|
gptkbp:instanceOf
|
mainframe computer
|
|
gptkbp:announced
|
gptkb:Elon_Musk
2021
|
|
gptkbp:architecture
|
custom D1 chip
|
|
gptkbp:chipset
|
gptkb:D1
|
|
gptkbp:D1Announced
|
2021
|
|
gptkbp:D1Cores
|
354
|
|
gptkbp:D1DesignedBy
|
gptkb:Tesla,_Inc.
|
|
gptkbp:D1DieSize
|
645 mm²
|
|
gptkbp:D1InterconnectBandwidth
|
4 TB/s
|
|
gptkbp:D1Memory
|
576 MB SRAM
|
|
gptkbp:D1NotableFeature
|
energy efficiency
modular design
low latency
high throughput
FP16 and BF16 support
chip-to-chip interconnect
custom instruction set
custom software stack
direct liquid cooling
high compute density
high-speed I/O
in-package memory
large on-chip memory
multi-chip module
optimized for Tesla workloads
scalable mesh network
|
|
gptkbp:D1Package
|
BGA
|
|
gptkbp:D1Performance
|
362 teraflops
|
|
gptkbp:D1ProcessNode
|
7nm
|
|
gptkbp:D1Purpose
|
neural network training
|
|
gptkbp:D1TransistorCount
|
50 billion
|
|
gptkbp:D1Type
|
AI training chip
|
|
gptkbp:developedBy
|
gptkb:Tesla,_Inc.
|
|
gptkbp:firstRevealedAt
|
gptkb:Tesla_AI_Day_2021
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
Tesla Dojo
|
|
gptkbp:location
|
gptkb:Palo_Alto,_California
|
|
gptkbp:notableFeature
|
scalable architecture
high bandwidth interconnect
in-house chip design
|
|
gptkbp:platform
|
mainframe computer
Neural Engine
|
|
gptkbp:purpose
|
AI training
neural network training
|
|
gptkbp:relatedTo
|
gptkb:Tesla_Autopilot
gptkb:Tesla_AI_Day
gptkb:Tesla_FSD
|
|
gptkbp:status
|
in development
|
|
gptkbp:usedFor
|
gptkb:Tesla_Autopilot
gptkb:Full_Self-Driving_(FSD)
|
|
gptkbp:bfsParent
|
gptkb:Tesla_Autopilot
|
|
gptkbp:bfsLayer
|
5
|