|
gptkbp:instance_of
|
gptkb:Graphics_Processing_Unit
|
|
gptkbp:bfsLayer
|
3
|
|
gptkbp:bfsParent
|
gptkb:Graphics_Processing_Unit
|
|
gptkbp:application
|
deep learning
|
|
gptkbp:architectural_style
|
gptkb:football_match
parallel processing
|
|
gptkbp:availability
|
gptkb:Google_Cloud
|
|
gptkbp:collaborations
|
academic institutions
research labs
|
|
gptkbp:connects
|
high-speed network
|
|
gptkbp:data_type
|
gptkb:Graphics_Processing_Unit
|
|
gptkbp:design
|
accelerate ML workloads
|
|
gptkbp:designed_by
|
gptkb:Job_Search_Engine
|
|
gptkbp:feedback
|
positive
|
|
gptkbp:focuses_on
|
AI and ML
|
|
gptkbp:form_factor
|
PC Ie card
|
|
gptkbp:fuel_capacity
|
high
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
TPU v2
|
|
gptkbp:impact
|
AI development
|
|
gptkbp:integrates_with
|
with Google services
|
|
gptkbp:involved_units
|
2
|
|
gptkbp:is_compatible_with
|
gptkb:Cloud_Computing_Service
|
|
gptkbp:is_scalable
|
high
|
|
gptkbp:is_used_for
|
gptkb:software_framework
|
|
gptkbp:language
|
gptkb:Graphics_Processing_Unit
gptkb:Library
|
|
gptkbp:market
|
leading in AI hardware
|
|
gptkbp:market_segment
|
enterprise
|
|
gptkbp:notable_users
|
gptkb:Google_services
|
|
gptkbp:part_of
|
Google AI ecosystem
|
|
gptkbp:performance
|
low
high
teraflops
ML Perf
up to 180 teraflops
|
|
gptkbp:power_consumption
|
up to 40 watts
|
|
gptkbp:predecessor
|
gptkb:TPU_v1
|
|
gptkbp:price
|
variable based on usage
|
|
gptkbp:primary_source
|
neural network training
|
|
gptkbp:project
|
ongoing
|
|
gptkbp:ram
|
8 GBHBM
|
|
gptkbp:release_date
|
gptkb:2017
|
|
gptkbp:released_in
|
gptkb:2017
|
|
gptkbp:research_focus
|
gptkb:Research_Institute
|
|
gptkbp:successor
|
gptkb:TPU_v3
|
|
gptkbp:supports
|
large datasets
float32 and bfloat16
|
|
gptkbp:target_market
|
AI researchers
|
|
gptkbp:technology
|
gptkb:Google_Cloud_AI
custom silicon
|
|
gptkbp:training
|
gptkb:Graphics_Processing_Unit
faster than GP Us
|
|
gptkbp:use_case
|
inference
|
|
gptkbp:user_base
|
developers and researchers
|