Statements (23)
| Predicate | Object |
|---|---|
| gptkbp:instanceOf |
application-specific integrated circuit
|
| gptkbp:announced |
2017
|
| gptkbp:architecture |
matrix multiply unit
|
| gptkbp:availableOn |
gptkb:Cloud_TPU
|
| gptkbp:connects |
high-speed network
|
| gptkbp:developedBy |
gptkb:Google
|
| gptkbp:formFactor |
PCIe card
|
| https://www.w3.org/2000/01/rdf-schema#label |
TPU v2
|
| gptkbp:memoryPerChip |
8 GB HBM
|
| gptkbp:performanceContext |
45 teraflops
|
| gptkbp:powerSource |
280W per board
|
| gptkbp:precededBy |
gptkb:TPU_v3
|
| gptkbp:successor |
gptkb:TPU_v1
|
| gptkbp:supports |
FP32
floating point operations bfloat16 |
| gptkbp:usedFor |
gptkb:artificial_intelligence
gptkb:machine_learning inference training neural networks |
| gptkbp:usedIn |
gptkb:Google_Cloud_Platform
|
| gptkbp:bfsParent |
gptkb:Neural_Engine
|
| gptkbp:bfsLayer |
6
|