ThinkSystem NVIDIA A100 PCIe 4.0 GPU

NVIDIA A100 Tensor Core GPU deliver outstanding acceleration and flexibility to power the world’s highest-performing elastic data centers for AI, data analytics, and HPC applications. As the engine of the NVIDIA data center platform, A100 provides up to 20X higher performance over V100 GPUs and can efficiently scale up to thousands of GPUs, or be partitioned into seven isolated GPU instances with new Multi-Instance GPU (MIG) capability to accelerate workloads of all sizes.

The third-generation Tensor Core technology supports a broad range of math precisions providing a unified workload accelerator for data analytics, AI training, AI inference, and HPC. Accelerating both scale-up and scale-out workloads on one platform enables elastic data centers that can dynamically adjust to shifting application workload demands. This simultaneously boosts throughput and drives down the cost of data centers.

Figure 1. ThinkSystem NVIDIA A100 PCIe 4.0 Passive GPU

The NVIDIA A100 is available in both double-wide PCIe adapter form factor and in SXM form factor. SXM is used in Lenovo's Neptune direct-water-cooled ThinkSystem SD650-N V2 server for the ultimate in GPU performance and heat management.

The following table shows the part numbers for the A100 GPUs.

The PCIe option part numbers includes the following:

• One NVIDIA A100 GPU with full-height (3U) adapter bracket attached
• Documentation

GPUs without a CEC chip: The NVIDIA A100 GPU is offered without a CEC chip (look for "w/o CEC" in the name). The CEC is a secondary Hardware Root of Trust (RoT) module that provides an additional layer of security, which can be used by customers who have high regulatory requirements or high security standards. NVIDIA uses a multi-layered security model and hence the protection offered by the primary Root of Trust embedded in the GPU is expected to be sufficient for most customers. The CEC defeatured products still offer Secure Boot, Secure Firmware Update, Firmware Rollback Protection, and In-Band Firmware Update Disable. Specifically, without the CEC chip, the GPU does not support Key Revocation or Firmware Attestation. CEC and non-CEC GPUs of the same type of GPU can be mixed in field upgrades.

The following figure shows the NVIDIA HGX A100 4-GPU Board in the water-cooled ThinkSystem SD650-N V2 server

Figure 2. NVIDIA HGX A100 4-GPU Board in the water-cooled ThinkSystem SD650-N V2 server

The NVIDIA A100 Tensor Core GPU delivers unprecedented acceleration—at every scale—to power the world’s highest-performing elastic data centers for AI, data analytics, and high-performance computing (HPC) applications. As the engine of the NVIDIA data center platform, A100 provides up to 20X higher performance over the prior NVIDIA Volta™ generation. A100 can efficiently scale up or be partitioned into seven isolated GPU instances with Multi-Instance GPU (MIG), providing a unified platform that enables elastic data centers to dynamically adjust to shifting workload demands.

NVIDIA A100 Tensor Core technology supports a broad range of math precisions, providing a single accelerator for every workload. The latest generation A100 80GB doubles GPU memory and debuts the world’s fastest memory bandwidth at 2 terabytes per second (TB/s), speeding time to solution for the largest models and most massive datasets.

A100 is part of the complete NVIDIA data center solution that incorporates building blocks across hardware, networking, software, libraries, and optimized AI models and applications from the NVIDIA NGC™ catalog. Representing the most powerful end-to-end AI and HPC platform for data centers, it allows researchers to deliver real-world results and deploy solutions into production at scale.

• NVIDIA Ampere Architecture

  Whether using MIG to partition an A100 GPU into smaller instances or NVLink to connect multiple GPUs to speed large-scale workloads, A100 can readily handle different-sized acceleration needs, from the smallest job to the biggest multi-node workload. A100’s versatility means IT managers can maximize the utility of every GPU in their data center, around the clock.

• Third-Generation Tensor Cores

  NVIDIA A100 delivers 312 teraFLOPS (TFLOPS) of deep learning performance. That’s 20X the Tensor floating-point operations per second (FLOPS) for deep learning training and 20X the Tensor tera operations per second (TOPS) for deep learning inference compared to NVIDIA Volta GPUs.

• Next-Generation NVLink

  NVIDIA NVLink in A100 delivers 2X higher throughput compared to the previous generation. When combined with NVIDIA NVSwitch, up to 16 A100 GPUs can be interconnected at up to 600 gigabytes per second (GB/sec), unleashing the highest application performance possible on a single server. NVLink is available in A100 SXM GPUs via HGX A100 server boards and in PCIe GPUs via an NVLink Bridge for up to 2 GPUs.

• Multi-Instance GPU (MIG)

  An A100 GPU can be partitioned into as many as seven GPU instances, fully isolated at the hardware level with their own high-bandwidth memory, cache, and compute cores. MIG gives developers access to breakthrough acceleration for all their applications, and IT administrators can offer right-sized GPU acceleration for every job, optimizing utilization and expanding access to every user and application.

• High-Bandwidth Memory (HBM2E)

  With up to 80 GB of HBM2e, A100 delivers the world’s fastest GPU memory bandwidth of over 2TB/s, as well as a DRAM utilization efficiency of 95%. A100 delivers 1.7X higher memory bandwidth over the previous generation.

• Structural Sparsity

  AI networks have millions to billions of parameters. Not all of these parameters are needed for accurate predictions, and some can be converted to zeros, making the models “sparse” without compromising accuracy. Tensor Cores in A100 can provide up to 2X higher performance for sparse models. While the sparsity feature more readily benefits AI inference, it can also improve the performance of model training.

The following table lists the NVIDIA A100 GPU specifications.

The following tables list the ThinkSystem servers that are compatible.

The following table lists the supported operating systems:

• ThinkSystem NVIDIA A100 40GB PCIe Gen4 Passive GPU, 4X67A13135
• ThinkSystem NVIDIA A100 80GB PCIe Gen4 Passive GPU, 4X67A76715

This section lists the NVIDIA software that is available from Lenovo.

• NVIDIA vGPU Software (vApps, vPC, RTX vWS)
• NVIDIA Omniverse Software (OVE)
• NVIDIA AI Enterprise Software
• NVIDIA HPC Compiler Software

• Virtual PC (vPC)

  This product is ideal for users who want a virtual desktop but need great user experience leveraging PC Windows® applications, browsers and high-definition video. NVIDIA Virtual PC delivers a native experience to users in a virtual environment, allowing them to run all their PC applications at full performance.

• NVIDIA RTX Virtual Workstation (RTX vWS)

  NVIDIA RTX vWS is the only virtual workstation that supports NVIDIA RTX technology, bringing advanced features like ray tracing, AI-denoising, and Deep Learning Super Sampling (DLSS) to a virtual environment. Supporting the latest generation of NVIDIA GPUs unlocks the best performance possible, so designers and engineers can create their best work faster. IT can virtualize any application from the data center with an experience that is indistinguishable from a physical workstation — enabling workstation performance from any device.

The following license types are offered:

• Perpetual license

  A non-expiring, permanent software license that can be used on a perpetual basis without the need to renew. Each Lenovo part number includes a fixed number of years of Support, Upgrade and Maintenance (SUMS).

• Annual subscription

  A software license that is active for a fixed period as defined by the terms of the subscription license, typically yearly. The subscription includes Support, Upgrade and Maintenance (SUMS) for the duration of the license term.

• Concurrent User (CCU)

  A method of counting licenses based on active user VMs. If the VM is active and the NVIDIA vGPU software is running, then this counts as one CCU. A vGPU CCU is independent of the connection to the VM.

The following table lists the ordering part numbers and feature codes.

NVIDIA Omniverse Software (OVE)

NVIDIA Omniverse™ Enterprise is an end-to-end collaboration and simulation platform that fundamentally transforms complex design workflows, creating a more harmonious environment for creative teams.

NVIDIA and Lenovo offer a robust, scalable solution for deploying Omniverse Enterprise, accommodating a wide range of professional needs. This document details the critical components, deployment options, and support available, ensuring an efficient and effective Omniverse experience.

Deployment options cater to varying team sizes and workloads. Using Lenovo NVIDIA-Certified Systems™ and Lenovo OVX nodes which are meticulously designed to manage scale and complexity, ensures optimal performance for Omniverse tasks.

Deployment options include:

• Workstations: NVIDIA-Certified Workstations with RTX 6000 Ada GPUs for desktop environments.
• Data Center Solutions: Deployment with Lenovo OVX nodes or NVIDIA-Certified Servers equipped with L40, L40S or A40 GPUs for centralized, high-capacity needs.

NVIDIA Omniverse Enterprise includes the following components and features:

• Platform Components: Kit, Connect, Nucleus, Simulation, RTX Renderer.
• Foundation Applications: USD Composer, USD Presenter.
• Omniverse Extensions: Connect Sample & SDK.
• Integrated Development Environment (IDE)
• Nucleus Configuration: Workstation, Enterprise Nucleus Server (supports up to 8 editors per scene); Self-Service Public Cloud Hosting using Containers.
• Omniverse Farm: Supports batch workloads up to 8 GPUs.
• Enterprise Services: Authentication (SSO/SSL), Navigator Microservice, Large File Transfer, User Accounts SAML/Account Directory.
• User Interface: Workstation & IT Managed Launcher.
• Support: NVIDIA Enterprise Support.
• Deployment Scenarios: Desktop to Data Center: Workstation deployment for building and designing, with options for physical or virtual desktops. For batch tasks, rendering, and SDG workloads that require headless compute, Lenovo OVX nodes are recommended.

The following part numbers are for a subscription license which is active for a fixed period as noted in the description. The license is for a named user which means the license is for named authorized users who may not re-assign or share the license with any other person.

NVIDIA AI Enterprise Software

Lenovo offers the NVIDIA AI Enterprise (NVAIE) cloud-native enterprise software. NVIDIA AI Enterprise is an end-to-end, cloud-native suite of  AI and data analytics software, optimized, certified, and supported by NVIDIA to run on VMware vSphere and bare-metal  with NVIDIA-Certified  Systems™.  It includes key enabling technologies from NVIDIA for rapid deployment, management, and scaling of AI workloads in the modern hybrid cloud.

NVIDIA AI Enterprise is licensed on a per-GPU basis. NVIDIA AI Enterprise products can be purchased as either a perpetual license with support services, or as an annual or multi-year subscription.

The features of NVIDIA AI Enterprise Software are listed in the following table.

Note: Maximum 10 concurrent VMs per product license

The following table lists the ordering part numbers and feature codes.

Find more information in the NVIDIA AI Enterprise Sizing Guide.

NVIDIA HPC Compiler Software

The power cables needed for the A100 SXM GPUs are included with the supported servers.

The A100 PCIe adapter option part number do not ship with auxiliary power cables. Cables are server-specific due to length requirements. For CTO orders, auxiliary power cables are derived by the configurator. For field upgrades, cables will need to be ordered separately as listed in the table below.

The NVIDIA A100 GPU has the following regulatory approvals:

• RCM
• BSMI
• CE
• FCC
• ICES
• KCC
• cUL, UL
• VCCI

The NVIDIA A100 GPU has the following operating characteristics:

• Ambient temperature
  • Operational: 0°C to 50°C (-5°C to 55°C for short term*)
  • Storage: -40°C to 75°C
• Relative humidity:
  • Operational: 5-85% (5-93% short term*)
  • Storage: 5-95%

* A period not more than 96 hours consecutive, not to exceed 15 days per year.

One year limited warranty. When installed in a Lenovo server, the GPU assumes the server’s base warranty and any warranty upgrades.

For more information, refer to these documents:

• ThinkSystem and ThinkAgile GPU Summary:
  https://lenovopress.lenovo.com/lp0768-thinksystem-thinkagile-gpu-summary
• ServerProven compatibility:
  https://serverproven.lenovo.com/
• NVIDIA A100 product page:
  https://www.nvidia.com/en-us/data-center/a100/
• NVIDIA Ampere Architecture page
  https://www.nvidia.com/en-us/data-center/ampere-architecture/