This AI list of terms and definitions helps businesses and customers regardless of their technical expertise, to comprehend the significance of generative AI. Let’s explore some of the common AI terms and definitions that are crucial for understanding a number of AI-related concepts.
AI (Artificial Intelligence)
Simulation of human cognitive skills in order to perform tasks such as problem solving, learning, and self-correcting.
AI Accelerator
Hardware or system designed specifically for the purpose of executing AI-related functions. Examples include NVIDIA GPUs.
Algorithm
Instruction sequence designed for AI training, problem-solving, learning, and reasoning.
Ampere Architecture
NVIDIA’s GPU design made specifically for accelerating AI, big data analytics, and HPC.
Bare Metal
A server that is physically located on a user’s premises as opposed to being in a data center.
Benchmark
A reference point or standard measurement used for comparison.
Big Data
Large volumes of data that require complex processing in order to extract meaningful information about patterns.
Black Box
An AI system whose decision-making mechanism is opaque despite being able to produce desirable results.
Cluster
Several computers connected together in order to perform specific tasks as a single system.
Colocation
The practice of placing privately-owned servers at a co-location facility.
Compute
Processing capability offered in the cloud or through data centers.
Containerization
Virtualization at the application level. Enables isolation of apps in order to run them on different OSs.
Conversational AI
Technology capable of sustaining a conversation by utilizing natural language processing (NLP).
CUDA (Compute Unified Device Architecture)
Parallel computing software created by NVIDIA. It enables GPUs to perform general computing operations.
CUDA Core
Processing element responsible for executing CUDA instructions on NVIDIA GPUs.
Deep Learning
ML discipline which utilizes multi-layer neural networks in order to process and analyze data with high levels of abstraction.
DGX System
NVIDIA’s AI supercomputers capable of performing deep learning.
Finetuning
The process of improving the performance of an already trained AI model on a specific set of data.
FP16 (Half Precision Floating Point)
Standardized format of binary representation of real numbers which utilizes 16 bits. It is used for training AI models.
FP32 (Single Precision Floating Point)
Format of binary representation of real numbers encoded with 32 bits. Used in gaming and training AI models.
GDDR Memory
Video memory type which is typically found in GPUs. It is used for high-bandwidth data access needed in gaming and AI.
Generative AI
An AI paradigm which enables machines to create new data such as text, images, and videos from existing data.
GPU (Graphics Processing Unit)
Processor which is designed to perform parallel processing tasks. It is commonly used in AI and ML.
HPC (High-Performance Computing)
Broad discipline which focuses on using supercomputers to solve complex problems by employing large amounts of processing power.
Hypervisor
Computer software which enables a guest operating system to run on another host computer.
InfiniBand (IB)
Networking standard which enables high-performance computing, particularly large-scale parallel processing.
Internet of Things (IoT)
Network of physical objects which are connected and exchange data over the internet. AI enhances IoT devices’ capabilities, such as enabling them to make predictions.
IOPS (Input/Output Operations Per Second)
Metric used in measuring the performance of storage devices.
Kubernetes
Open-source software which enables the management of workloads running in containers.
Latency
Delay which occurs between the time that information is sent and the time that it is received.
LLM (Large Language Model)
An AI program which is capable of understanding and generating realistic-sounding text.
Load Balancing
The process of distributing workloads among resources in a manner that ensures optimal use.
Machine Learning (ML)
AI discipline which enables computer systems to learn from data using a set of mathematical algorithms.
MIG (Multi-Instance GPU)
Technology which allows for partitioning of GPUs into instances.
Natural Language Processing
AI technology which enables computers to understand, interpret, and generate human languages.
NIC (Network Interface Controller)
A circuit which manages a computer’s communication with other devices connected to the network.
Node
An autonomous computer in a network or cluster of computers which processes and stores data.
NVLink
Interconnect technology developed by NVIDIA which enables faster data transfer between GPUs and a CPU.
NVSwitch
Technology developed by NVIDIA which provides high-speed connectivity between GPUs at the switch level.
OOB (Out of Band)
Technique of managing network devices by using networks which are different from data networks.
Optimization
The process of maximizing performance by altering certain parameters of a system.
Pre-training
The process of training AI models which allows for the acquisition of general knowledge.
RAID (Redundant Array of Independent Disks)
Technology which enables combining several hardware disks into one logical unit for data storage purposes.
SLA (Service Level Agreement)
Agreement which stipulates the standards which a service provider must uphold.
SLM (Small Language Model)
Language model designed to perform simple language-related tasks.
Tensor
Mathematical object which is used in linear algebra. They can be represented in the form of tensors, hence the name tensor processing unit (TPU).
Tensor Core
Specialized processing core which is in NVIDIA GPUs and is used for AI processing. It performs key functions, such as matrix multiplication.
TensorRT
High-performance deep learning inference library from NVIDIA.
Text-to-Image Generator
AI paradigm which transforms text commands into an image.
Text-to-Speech
AI technology which converts text into speech.
Throughput
Measurement of how much information can be processed by a system per unit of time. Its standard unit is bits per second (bps).
Virtualization
Technology which allows for the emulation of a computer system in another computer system.
Volta Architecture
GPU design by NVIDIA which is used in most AI applications due to containing tensor cores.
Conclusion
Generative AI has the potential to turbo-charge employees, processes, customers and your services. Generative AI is so powerful that it can make almost every aspect of your business more efficient. Sales, service, marketing, and commerce are already leveraging the power of generative AI to bring more innovative and customised solutions to market, faster. By harnessing AI to handle the day-to-day of supporting customers, you can free up time for humans to drive creativity and collaboration, building those vital connections that only people can provide.
