gptkbp:instanceOf
|
artificial intelligence hardware system
|
gptkbp:announced
|
2021
|
gptkbp:countryOfOrigin
|
gptkb:United_States
|
gptkbp:designedFor
|
large language models
deep learning workloads
|
https://www.w3.org/2000/01/rdf-schema#label
|
CS-2 systems
|
gptkbp:main_processor
|
Wafer Scale Engine 2 (WSE-2)
|
gptkbp:manufacturer
|
gptkb:Cerebras_Systems
|
gptkbp:memoryBusWidth
|
20 PB/s
|
gptkbp:networkInterface
|
100 GbE
|
gptkbp:notableClient
|
gptkb:Argonne_National_Laboratory
gptkb:Lawrence_Livermore_National_Laboratory
gptkb:Pittsburgh_Supercomputing_Center
GlaxoSmithKline
|
gptkbp:notableFeature
|
massive parallelism
high bandwidth memory
largest chip ever built
single wafer-scale processor
|
gptkbp:peak_performance
|
up to 1.2 PFLOPS (FP16)
|
gptkbp:powerSource
|
20 kW
|
gptkbp:predecessor
|
CS-1 system
|
gptkbp:primaryUse
|
artificial intelligence training
artificial intelligence inference
|
gptkbp:rack_size
|
15U
|
gptkbp:supports
|
gptkb:TensorFlow
gptkb:PyTorch
gptkb:ONNX
|
gptkbp:WSE-2_core_count
|
850,000
|
gptkbp:WSE-2_memory
|
40 GB SRAM
|
gptkbp:WSE-2_process_technology
|
7nm
|
gptkbp:WSE-2_transistor_count
|
2.6 trillion
|
gptkbp:bfsParent
|
gptkb:C1_supercomputer
|
gptkbp:bfsLayer
|
8
|