Statements (46)
| Predicate | Object |
|---|---|
| gptkbp:instanceOf |
gptkb:Graphics_Processing_Unit
|
| gptkbp:architect |
gptkb:ASIC
|
| gptkbp:availableIn |
gptkb:Google_Cloud
|
| gptkbp:competesWith |
NVIDIA GPUs
|
| gptkbp:designedBy |
gptkb:Google
|
| gptkbp:enables |
faster training
faster inference |
| gptkbp:features |
matrix multiplication
|
| https://www.w3.org/2000/01/rdf-schema#label |
TPU v3
|
| gptkbp:includes |
high-bandwidth memory
|
| gptkbp:isAvailableIn |
gptkb:TPU_Pods
multiple regions |
| gptkbp:isCompatibleWith |
Docker
Kubernetes |
| gptkbp:isDesignedFor |
high-performance computing
|
| gptkbp:isIntegratedWith |
gptkb:Google_Cloud_services
|
| gptkbp:isKnownFor |
energy efficiency
high performance cost-effectiveness |
| gptkbp:isOptimizedFor |
large-scale models
|
| gptkbp:isPartOf |
AI research projects
Google's_AI_hardware_offerings Google_AI_infrastructure |
| gptkbp:isSuitableFor |
to multiple chips
|
| gptkbp:isSupportedBy |
Google's_software_ecosystem
|
| gptkbp:isUsedBy |
large enterprises
research institutions startups |
| gptkbp:isUsedFor |
computer vision
natural language processing reinforcement learning |
| gptkbp:isUsedIn |
AI model training
AI model deployment |
| gptkbp:offers |
low latency
|
| gptkbp:performance |
deep learning
up to 420 teraflops |
| gptkbp:provides |
hardware acceleration
high throughput |
| gptkbp:RAM |
up to 128 GB
|
| gptkbp:releasedIn |
2018
|
| gptkbp:successor |
gptkb:TPU_v4
|
| gptkbp:supports |
gptkb:bfloat16
TensorFlow float32 |
| gptkbp:usedIn |
machine learning
|
| gptkbp:utilizes |
systolic array architecture
|