
| Tensor cores | 320 | 640 |
| CUDA® Cores | 2560 | 5120 |
| Memory | 16 GB GDDR6 | 32 GB HBM2 |
| FP 16/FP 32 (mixed precision) | 64 Teraflops | 125 Teraflops |
| FP 32 (single precision) | 8.1 Teraflops | 125 Teraflops |
| FP 64 (double precision) | - | 7.8 Teraflops |
| NVLink: GPUs per VM | - | Up to 8 |
| Error correction and page removal code | yes | yes |
| Number of GPUs per virtual machine | Up to 16 | Up to 16 |
Virtualization Partners
Frequently asked questions
How is vComputeServer different from GRID vPC/vApps and Quadro vDWS?
GRID vPC/vApps and Quadro vDWS are client computing solutions for business virtualization and working with graphics and video. vComputeServer is designed for resource-intensive server workloads: AI tasks, deep learning and data processing.
Are the license terms for vComputeServer the same as GRID vPC/vApps and Quadro vDWS?
No, vComputeServer is licensed differently. GRID vPC/vApps and Quadro vDWS are purchased as a perpetual or annual subscription based on the number of concurrent users. Since vComputeServer is designed for server workloads, the license is tied to the GPU, not the user. vComputeServer is purchased as an annual subscription based on the GPU license. For more information on licensing, see the NVIDIA Virtual GPU Packages, Pricing, and Licensing guide.
Which NVIDIA GPUs are supported by vComputeServer?
The recommended NVIDIA GPUs for vComputeServer are NVIDIA V100 and NVIDIA T4. Quadro RTX™ 6000 and RTX 8000, and Pascal-based graphics cards: NVIDIA P40, P100, and P6 are also supported.
Which servers are certified to run vComputeServer?
A full list of certified servers for all vGPU products can be found here.
Can I use containers with vComputeServer?
Yes, you can run containers on VMs with vComputeServer. NVIDIA NGC provides a comprehensive catalog of GPU-accelerated containers for deep learning, machine learning, and HPC. You can also run workloads directly on a VM without containers using vComputeServer.