TPU

E97074

A TPU (Tensor Processing Unit) is a specialized hardware accelerator designed by Google to efficiently perform large-scale machine learning and deep learning computations.

Aliases (8)

Statements (51)
Predicate Object
instanceOf AI accelerator
application-specific integrated circuit
tensor processing unit
architectureType systolic array
availableVia Google Cloud Platform
competesWith FPGA-based accelerators
GPU
componentOf Google Cloud AI infrastructure
deploymentModel cloud service
on-premise appliance (TPU pods)
designedBy Google
firstDeploymentContext Google data centers
firstIntroducedBy Google
hasGeneration TPU v1
TPU v2
TPU v3
TPU v4
TPU v5e
TPU v5lite
TPU v5p
keyFeature energy efficiency for ML workloads
high throughput for matrix operations
tight integration with Google data center network
manufacturer Google
offeredAs TPU VM
TPU node
TPU pod
optimizedFor large-scale linear algebra
matrix multiplication
tensor operations
programmedWith TensorFlow high-level APIs
XLA (Accelerated Linear Algebra)
supportsFramework JAX
PyTorch (via XLA or specialized backends)
TensorFlow
supportsNumericFormat bfloat16
float32
int8
targetWorkload computer vision models
natural language processing models
recommendation systems
speech recognition models
training large language models
usedByService Google Assistant
Google Photos
Google Search
Google Translate
usedFor deep learning workloads
machine learning workloads
neural network inference
neural network training

Referenced by (9)
Subject (surface form when different) Predicate
TPU ("TPU v3")
TPU ("TPU v4")
TPU ("TPU v5e")
TPU ("TPU v5p")
TPU ("TPU v5lite")
hasGeneration
TPU ("TPU VM")
TPU ("TPU pod")
offeredAs
Pixel 8 Pro ("Tensor G3 TPU")
neuralProcessingUnit
TensorFlow
supportsHardware

Please wait…