vcfclassnotes_quiz5

<p>Automated memory allocation for AI workloads. (C)</p> Signup and view all the answers

<p>They must be manually powered off before operations. (B)</p> Signup and view all the answers

<p>SR-IOV (D)</p> Signup and view all the answers

<p>It supports maintenance operations only. (A)</p> Signup and view all the answers

<p>To provision AI workloads on ESXi hosts with enhancing resource access. (A)</p> Signup and view all the answers

<p>Both communication traffic and CPU overhead are significantly reduced. (C)</p> Signup and view all the answers

<p>Using familiar tools without managing isolated AI resources. (D)</p> Signup and view all the answers

<p>GPUs have more cores that facilitate parallel processing. (D)</p> Signup and view all the answers

<p>Enable MIG Mode (D)</p> Signup and view all the answers

<p>To facilitate direct communication between GPUs. (B)</p> Signup and view all the answers

<p>High-speed connection between multiple GPUs. (D)</p> Signup and view all the answers

<p>Equal shares of GPU resources based on preconfigured profiles. (B)</p> Signup and view all the answers

<p>Run the SDDC Manager configuration. (A)</p> Signup and view all the answers

<p>Shared GPU slices can range from 1 to 7. (B)</p> Signup and view all the answers

<p>Install the NVIDIA Guest Driver. (A)</p> Signup and view all the answers

<p>Tolerance of memory latency due to parallel processing. (D)</p> Signup and view all the answers

<p>Dynamic DirectPath passthrough mode (C)</p> Signup and view all the answers

<p>Time-Slicing Mode (C)</p> Signup and view all the answers

<p>7 (B)</p> Signup and view all the answers

<p>Time-Slicing Mode (C)</p> Signup and view all the answers

<p>To run multiple workloads in parallel on GPU resources (A)</p> Signup and view all the answers

<p>nvidia-smi (D)</p> Signup and view all the answers

<p>MIG Mode (A)</p> Signup and view all the answers

<p>NVIDIA Host software (VIB) (A)</p> Signup and view all the answers

<p>High-demand, GPU-intensive workloads (A)</p> Signup and view all the answers

<p>Workloads operate in series with shared access to the GPU (B)</p> Signup and view all the answers

<p>GPUs are designed for parallel processing tasks. (A)</p> Signup and view all the answers

<p>Genetic algorithms (C)</p> Signup and view all the answers

<p>NVLink (A)</p> Signup and view all the answers

<p>They prioritize higher throughput over cache size. (C)</p> Signup and view all the answers

<p>Generative AI (B)</p> Signup and view all the answers

<p>Model optimization for specific tasks (B)</p> Signup and view all the answers

<p>To enhance the training and inference speed. (B)</p> Signup and view all the answers

<p>NVIDIA GPUs are optimized for high-throughput calculations. (B)</p> Signup and view all the answers

<p>Prompt completion (B)</p> Signup and view all the answers

<p>It allows computers to learn from data independently. (B)</p> Signup and view all the answers

<p>A technology that generates human-like creativity and reasoning. (C)</p> Signup and view all the answers

<p>GPUs have more cores, enabling parallel processing. (C)</p> Signup and view all the answers

<p>They enhance the speed of computations and processing. (B)</p> Signup and view all the answers

<p>Deep Learning (A)</p> Signup and view all the answers

<p>Deep-learning neural networks based on transformers. (A)</p> Signup and view all the answers

<p>By training on vast and dynamic datasets. (B)</p> Signup and view all the answers

<p>Learning and improving from data without explicit programming. (D)</p> Signup and view all the answers

<p>An entire GPU is allocated to a specific VM. (A)</p> Signup and view all the answers

<p>Workloads share the physical GPU and operate in series. (A)</p> Signup and view all the answers

<p>Their design for managing higher throughput. (D)</p> Signup and view all the answers

<p>MIG Mode (Multi-Instance GPU Mode) (B)</p> Signup and view all the answers

<p>Refining a pre-trained model for specific applications. (C)</p> Signup and view all the answers

<p>Completes prompts based on learned knowledge. (D)</p> Signup and view all the answers

<p>Maximizes GPU utilization by running several workloads. (A)</p> Signup and view all the answers

<p>Allocating a full GPU to a single VM-based workload. (A)</p> Signup and view all the answers

<p>vGPU Time-Slicing Mode (C)</p> Signup and view all the answers

<p>Dedicated and predictable performance for multiple instances. (C)</p> Signup and view all the answers

<p>vGPU Time-Slicing with equal shares (C)</p> Signup and view all the answers

<p>Higher throughput capabilities (D)</p> Signup and view all the answers

<p>Multiple workloads sharing a single GPU (B)</p> Signup and view all the answers

<p>It enables direct access to GPU memory. (B)</p> Signup and view all the answers

<p>Equal GPU shares (C)</p> Signup and view all the answers

<p>High-speed connections between GPUs (B)</p> Signup and view all the answers

<p>When leveraging GPU capabilities in TKG (A)</p> Signup and view all the answers

<p>GPUs excel in parallel processing with many cores. (B)</p> Signup and view all the answers

<p>Equal sharing of GPU resources among VMs (A)</p> Signup and view all the answers

<p>Install and configure the NVIDIA guest driver (C)</p> Signup and view all the answers

<p>Dedicated GPU Mode (B)</p> Signup and view all the answers

<p>It provides all-to-all communication at full NVLink speed. (B)</p> Signup and view all the answers

<p>GPUs can be allocated individually or as a group to virtual machines. (A)</p> Signup and view all the answers

<p>It ensures all hosts in a cluster have a consistent GPU device and image. (C)</p> Signup and view all the answers

<p>Manually power off the GPU-enabled TKG VMs. (D)</p> Signup and view all the answers

<p>To migrate workloads without needing VM downtime. (B)</p> Signup and view all the answers

<p>It treats a single PCIe device as multiple separate physical devices. (C)</p> Signup and view all the answers

<p>They are significantly reduced, enhancing overall system performance. (B)</p> Signup and view all the answers

<p>A structured platform for provisioning AI workloads on ESXi hosts. (A)</p> Signup and view all the answers

<p>GPU-enabled TKG VMs must be powered off before Lifecycle Manager operations. (D)</p> Signup and view all the answers

<p>By allowing them to provision AI workloads on-demand with optimized resources. (B)</p> Signup and view all the answers

<p>A GPU can process tasks in parallel due to significantly more cores. (C)</p> Signup and view all the answers

<p>To understand, generate, and interact with human language in a human-like manner. (B)</p> Signup and view all the answers

<p>Neural networks found in the human brain. (C)</p> Signup and view all the answers

<p>Traditional rule-based systems. (B)</p> Signup and view all the answers

<p>Significantly more cores for parallel processing. (C)</p> Signup and view all the answers

<p>Improved performance for intensive computational tasks. (D)</p> Signup and view all the answers

<p>Fine-tuning tasks. (D)</p> Signup and view all the answers

<p>Offers human-like creativity and reasoning. (D)</p> Signup and view all the answers

<p>They are designed for higher throughput. (C)</p> Signup and view all the answers

<p>NVIDIA NVLink. (A)</p> Signup and view all the answers

<p>To allow direct communication between NVIDIA GPUs for improved performance (B)</p> Signup and view all the answers

<p>It enables parallel operation of multiple workloads sharing a GPU (C)</p> Signup and view all the answers

<p>Ability to process tasks in parallel due to more cores (D)</p> Signup and view all the answers

<p>Share GPU resources equally based on preconfigured profiles (B)</p> Signup and view all the answers

<p>Install the NVIDIA Guest Driver (C)</p> Signup and view all the answers

<p>To facilitate high-speed connections between multiple GPUs on a single server (D)</p> Signup and view all the answers

<p>Through a design that supports efficient parallel processing (C)</p> Signup and view all the answers

<p>Pre-configure vGPU profiles for time-sharing or MIG (A)</p> Signup and view all the answers

<p>It enhances the GPU's parallel processing capabilities (D)</p> Signup and view all the answers

<p>Commission the hosts into the centralized management system (C)</p> Signup and view all the answers

<p>7 slices (C)</p> Signup and view all the answers

<p>Workloads operate in series, sharing the GPU based on scheduled time. (A)</p> Signup and view all the answers

<p>When resources need to be shared among multiple workloads efficiently. (C)</p> Signup and view all the answers

<p>For multiple workloads requiring high throughput and parallel processing. (A)</p> Signup and view all the answers

<p>To enable MIG Mode at the ESXi host level. (A)</p> Signup and view all the answers

<p>Sharing GPU resources to maximize throughput. (D)</p> Signup and view all the answers

<p>When a workload requires dedicated access to the GPU's resources. (A)</p> Signup and view all the answers

<p>For workloads that need a secured and predictable performance. (C)</p> Signup and view all the answers

<p>To maximize the availability and performance of VM workloads. (D)</p> Signup and view all the answers

<p>All-to-all GPU communication at full NVLink speed in a single node and between nodes (B)</p> Signup and view all the answers

<p>Requires all hosts in a cluster to have the same GPU device and image (D)</p> Signup and view all the answers

<p>Only NVIDIA GPUs with specific licensing are eligible (A)</p> Signup and view all the answers

<p>Migration that involves powering off the GPU workload to ensure safety (C)</p> Signup and view all the answers

<p>Manually power off the GPU-enabled VMs. (C)</p> Signup and view all the answers

<p>It facilitates high bandwidth connections between multiple GPUs for optimal performance. (B)</p> Signup and view all the answers

<p>Provisioning AI workloads like Retrieval-augmented Generation (RAG) using deep learning (C)</p> Signup and view all the answers

<p>The efficiency of task allocation and overall system performance (D)</p> Signup and view all the answers

<p>Single PCIe devices to appear as multiple separate physical devices to the hypervisor (B)</p> Signup and view all the answers

<p>Optimized resource management with comprehensive lifecycle controls (A)</p> Signup and view all the answers

<p>It mimics the structure of the human brain. (D)</p> Signup and view all the answers

<p>Deep-learning neural networks (B)</p> Signup and view all the answers

<p>They can process tasks in parallel with many cores. (B)</p> Signup and view all the answers

<p>To complete prompts and generate outputs. (C)</p> Signup and view all the answers

<p>Machine Learning (D)</p> Signup and view all the answers

<p>Understanding and generating human-like responses. (D)</p> Signup and view all the answers

<p>GPUs excel at parallel processing with more cores. (A)</p> Signup and view all the answers

<p>To enhance model performance on specific tasks. (A)</p> Signup and view all the answers

<p>Pre-training tasks (A)</p> Signup and view all the answers

<p>They are designed for higher throughput. (D)</p> Signup and view all the answers

<p>To increase the number of virtual devices presented to a workload (B)</p> Signup and view all the answers

<p>Creating a VM class with appropriate resource settings (B)</p> Signup and view all the answers

<p>Creation of multiple vGPU instances from a single physical GPU (A)</p> Signup and view all the answers

<p>Increased bandwidth by allowing direct GPU memory access (C)</p> Signup and view all the answers

<p>Enhanced performance for applications requiring high throughput (D)</p> Signup and view all the answers

<p>Dedicated components for parallel computation (A)</p> Signup and view all the answers

<p>By enabling parallel processing of tasks with many cores (C)</p> Signup and view all the answers

<p>It simplifies the interconnection of multiple PCIe devices (D)</p> Signup and view all the answers

<p>It enables communication between the VM and the GPU (B)</p> Signup and view all the answers

<p>7 slices (A)</p> Signup and view all the answers

<p>It enables all-to-all GPU communication at full NVLink speed. (A)</p> Signup and view all the answers

<p>It enables the allocation of all or a subset of GPUs to a VM. (D)</p> Signup and view all the answers

<p>All hosts require the same GPU device and image. (A)</p> Signup and view all the answers

<p>It is restricted to maintenance operations only. (B)</p> Signup and view all the answers

<p>They must be manually powered off. (B)</p> Signup and view all the answers

<p>It provides on-demand access to AI and ML optimized resources. (D)</p> Signup and view all the answers

<p>It enables a single physical GPU to act as multiple distinct logical devices. (A)</p> Signup and view all the answers

<p>To provision production-ready AI workloads for development teams. (A)</p> Signup and view all the answers

<p>It enables workloads to run in a sequential manner, sharing the available GPU resources. (C)</p> Signup and view all the answers

<p>It enables seamless migration of GPU workloads during maintenance. (D)</p> Signup and view all the answers

<p>MIG Mode (Multi-Instance GPU Mode) (D)</p> Signup and view all the answers

<p>It guarantees equal shares for multiple workloads. (C)</p> Signup and view all the answers

<p>Allows up to 7 allocations of GPU slices for workloads. (C)</p> Signup and view all the answers

<p>For a single VM-based workload demanding the entire GPU. (D)</p> Signup and view all the answers

<p>Nvidia vGPU (Shared GPU) (D)</p> Signup and view all the answers

<p>Demand isolation of resources with predictable performance. (C)</p> Signup and view all the answers

<p>They operate in series with processing scheduled in turns. (B)</p> Signup and view all the answers

<p>Cannot exceed the GPU's physical core limits. (D)</p> Signup and view all the answers

<p>It mimics the brain's network of neurons for processing. (C)</p> Signup and view all the answers

<p>Memory bandwidth optimization techniques (D)</p> Signup and view all the answers

<p>GPUs have significantly more cores for parallel processing. (A)</p> Signup and view all the answers

<p>Generation of human-like language responses. (C)</p> Signup and view all the answers

<p>It can produce human-like creativity and language understanding. (D)</p> Signup and view all the answers

<p>It facilitates parallel processing of data. (C)</p> Signup and view all the answers

<p>7 (C)</p> Signup and view all the answers

<p>By dedicating more components to computation. (D)</p> Signup and view all the answers

<p>To adjust the model to perform better on specific tasks. (C)</p> Signup and view all the answers

<p>Operating workloads in series with shared access (A)</p> Signup and view all the answers

<p>MIG Mode (C)</p> Signup and view all the answers

<p>They leverage vast amounts of text data for training. (D)</p> Signup and view all the answers

<p>vGPU configuration (A)</p> Signup and view all the answers

<p>They enhance the performance and efficiency of computations. (B)</p> Signup and view all the answers

<p>Running workloads in a parallel fashion (C)</p> Signup and view all the answers

<p>Maximizes utilization by running multiple workloads concurrently (C)</p> Signup and view all the answers

<p>Best effort scheduling in Time-Slicing Mode (D)</p> Signup and view all the answers

<p>Time-Slicing Mode with equal shares (A)</p> Signup and view all the answers

<p>VGPU Profile (D)</p> Signup and view all the answers

<p>Reduces the need for CPU intervention (D)</p> Signup and view all the answers

<p>Nvidia NVLINK (A)</p> Signup and view all the answers

<p>VM Class (A)</p> Signup and view all the answers

<p>1-7 (D)</p> Signup and view all the answers

<p>Equal shares based on profiles (A)</p> Signup and view all the answers

<p>To allow virtualization of network devices (B)</p> Signup and view all the answers

<p>Higher memory latency tolerance (B)</p> Signup and view all the answers

<p>Scalability of resource allocation (D)</p> Signup and view all the answers

<p>Optimized for high throughput volumes (D)</p> Signup and view all the answers

<p>It enables all-to-all GPU communication at full NVLink speed. (B)</p> Signup and view all the answers

<p>It enables a single VM to allocate up to 8 GPUs simultaneously. (B)</p> Signup and view all the answers

<p>NVIDIA AI Enterprise Suite licensing (C)</p> Signup and view all the answers

<p>It supports migration and maintenance operations only. (C)</p> Signup and view all the answers

<p>Manually power off the VMs. (B)</p> Signup and view all the answers

<p>To require uniform GPU device and image across all hosts. (D)</p> Signup and view all the answers

<p>Configuring access to AI-optimized resources. (A)</p> Signup and view all the answers

<p>More efficient GPU-to-GPU communication for larger workloads. (D)</p> Signup and view all the answers

<p>To support production-ready AI workloads on Tanzu Kubernetes Grid clusters. (C)</p> Signup and view all the answers

<p>SR-IOV (Single Root I/O Virtualization) (B)</p> Signup and view all the answers

<p>Deep learning mimics the neural network structure of the brain. (B)</p> Signup and view all the answers

<p>Deep-learning neural networks (transformers) (D)</p> Signup and view all the answers

<p>GPUs are optimized for parallel processing with multiple cores. (C)</p> Signup and view all the answers

<p>Generative AI (C)</p> Signup and view all the answers

<p>By enabling faster computations and data processing. (A)</p> Signup and view all the answers

<p>Fine-tuning (C)</p> Signup and view all the answers

<p>Establishing language understanding and context (C)</p> Signup and view all the answers

<p>GPUs are designed for high throughput rather than low latency. (A)</p> Signup and view all the answers

<p>Transformers and deep learning techniques (A)</p> Signup and view all the answers

<p>Generative AI can produce creative outputs and mimic human-like reasoning. (B)</p> Signup and view all the answers

<p>Provides increased speed for GPU-to-GPU communication (D)</p> Signup and view all the answers

<p>vSphere Lifecycle Manager (A)</p> Signup and view all the answers

<p>Manually power off the GPU-enabled VMs (D)</p> Signup and view all the answers

<p>Development of AI workloads, such as deep learning (B)</p> Signup and view all the answers

<p>All hosts need the same GPU device and image (B)</p> Signup and view all the answers

<p>Non-maintenance operations (C)</p> Signup and view all the answers

<p>NVIDIA NVSwitch architecture (A)</p> Signup and view all the answers

<p>Provision of production-ready AI workloads (A)</p> Signup and view all the answers

<p>NVIDIA NVSwitch (C)</p> Signup and view all the answers

<p>It allows a single PCIe device to mimic multiple devices. (D)</p> Signup and view all the answers

<p>VM Class (C)</p> Signup and view all the answers

<p>GPUs accommodate more components for computation than memory. (D)</p> Signup and view all the answers

<p>Divides a physical GPU into multiple smaller GPU instances (C)</p> Signup and view all the answers

<p>To maximize GPU utilization by running many workloads simultaneously (B)</p> Signup and view all the answers

<p>It facilitates direct memory access between NVIDIA GPUs. (D)</p> Signup and view all the answers

<p>It segments the GPU into multiple slices for parallel workloads. (A)</p> Signup and view all the answers

<p>7 (A)</p> Signup and view all the answers

<p>Time-Slicing Mode (B)</p> Signup and view all the answers

<p>Equal shares of GPU resources based on preconfigured profiles. (B)</p> Signup and view all the answers

<p>It enables high-speed connections between multiple GPUs. (C)</p> Signup and view all the answers

<p>NVIDIA Host vGPU Manager Driver (A)</p> Signup and view all the answers

<p>Sharing GPU resources among multiple virtual machines sequentially. (B)</p> Signup and view all the answers

<p>Complete GPU allocation to a specific workload (C)</p> Signup and view all the answers

<p>To configure the level of GPU resource sharing among VMs. (B)</p> Signup and view all the answers

<p>Multiple workloads that need to operate in parallel (B)</p> Signup and view all the answers

<p>GPUs can handle high throughput volumes for parallel processing. (C)</p> Signup and view all the answers

<p>Equal shares (C)</p> Signup and view all the answers

<p>Enables multiple workloads to run on shared GPU resources (D)</p> Signup and view all the answers

<p>Ability to execute multiple operations in parallel (D)</p> Signup and view all the answers

Connects multiple NVLinks for GPU communication. Provides fast GPU-to-GPU communication within a single node or between nodes.

Up to 8 GPUs can be assigned to a single virtual machine (VM) on a host using vSphere device groups.

A platform for deploying AI workloads on vSphere hosts with NVIDIA GPUs.

Manages the lifecycle of GPU-enabled hosts in a cluster. Must use the identical NVIDIA GPU device and image for all hosts and requires NVIDIA AI licensing.

GPU-enabled VMs in TKG must be manually powered down before vSphere Lifecycle Manager operations. Re-instantiated on another host.

GPU-enabled VMs need manual shutdowns prior to vSphere Lifecycle Manager operations. vMotion supported only for maintenance.

A method for high-speed communication between the GPU and the host CPU, bypassing the hypervisor.

Single PCIe device appears as multiple devices to the hypervisor or guest OS.

Processes execute in sequence, not simultaneously in hardware.

Tasks like deep learning, often requiring large-scale GPU computations.

Setting up an ESXi host for NVIDIA GPUs, including adding devices, enabling acceleration, and configuring drivers.

Configures vGPU resource allocation for a virtual machine (time-sharing or MIG).

Multiple instances GPU mode, allowing multiple virtual machines to share a physical GPU.

Adding a host to the vCenter inventory and preparing for use within the virtual data center environment.

A technology that allows direct communication between NVIDIA GPUs, improving performance by 10x.

Software that enables a virtual machine to interact with NVIDIA GPUs.

A template for creating virtual machines, specifying their configuration, resources and requirements from workloads/containers.

Assigning equal shares of GPU resources to multiple virtual machines, based on profiles.

A high-speed connection between multiple GPUs on a server, simplifying device access, usable with vCF 5.1.

A configuration mode where an entire GPU is allocated to a single VM workload.

A configuration mode allowing multiple VMs to access parts of a physical GPU concurrently.

A vGPU mode where VMs take turns using the GPU in series.

A vGPU mode that divides a physical GPU into multiple smaller instances.

A representation of a fractioned GPU instance in MIG Mode.

Command used to enable and manage MIG Mode at the ESXi host level.

Scheduling workloads on a shared GPU, using different approaches (best effort, equal or fixed shares).

A time-slicing mode where each workload gets GPU time as needed.

A time-slicing mode where each workload is granted a fixed amount of GPU time.

A time-slicing mode where each workload has access to the GPU with equal priority.

Mimicking human or other living entity intelligence or behavior

Computers learning from data without explicit rules

Machine learning technique using brain-like networks

AI that creates new content

Large language model; processes text to understand and create language

Graphics Processing Unit; a specialized processor for parallel tasks

Running multiple virtual computers on one physical computer using CPUs

Using GPUs to speed up tasks

Processing multiple tasks at the same time

Dedicated hardware components to speed up tasks

A type of AI that understands and generates human language. It can process vast amounts of text and produce coherent responses.

GPUs are specialized processors designed for parallel tasks, while CPUs are more general-purpose processors.

Training a language model on vast amounts of data to learn the general structure and patterns of human language.

Adapting a pre-trained language model to perform specific tasks, like generating different types of creative text formats, by providing specialized data.

Using a trained language model to generate responses or complete tasks based on user prompts.

A technology that enables direct communication between NVIDIA GPUs, resulting in 10x performance improvement.

A high-speed connection between multiple GPUs on the same server, allowing for fast communication and simplified device management.

CPUs are better at handling low-latency tasks, processing tasks in a sequential manner, while GPUs excel at high throughput, capable of parallel processing.

A hardware component that connects multiple NVLink connections, enabling high-speed communication among GPUs in a single node or across nodes. It allows all-to-all communication at full NVLink speed, crucial for large AI/ML workloads.

A feature in vSphere that allows you to allocate multiple GPUs to a single virtual machine (VM). This enables better performance across all the allocated GPUs for intensive workloads, such as AI/ML, by consolidating GPU resources for a single purpose.

A tool that manages the lifecycle of hosts with GPUs. It enforces consistency by requiring all hosts in a cluster to have the same GPU device and image, and also requires NVIDIA AI licensing for proper operation.

A technology that allows a single PCIe device (like a GPU) to be presented to the host system as multiple separate physical devices. Each device is independently managed for better resource allocation.

GPU-enabled virtual machines (VMs) require manual powering off before performing vSphere Lifecycle Manager operations. This is due to the potential for conflicts during host operations, ensuring stability for both GPU VMs and Lifecycle Manager processes.

vMotion (live migration) for GPU-enabled VMs is supported only for maintenance operations and not for vSphere Lifecycle Manager operations. vMotion allows the movement of running VMs to another host without downtime but has limitations with Lifecycle Manager updates.

vMotion, when used for GPU-enabled VMs, enables the live migration of these VMs to another host without downtime. This is specifically designed for maintenance tasks and does not apply to vSphere Lifecycle Manager operations.

Private AI Foundation with NVIDIA components offers flexibility and control over AI workloads on vSphere hosts. It allows for the development of AI applications (Retrieval-augmented Generation, data science) and the deployment of production-ready AI workloads on Tanzu Kubernetes Grid.

Using a trained language model to generate responses or complete tasks.

A GPU resource allocation method where VMs take turns using the GPU, sharing the resource based on predefined profiles.

A vGPU mode where a physical GPU is divided into smaller instances, allowing multiple VMs to each use a dedicated portion of the GPU.

A template for creating virtual machines that use GPUs, defining their configuration and resource requirements for Tanzu Kubernetes Grid.

GPUs are preferred over CPUs for accelerating computational workloads in machine learning due to their parallel processing capabilities, high throughput, and tolerance for memory latency.

Software installed on the ESXi host that provides the underlying foundation for NVIDIA vGPU functionality and management.

AI aims to mimic the intelligence or behaviour of humans and living entities. It's about creating systems that can reason, learn, and solve problems like humans do.

A GPU configuration mode where the entire GPU is dedicated to a single virtual machine (VM). This provides the VM with full access to the GPU's resources.

A GPU configuration mode that allows multiple VMs to share a physical GPU. It divides the GPU's resources among the VMs, allowing them to use the GPU concurrently.

Assigning an ESXi host to a specific workload domain cluster. This helps to group similar hosts together for efficient resource management and load balancing.

The ability to assign multiple GPUs to a single virtual machine (VM) using vSphere Device Groups.

A tool that manages the lifecycle of hosts with GPUs, ensuring all hosts within a cluster have the same GPU device and image.

A system specifically designed and tested to work optimally with NVIDIA GPUs.

Configuring an ESXi host to use NVIDIA GPUs, including adding devices, enabling SR-IOV, and installing NVIDIA drivers.

Adding an ESXi host to the vCenter inventory, making it part of your virtual data center.

Artificial Intelligence aims to replicate the thinking and behavior of humans or other living beings. It involves creating systems that can reason, learn, and solve problems like humans do.

Machine learning is a method where computers learn from data without needing explicit rules programmed. It involves training models on datasets to discover patterns and make predictions.

Deep learning is a specific technique within machine learning that draws inspiration from the network of neurons in our brain. It uses layers of interconnected nodes to learn complex patterns in data.

Generative AI is a type of AI that focuses on creating new content, like text, images, or music. It uses learned patterns to generate novel outputs.

A Graphics Processing Unit (GPU) is a specialized processor designed for parallel tasks, making it ideal for processing large amounts of data simultaneously.

GPUs excel at parallel processing, handling many tasks simultaneously, while CPUs are better at sequential tasks and managing low-latency operations.

In LLM training, pre-training involves feeding a model vast amounts of data to learn the fundamental structure and patterns of language.

Fine-tuning takes a pre-trained LLM and adjusts it for specific tasks, like generating different creative text formats, using specialized datasets.

Inference is when you use a trained LLM to generate responses or complete tasks based on prompts or questions. It uses its knowledge to provide answers.

GPUDirect RDMA allows direct communication between NVIDIA GPUs, bypassing the host CPU for faster data transfer. It leads to a 10x performance boost compared to traditional methods.

GPUs are preferred over CPUs for Machine Learning due to their specialized architecture designed for parallel processing. They have more cores, handle memory latency better, and focus on high throughput.

NVLINK is a high-speed interconnect that connects multiple GPUs on the same server, enabling fast communication and streamlined device management.

SR-IOV (Single Root I/O Virtualization) allows a single PCIe device, like a GPU, to be presented to the host as multiple independent devices, improving resource allocation and management.

A VGPU profile determines how GPU resources are allocated for a virtual machine (VM), either using time-slicing (sharing) or MIG mode (dedicated instances).

MIG (Multiple Instance GPU) mode allows a single physical GPU to be divided into multiple smaller instances, each dedicated to a particular virtual machine.

A VM class is a template used to create virtual machines that include GPUs for Tanzu Kubernetes Grid (TKG) workloads. It defines the configuration and resource requirements for the VM.

Commissioning a host involves adding it to the vCenter inventory, preparing it for use within the virtual data center environment.

ESXi host configuration is crucial for properly using NVIDIA GPUs. It involves adding devices, enabling acceleration features like SR-IOV, and installing the necessary NVIDIA drivers.

Latency refers to the delay in processing a task, while throughput measures the rate at which tasks are completed. CPUs excel at low latency, processing tasks sequentially, while GPUs are optimized for high throughput, handling many tasks in parallel.