Tensor Cores

E223691

Tensor Cores are specialized processing units in NVIDIA GPUs designed to accelerate matrix operations for deep learning and AI workloads.

All labels observed (1)

Label Occurrences
Tensor Cores canonical 1

How this entity was disambiguated

Statements (49)

Predicate Object
instanceOf GPU functional unit
hardware accelerator
specialized processing unit
accelerates convolution operations
matrix multiplication
neural network inference
neural network training
benefits energy efficiency
latency reduction
throughput
designedFor AI workloads
deep learning workloads
matrix operations
developedBy NVIDIA Corporation
surface form: NVIDIA
exposedThrough NVIDIA CUDA
surface form: CUDA

NVIDIA TensorRT
surface form: TensorRT

WMMA API
cuDNN
introducedIn Volta architecture
introducedWithGPU NVIDIA Pascal architecture
surface form: NVIDIA Tesla V100
optimizesFor mixed-precision computation
performsOperation matrix multiply-accumulate
presentInArchitecture Ada Lovelace
Ampere
Hopper
Turing (for selected generations)
surface form: Turing

Volta
requires GPU
surface form: CUDA-capable NVIDIA GPU

tensor-friendly data layout
supportsFeature sparsity acceleration
structured sparsity
supportsPrecision BF16
FP16
FP8
INT4
INT8
TF32
targetDomain data center AI
high-performance computing
real-time graphics AI effects
usedByFramework JAX
MXNet
PyTorch
TensorFlow
usedInProductLine NVIDIA A100
RTX
surface form: NVIDIA GeForce RTX

NVIDIA Ampere architecture
surface form: NVIDIA H100

RTX
surface form: NVIDIA RTX (professional)

NVIDIA Tesla data center GPUs
surface form: NVIDIA Tesla

How these facts were elicited

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

RTX includesFeature Tensor Cores