Exam glossary
NVIDIA NCA-AIIO glossary
The terms you meet in the NVIDIA NCA-AIIO exam, with a short definition. The exam is in English: the German equivalent helps if you study in German. Portuguese, Spanish and French will follow.
39 terms · Definitions checked against official documentation on November 13, 2026 · Auf Deutsch
A
- Artificial intelligence(AI)#German: Künstliche Intelligenz (KI)
- The broad field of computer systems that perform tasks normally requiring human intelligence. Machine learning is a subset of AI, and deep learning is a subset of machine learning.
- Exam domain: Essential AI Knowledge
B
- Base Command Manager(BCM)#German: Base Command Manager
- NVIDIA software for provisioning, managing and monitoring AI clusters, including setup of Slurm or Kubernetes.
- Exam domain: AI Operations
C
- CUDA#German: CUDA
- NVIDIA's parallel computing platform and programming model for GPUs. Most of the NVIDIA AI software stack builds on CUDA and its libraries.
- Exam domain: Essential AI Knowledge
D
- Data Center GPU Manager(DCGM)#German: Data Center GPU Manager (DCGM)
- NVIDIA tool suite for monitoring, health checks and diagnostics of GPUs in clusters. Its metrics are often exported to Prometheus and Grafana.
- Exam domain: AI Operations
- Data processing unit(DPU)#German: Datenverarbeitungseinheit (DPU)
- A programmable processor on the network card that offloads networking, storage and security tasks from the host CPU. NVIDIA's DPUs are the BlueField family.
- Exam domain: AI Infrastructure
- Deep learning(DL)#German: Deep Learning (tiefes Lernen)
- Machine learning with neural networks of many layers. Its large matrix calculations are the main reason GPUs are used for AI.
- Exam domain: Essential AI Knowledge
- DGX#German: DGX
- NVIDIA's integrated AI systems: servers with multiple NVIDIA GPUs, NVLink, networking and software, delivered as one tested platform.
- Exam domain: AI Infrastructure
- DGX SuperPOD#German: DGX SuperPOD
- NVIDIA's reference architecture for large AI clusters built from DGX systems, with defined networking, storage and management software.
- Exam domain: AI Infrastructure
E
- East-west traffic#German: Ost-West-Verkehr
- Network traffic between servers inside the data center, such as GPU-to-GPU traffic during distributed training. North-south traffic flows in and out of the data center.
- Exam domain: AI Infrastructure
G
- GPU utilization#German: GPU-Auslastung
- The share of time a GPU is busy. High utilisation alone does not prove efficient use, so it is read together with memory use, power, temperature and error counters.
- Exam domain: AI Operations
- GPUDirect#German: GPUDirect
- NVIDIA technologies that let network cards (GPUDirect RDMA) and storage (GPUDirect Storage) move data straight to and from GPU memory, bypassing a copy through CPU memory.
- Exam domain: AI Infrastructure
- Graphics processing unit(GPU)#German: Grafikprozessor (GPU)
- A processor with thousands of cores built for running many calculations in parallel. A CPU has fewer, more complex cores optimised for sequential, general-purpose work.
- Exam domain: Essential AI Knowledge
H
- HGX#German: HGX
- NVIDIA's GPU baseboard platform that server manufacturers build their own AI servers around.
- Exam domain: AI Infrastructure
I
- Inference#German: Inferenz
- Running a trained model to produce predictions or answers. Inference is usually judged by latency, throughput and cost per request.
- Exam domain: Essential AI Knowledge
- InfiniBand(IB)#German: InfiniBand
- A low-latency, high-bandwidth network technology with native RDMA, widely used to connect the servers of AI training clusters. NVIDIA's InfiniBand family is called Quantum.
- Exam domain: AI Infrastructure
K
- Kubernetes(K8s)#German: Kubernetes
- Open-source platform for orchestrating containers. It is common for inference services and increasingly for training, with GPUs exposed through the NVIDIA GPU Operator.
- Exam domain: AI Operations
L
- Liquid cooling#German: Flüssigkeitskühlung
- Removing heat with liquid, for example cold plates on the processors (direct-to-chip). It handles the high rack power densities of modern GPU systems better than air.
- Exam domain: AI Infrastructure
M
- Machine learning(ML)#German: Maschinelles Lernen
- AI methods that learn patterns from data instead of following hand-written rules.
- Exam domain: Essential AI Knowledge
- Multi-Instance GPU(MIG)#German: Multi-Instance GPU (MIG)
- Partitioning one supported data center GPU into up to seven isolated instances, each with its own memory, cache and compute.
- Exam domain: AI Operations
N
- NGC catalog(NGC)#German: NGC-Katalog
- NVIDIA's hub for GPU-optimised containers, pretrained models, Helm charts and SDKs.
- Exam domain: Essential AI Knowledge
- NVIDIA AI Enterprise#German: NVIDIA AI Enterprise
- NVIDIA's licensed software platform for developing and running AI in production, with enterprise support. It includes frameworks, pretrained models and NIM microservices.
- Exam domain: Essential AI Knowledge
- NVIDIA GPU Operator#German: NVIDIA GPU Operator
- A Kubernetes operator that installs and manages the software GPUs need on each node: driver, container toolkit, device plugin and DCGM monitoring.
- Exam domain: AI Operations
- NVIDIA NIM(NIM)#German: NVIDIA NIM
- Prebuilt, containerised inference microservices that package a model with an optimised runtime and a standard API.
- Exam domain: Essential AI Knowledge
- NVIDIA Run:ai#German: NVIDIA Run:ai
- GPU orchestration software on Kubernetes that pools GPUs and shares them between teams with quotas and scheduling policies.
- Exam domain: AI Operations
- nvidia-smi#German: nvidia-smi
- Command-line tool that shows and manages the state of the NVIDIA GPUs in one server: utilisation, memory, temperature, power and running processes.
- Exam domain: AI Operations
- NVLink#German: NVLink
- NVIDIA's high-bandwidth GPU-to-GPU interconnect, much faster than PCIe, used inside a server and, with NVLink Switch, across a rack.
- Exam domain: AI Infrastructure
- NVSwitch#German: NVSwitch
- A switch chip that connects many GPUs over NVLink so that every GPU can talk to every other GPU at full NVLink speed.
- Exam domain: AI Infrastructure
P
- PCI Express(PCIe)#German: PCI Express
- The standard expansion bus that connects GPUs, network cards and storage to the CPU in a server.
- Exam domain: AI Infrastructure
- Power usage effectiveness(PUE)#German: PUE (Energieeffizienz-Kennzahl)
- Total facility power divided by the power used by IT equipment. A value closer to 1.0 means less energy goes to cooling and other overhead.
- Exam domain: AI Infrastructure
R
- RAPIDS#German: RAPIDS
- Open-source GPU-accelerated libraries for data science and data preparation, with APIs modelled on pandas and scikit-learn.
- Exam domain: Essential AI Knowledge
- RDMA over Converged Ethernet(RoCE)#German: RDMA über Ethernet (RoCE)
- RDMA carried over Ethernet networks. It needs a carefully configured, lossless or congestion-controlled Ethernet fabric to perform well.
- Exam domain: AI Infrastructure
- Remote direct memory access(RDMA)#German: Direkter Speicherzugriff über das Netz (RDMA)
- Moving data directly between the memory of two computers without involving their CPUs or operating systems, which lowers latency and CPU load.
- Exam domain: AI Infrastructure
S
- Slurm#German: Slurm
- Open-source workload manager common in HPC and AI training clusters. It queues batch jobs and allocates nodes and GPUs to them.
- Exam domain: AI Operations
- Spectrum-X#German: Spectrum-X
- NVIDIA's Ethernet platform for AI, combining Spectrum switches with BlueField SuperNICs to improve RoCE performance in AI clusters.
- Exam domain: AI Infrastructure
T
- Tensor Core#German: Tensor Core (Tensorkern)
- Specialised units inside NVIDIA GPUs that accelerate the matrix multiply-accumulate operations of deep learning, especially at reduced precision.
- Exam domain: Essential AI Knowledge
- TensorRT#German: TensorRT
- NVIDIA SDK that optimises trained models for faster inference on NVIDIA GPUs, for example through layer fusion and reduced precision.
- Exam domain: Essential AI Knowledge
- Training#German: Training
- The phase in which a model learns its parameters from data. Training large models needs many GPUs working together, fast GPU-to-GPU links and high storage throughput.
- Exam domain: Essential AI Knowledge
- Triton Inference Server#German: Triton Inference Server
- Open-source inference server that serves models from several frameworks on GPUs and CPUs. NVIDIA now names it NVIDIA Dynamo-Triton.
- Exam domain: Essential AI Knowledge
V
- Virtual GPU(vGPU)#German: Virtuelle GPU (vGPU)
- NVIDIA software that shares a physical GPU between virtual machines through the hypervisor.
- Exam domain: AI Operations