NVIDIA AI Enterprise Release Notes

Release information for all users of NVIDIA AI Enterprise.

1. What's New in NVIDIA AI Enterprise

Features in this release of NVIDIA AI Enterprise are as follows:

  • Support for VMware vSphere with Tanzu.

  • Added support for NVIDIA V100 GPUs.

  • Added support for the Hypervisor software, VMware vSphere 6.7 on NVIDIA V100 and NVIDIA T4 GPUs.

  • Added support for the Guest operating system, Red Hat Enterprise Linux 8.4.

  • Added support for Bare-Metal host operating systems with Red Hat Enterprise Linux 8.4 and Ubuntu 20.04.

2. Supported Hardware and Software

NVIDIA GPUs:

  • NVIDIA V100

  • NVIDIA A100 PCIe 40GB

  • NVIDIA A100 HGX 40GB

  • NVIDIA A100 PCIe 80GB

  • NVIDIA A100 HGX 80GB

  • NVIDIA A40 (SR-IOV Mode Only)

  • NVIDIA A30

  • NVIDIA A10

  • NVIDIA A16

  • NVIDIA RTX A6000

  • NVIDIA RTX A5000

  • NVIDIA T4

NVIDIA-certified systems for NVIDIA GPU Cloud that support the supported NVIDIA GPUs and are also certified for use with VMware vSphere ESXi hypervisor

Multi-node scaling requires an ethernet NIC that supports RoCE. For best performance, NVIDIA recommends using an NVIDIA® Mellanox® ConnectX®-6 Dx and an NVIDIA A100 GPU in each VM used for multi-node scaling. Refer to the Sizing guide and the Multi-Node Training solution guide for further information.

Hypervisor software:

  • VMware vSphere Hypervisor (ESXi) Enterprise Plus Edition 7.0 Update 3
  • VMware vCenter Server 7.0 Update 3
  • VMware vSphere 6.7

Guest operating systems:

  • Ubuntu 20.04 LTS
  • Red Hat Enterprise Linux 8.4

2.1. NVIDIA AI Enterprise Software Components

Software Component NVIDIA Release
NVIDIA vGPU Software 13.2
TensorFlow 2 21.08-tf2-py3
TensorFlow 1 21.08-tf1-py3
PyTorch 21.08-py3
NVIDIA Triton Inference Server 21.08-py3 and 21.08-py3-sdk
NVIDIA TensorRT 21.08-py3
RAPIDS 21.08-cuda11.4-base-ubuntu20.04

2.2. Switching the Mode of a GPU that Supports Multiple Display Modes

Some GPUs support displayless and display-enabled modes but must be used in NVIDIA AI Enterprise deployments in displayless mode.

The GPUs listed in the following table support multiple display modes. As shown in the table, some GPUs are supplied from the factory in displayless mode, but other GPUs are supplied in a display-enabled mode.

GPU Mode as Supplied from the Factory
NVIDIA A40 Displayless
NVIDIA RTX A5000 Display enabled
NVIDIA RTX A6000 Display enabled

A GPU that is supplied from the factory in displayless mode, such as the NVIDIA A40 GPU, might be in a display-enabled mode if its mode has previously been changed.

To change the mode of a GPU that supports multiple display modes, use the displaymodeselector tool, which you can request from the NVIDIA Display Mode Selector Tool page on the NVIDIA Developer website.

Note:

Only the following GPUs support the displaymodeselector tool:

  • NVIDIA A40
  • NVIDIA RTX A5000
  • NVIDIA RTX A6000

Other GPUs that support NVIDIA AI Enterprise do not support the displaymodeselector tool and, unless otherwise stated, do not require display mode switching.

2.3. Requirements for Using C-Series vCS vGPUs

Because C-Series vCS vGPUs have large BAR memory settings, using these vGPUs has some restrictions on VMware ESXi.

2.4. Requirements for Using vGPU on GPUs Requiring 64 GB or More of MMIO Space with Large-Memory VMs

Some GPUs require 64 GB or more of MMIO space. When a vGPU on a GPU that requires 64 GB or more of MMIO space is assigned to a VM with 32 GB or more of memory on ESXi , the VM’s MMIO space must be increased to the amount of MMIO space that the GPU requires.

For more information, refer to VMware Knowledge Base Article: VMware vSphere VMDirectPath I/O: Requirements for Platforms and Devices (2142307).

With ESXi 6.7 or later, no extra configuration is needed.

No extra configuration is needed.

The following table lists the GPUs that require 64 GB or more of MMIO space and the amount of MMIO space that each GPU requires.

2.5. Linux Only: Error Messages for Misconfigured GPUs Requiring Large MMIO Space

In a Linux VM, if the requirements for using C-Series vCS vGPUs or GPUs requiring large MMIO space in pass-through mode are not met, the following error messages are written to the VM's dmesg log during installation of the NVIDIA AI Enterprise graphics driver:

NVRM: BAR1 is 0M @ 0x0 (PCI:0000:02:02.0)
[  90.823015] NVRM: The system BIOS may have misconfigured your GPU.
[  90.823019] nvidia: probe of 0000:02:02.0 failed with error -1
[  90.823031] NVRM: The NVIDIA probe routine failed for 1 device(s).

2.6. NVIDIA CUDA Toolkit Version Support

The releases in this release family of NVIDIA AI Enterprise support NVIDIA CUDA Toolkit 11.4.

To build a CUDA application, the system must have the NVIDIA CUDA Toolkit and the libraries required for linking. For details of the components of NVIDIA CUDA Toolkit, refer to NVIDIA CUDA Toolkit Release Notes for CUDA 11.4.

To run a CUDA application, the system must have a CUDA-enabled GPU and an NVIDIA display driver that is compatible with the NVIDIA CUDA Toolkit release that was used to build the application. If the application relies on dynamic linking for libraries, the system must also have the correct version of these libraries.

For more information about NVIDIA CUDA Toolkit, refer to CUDA Toolkit 11.4 Documentation.

Note:

If you are using NVIDIA AI Enterprise with CUDA on Linux, avoid conflicting installation methods by installing CUDA from a distribution-independent runfile package. Do not install CUDA from a distribution-specific RPM or Deb package.

To ensure that the NVIDIA AI Enterprise graphics driver is not overwritten when CUDA is installed, deselect the CUDA driver when selecting the CUDA components to install.

For more information, see NVIDIA CUDA Installation Guide for Linux.

2.7.  Support

is supported only on a subset of supported GPUs, VMware vSphere Hypervisor (ESXi) releases, and guest operating systems.

is supported on all supported GPUs, hypervisor software releases, and guest operating systems.

Note: vGPU migration is disabled for a VM for which any of the following NVIDIA CUDA Toolkit features is enabled:
  • Unified memory
  • Debuggers
  • Profilers

Supported GPUs

  • Tesla M6
  • Tesla M10
  • Tesla M60
  • Tesla P4
  • Tesla P6
  • Tesla P40

Supported VMware vSphere Hypervisor (ESXi) Releases

Supported Guest OS Releases

Known Issues with Support

Use Case Affected GPUs Issue
Migration to or from a host running an NVIDIA AI Enterprise 11 release before 11.6 Tesla T4 index.html#bug-no-id-vm-hangs-after-vgpu-migration-11-x-13-x
  • Migration from a host running NVIDIA AI Enterprise 11.3 to a host running a different release
  • Migration to a host running NVIDIA AI Enterprise 11.3 from a host running a different release
Tesla T4 index.html#bug-200691763-vgpu-vm-migration-11-0-11-2-to-later-release-fails
Migration from a host that is running a vGPU manager 11 release to a host that is running a vGPU manager 13 release.
  • Tesla T4
  • Tesla V100
index.html#bug-200691445-vm-hangs-after-vgpu-migration-to-newer-vgpu-manager
Migration to or from a host running an NVIDIA AI Enterprise 11 release GPUs based on the NVIDIA Volta™ architecture index.html#bug-200707632-session-freeze-volta-vgpu-vm-migration-to-from-release-11
Migration to or from a host running an NVIDIA AI Enterprise 14 release
  • Tesla T4
  • Tesla V100
index.html#bug-3512790-session-freeze-linux-vm-migration-to-from-release-14
Migration between hosts with different ECC memory configuration All GPUs that support Migration of VMs configured with vGPU stops before the migration is complete

Known Issues with vGPU Migration Support

Use Case Affected GPUs Issue
Migration between hosts with different ECC memory configuration All GPUs that support vGPU migration Migration of VMs configured with vGPU stops before the migration is complete

2.8. Multiple vGPU Support

To support applications and workloads that are compute or graphics intensive, multiple vGPUs can be added to a single VM. The assignment of more than one vGPU to a VM is supported only on a subset of vGPUs and VMware vSphere Hypervisor (ESXi) releases.

Supported vGPUs

Only Q-series vGPUs that are allocated all of the physical GPU's frame buffer are supported.

Only time-sliced vGPUs that are allocated all of the physical GPU's frame buffer are supported. MIG-backed vGPUs are not supported.

Note:
  1. This type of vGPU cannot be assigned with other types of vGPU to the same VM.

Maximum vGPUs per VM

NVIDIA AI Enterprise supports up to a maximum of four vGPUs per VM on VMware vSphere Hypervisor (ESXi).

Supported Hypervisor Releases

Supported Hypervisor Releases

All hypervisors that support NVIDIA AI Enterprise are supported.

2.9. Peer-to-Peer CUDA Transfers over NVLink Support

Peer-to-peer CUDA transfers enable device memory between vGPUs on different GPUs that are assigned to the same VM to be accessed from within the CUDA kernels. NVLink is a high-bandwidth interconnect that enables fast communication between such vGPUs. Peer-to-Peer CUDA transfers over NVLink are supported only on a subset of vGPUs, VMware vSphere Hypervisor (ESXi) releases, and guest OS releases.

Supported vGPUs

Only Q-series vGPUs that are allocated all of the physical GPU's frame buffer on physical GPUs that support NVLink are supported.

Only time-sliced vGPUs that are allocated all of the physical GPU's frame buffer on physical GPUs that support NVLink are supported.

Supported Hypervisor Releases

Peer-to-Peer CUDA Transfers over NVLink are supported on all hypervisor releases that support the assignment of more than one vGPU to a VM. For details, see Multiple vGPU Support.

Supported Guest OS Releases

Linux only. Peer-to-Peer CUDA Transfers over NVLink are not supported on Windows.

Limitations

  • PCIe is not supported.
  • SLI is not supported.

2.10. Unified Memory Support

Unified memory is a single memory address space that is accessible from any CPU or GPU in a system. It creates a pool of managed memory that is shared between the CPU and GPU to provide a simple way to allocate and access data that can be used by code running on any CPU or GPU in the system. Unified memory is supported only on a subset of vGPUs and guest OS releases.

Note:Unified memory is disabled by default. If used, you must enable unified memory individually for each vGPU that requires it by setting a vGPU plugin parameter.

Supported vGPUs

Only Q-series vGPUs that are allocated all of the physical GPU's frame buffer on physical GPUs that support unified memory are supported.

Only time-sliced vGPUs that are allocated all of the physical GPU's frame buffer on physical GPUs that support unified memory are supported.

Supported Guest OS Releases

Linux only. Unified memory is not supported on Windows.

Limitations

  • When unified memory is enabled for a VM, NVIDIA CUDA Toolkit profilers are disabled.

2.11. NVIDIA GPU Operator Support

NVIDIA GPU Operator simplifies the deployment of NVIDIA AI Enterprise with software container platforms on immutable operating systems. An immutable operating system does not allow the installation of the NVIDIA AI Enterprise graphics driver directly on the operating system. NVIDIA GPU Operator is supported only on specific combinations of VMware vSphere Hypervisor (ESXi) release, container platform, and guest OS release.

VMware vSphere Hypervisor (ESXi) Release Container Platform Guest OS
VMware vSphere Hypervisor (ESXi) Release Container Platform Guest OS

3. Known Product Limitations

Known product limitations for this release of NVIDIA AI Enterprise are described in the following sections.

3.1. Issues occur when the channels allocated to a vGPU are exhausted

Description

Issues occur when the channels allocated to a vGPU are exhausted and the guest VM to which the vGPU is assigned fails to allocate a channel to the vGPU. A physical GPU has a fixed number of channels and the number of channels allocated to each vGPU is inversely proportional to the maximum number of vGPUs allowed on the physical GPU.

When the channels allocated to a vGPU are exhausted and the guest VM fails to allocate a channel, the following errors are reported on the hypervisor host or in an NVIDIA bug report:

Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0): Guest attempted to allocate channel above its max channel limit 0xfb
Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0): VGPU message 6 failed, result code: 0x1a
Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0):         0xc1d004a1, 0xff0e0000, 0xff0400fb, 0xc36f,
Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0):         0x1, 0xff1fe314, 0xff1fe038, 0x100b6f000, 0x1000,
Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0):         0x80000000, 0xff0e0200, 0x0, 0x0, (Not logged),
Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0):         0x1, 0x0
Jun 26 08:01:25 srvxen06f vgpu-3[14276]: error: vmiop_log: (0x0): , 0x0

Workaround

Use a vGPU type with more frame buffer, thereby reducing the maximum number of vGPUs allowed on the physical GPU. As a result, the number of channels allocated to each vGPU is increased.

3.2. Total frame buffer for vGPUs is less than the total frame buffer on the physical GPU

Some of the physical GPU's frame buffer is used by the hypervisor on behalf of the VM for allocations that the guest OS would otherwise have made in its own frame buffer. The frame buffer used by the hypervisor is not available for vGPUs on the physical GPU. In NVIDIA vGPU deployments, frame buffer for the guest OS is reserved in advance, whereas in bare-metal deployments, frame buffer for the guest OS is reserved on the basis of the runtime needs of applications.

If error-correcting code (ECC) memory is enabled on a physical GPU that does not have HBM2 memory, the amount of frame buffer that is usable by vGPUs is further reduced. All types of vGPU are affected, not just vGPUs that support ECC memory.

On all GPUs that support ECC memory and, therefore, dynamic page retirement, additional frame buffer is allocated for dynamic page retirement. The amount that is allocated is inversely proportional to the maximum number of vGPUs per physical GPU. All GPUs that support ECC memory are affected, even GPUs that have HBM2 memory or for which ECC memory is disabled.

The approximate amount of frame buffer that NVIDIA AI Enterprise reserves can be calculated from the following formula:

max-reserved-fb = vgpu-profile-size-in-mb÷16 + 16 + ecc-adjustments + page-retirement-allocation + compression-adjustment

max-reserved-fb
The maximum total amount of reserved frame buffer in Mbytes that is not available for vGPUs.
vgpu-profile-size-in-mb
The amount of frame buffer in Mbytes allocated to a single vGPU. This amount depends on the vGPU type. For example, for the T4-16Q vGPU type, vgpu-profile-size-in-mb is 16384.
ecc-adjustments
The amount of frame buffer in Mbytes that is not usable by vGPUs when ECC is enabled on a physical GPU that does not have HBM2 memory.
  • If ECC is enabled on a physical GPU that does not have HBM2 memory ecc-adjustments is fb-without-ecc/16, which is equivalent to 64 Mbytes for every Gbyte of frame buffer assigned to the vGPU. fb-without-ecc is total amount of frame buffer with ECC disabled.
  • If ECC is disabled or the GPU has HBM2 memory, ecc-adjustments is 0.
page-retirement-allocation
The amount of frame buffer in Mbytes that is reserved for dynamic page retirement.
  • On GPUs based on the NVIDIA Maxwell GPU architecture, page-retirement-allocation = 4÷max-vgpus-per-gpu.
  • On GPUs based on NVIDIA GPU architectures after the Maxwell architecture, page-retirement-allocation = 128÷max-vgpus-per-gpu
max-vgpus-per-gpu
The maximum number of vGPUs that can be created simultaneously on a physical GPU. This number varies according to the vGPU type. For example, for the T4-16Q vGPU type, max-vgpus-per-gpu is 1.
compression-adjustment

The amount of frame buffer in Mbytes that is reserved for the higher compression overhead in vGPU types with 12 Gbytes or more of frame buffer on GPUs based on the Turing architecture.

compression-adjustment depends on the vGPU type as shown in the following table.

vGPU Type Compression Adjustment (MB)

For all other vGPU types, compression-adjustment is 0.

3.3. Single vGPU benchmark scores are lower than pass-through GPU

Description

A single vGPU configured on a physical GPU produces lower benchmark scores than the physical GPU run in pass-through mode.

Aside from performance differences that may be attributed to a vGPU’s smaller frame buffer size, vGPU incorporates a performance balancing feature known as Frame Rate Limiter (FRL). On vGPUs that use the best-effort scheduler, FRL is enabled. On vGPUs that use the fixed share or equal share scheduler, FRL is disabled.

FRL is used to ensure balanced performance across multiple vGPUs that are resident on the same physical GPU. The FRL setting is designed to give good interactive remote graphics experience but may reduce scores in benchmarks that depend on measuring frame rendering rates, as compared to the same benchmarks running on a pass-through GPU.

3.4. VMs configured with large memory fail to initialize vGPU when booted

Description

When starting multiple VMs configured with large amounts of RAM (typically more than 32GB per VM), a VM may fail to initialize vGPU. In this scenario, the VM boots in VMware SVGA mode. The NVIDIA AI Enterprise GPU is present in Windows Device Manager but displays a warning sign, and the following device status:

Windows has stopped this device because it has reported problems. (Code 43)

When this error occurs, VGPU message failed messages and XID error messages are written to the VMware vSphere VM’s log file.

4. Resolved Issues in NVIDIA AI Enterprise

This is the first GA release of NVIDIA AI Enterprise. Resolved issues will be listed in future releases.

5. Known Issues

5.1. Migration of VMs configured with vGPU stops before the migration is complete

Description

When a VM configured with vGPU is migrated to another host, the migration stops before it is complete.

This issue occurs if the ECC memory configuration (enabled or disabled) on the source and destination hosts are different. The ECC memory configuration on both the source and destination hosts must be identical.

Workaround

Before attempting to migrate the VM again, ensure that the ECC memory configuration on both the source and destination hosts are identical.

Status

Not an NVIDIA bug

Ref. #

200520027

Notices

Notice

This document is provided for information purposes only and shall not be regarded as a warranty of a certain functionality, condition, or quality of a product. NVIDIA Corporation (“NVIDIA”) makes no representations or warranties, expressed or implied, as to the accuracy or completeness of the information contained in this document and assumes no responsibility for any errors contained herein. NVIDIA shall have no liability for the consequences or use of such information or for any infringement of patents or other rights of third parties that may result from its use. This document is not a commitment to develop, release, or deliver any Material (defined below), code, or functionality.

NVIDIA reserves the right to make corrections, modifications, enhancements, improvements, and any other changes to this document, at any time without notice.

Customer should obtain the latest relevant information before placing orders and should verify that such information is current and complete.

NVIDIA products are sold subject to the NVIDIA standard terms and conditions of sale supplied at the time of order acknowledgement, unless otherwise agreed in an individual sales agreement signed by authorized representatives of NVIDIA and customer (“Terms of Sale”). NVIDIA hereby expressly objects to applying any customer general terms and conditions with regards to the purchase of the NVIDIA product referenced in this document. No contractual obligations are formed either directly or indirectly by this document.

NVIDIA products are not designed, authorized, or warranted to be suitable for use in medical, military, aircraft, space, or life support equipment, nor in applications where failure or malfunction of the NVIDIA product can reasonably be expected to result in personal injury, death, or property or environmental damage. NVIDIA accepts no liability for inclusion and/or use of NVIDIA products in such equipment or applications and therefore such inclusion and/or use is at customer’s own risk.

NVIDIA makes no representation or warranty that products based on this document will be suitable for any specified use. Testing of all parameters of each product is not necessarily performed by NVIDIA. It is customer’s sole responsibility to evaluate and determine the applicability of any information contained in this document, ensure the product is suitable and fit for the application planned by customer, and perform the necessary testing for the application in order to avoid a default of the application or the product. Weaknesses in customer’s product designs may affect the quality and reliability of the NVIDIA product and may result in additional or different conditions and/or requirements beyond those contained in this document. NVIDIA accepts no liability related to any default, damage, costs, or problem which may be based on or attributable to: (i) the use of the NVIDIA product in any manner that is contrary to this document or (ii) customer product designs.

No license, either expressed or implied, is granted under any NVIDIA patent right, copyright, or other NVIDIA intellectual property right under this document. Information published by NVIDIA regarding third-party products or services does not constitute a license from NVIDIA to use such products or services or a warranty or endorsement thereof. Use of such information may require a license from a third party under the patents or other intellectual property rights of the third party, or a license from NVIDIA under the patents or other intellectual property rights of NVIDIA.

Reproduction of information in this document is permissible only if approved in advance by NVIDIA in writing, reproduced without alteration and in full compliance with all applicable export laws and regulations, and accompanied by all associated conditions, limitations, and notices.

THIS DOCUMENT AND ALL NVIDIA DESIGN SPECIFICATIONS, REFERENCE BOARDS, FILES, DRAWINGS, DIAGNOSTICS, LISTS, AND OTHER DOCUMENTS (TOGETHER AND SEPARATELY, “MATERIALS”) ARE BEING PROVIDED “AS IS.” NVIDIA MAKES NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE. TO THE EXTENT NOT PROHIBITED BY LAW, IN NO EVENT WILL NVIDIA BE LIABLE FOR ANY DAMAGES, INCLUDING WITHOUT LIMITATION ANY DIRECT, INDIRECT, SPECIAL, INCIDENTAL, PUNITIVE, OR CONSEQUENTIAL DAMAGES, HOWEVER CAUSED AND REGARDLESS OF THE THEORY OF LIABILITY, ARISING OUT OF ANY USE OF THIS DOCUMENT, EVEN IF NVIDIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. Notwithstanding any damages that customer might incur for any reason whatsoever, NVIDIA’s aggregate and cumulative liability towards customer for the products described herein shall be limited in accordance with the Terms of Sale for the product.

VESA DisplayPort

DisplayPort and DisplayPort Compliance Logo, DisplayPort Compliance Logo for Dual-mode Sources, and DisplayPort Compliance Logo for Active Cables are trademarks owned by the Video Electronics Standards Association in the United States and other countries.

HDMI

HDMI, the HDMI logo, and High-Definition Multimedia Interface are trademarks or registered trademarks of HDMI Licensing LLC.

OpenCL

OpenCL is a trademark of Apple Inc. used under license to the Khronos Group Inc.

Trademarks

NVIDIA, the NVIDIA logo, NVIDIA Maxwell, NVIDIA Pascal, NVIDIA Turing, NVIDIA Volta, Quadro, and Tesla are trademarks or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of the respective companies with which they are associated.