Statements (134)
Predicate | Object |
---|---|
gptkbp:instanceOf |
gptkb:artificial_intelligence
gptkb:microprocessor application-specific integrated circuit Artificial intelligence hardware |
gptkbp:abbreviation |
gptkb:NPU
gptkb:TPU |
gptkbp:alternativeName |
AI_accelerator
AI_engine AI_processor Neural_Processing_Unit Neural_processing_unit Tensor_Processing_Unit inference_engine neural_processing_unit |
gptkbp:alternativeTo |
gptkb:graphics_card
Neural Engine |
gptkbp:announced |
2016
|
gptkbp:architecture |
ASIC
Parallel processing Matrix multiplication optimization |
gptkbp:availableOn |
gptkb:Google_Cloud_Platform
|
gptkbp:benefit |
Energy efficiency
On-device AI processing Lower latency for AI tasks |
gptkbp:category |
gptkb:microprocessor
gptkb:personal_computer Neural Engine specialized processor Specialized hardware |
gptkbp:competitor |
gptkb:NVIDIA_GPU
gptkb:AMD_GPU gptkb:Intel_Nervana |
gptkbp:contrastsWith |
gptkb:microprocessor
gptkb:graphics_card Central processing unit |
gptkbp:designedFor |
gptkb:artificial_intelligence
gptkb:machine_learning deep learning Machine learning acceleration Artificial intelligence workloads |
gptkbp:developedBy |
gptkb:Apple
gptkb:Google gptkb:Qualcomm gptkb:Huawei gptkb:Samsung gptkb:MediaTek |
gptkbp:enables |
privacy-preserving AI
on-device AI processing energy-efficient AI real-time AI inference |
gptkbp:example |
gptkb:Huawei_Kirin_NPU
gptkb:Qualcomm_Hexagon_AI_Processor gptkb:Samsung_NPU gptkb:Apple_Neural_Engine MediaTek APU |
gptkbp:fabricationProcess |
16nm (v2)
28nm (v1) 7nm (v4) |
gptkbp:feature |
low power consumption
parallel processing high throughput |
gptkbp:function |
accelerate neural network computations
optimize AI workloads Accelerate neural network computations Improve AI inference speed Reduce power consumption for AI tasks |
gptkbp:generation |
gptkb:TPU_v4
gptkb:TPU_v1 gptkb:TPU_v2 gptkb:TPU_v3 |
gptkbp:introducedIn |
2017
|
gptkbp:location |
gptkb:United_States
|
gptkbp:marketedAs |
mobile device manufacturers
Chipset vendors Smartphone manufacturers |
gptkbp:memoryType |
gptkb:HBM_(High_Bandwidth_Memory)
|
gptkbp:notableFeature |
low power consumption
high throughput custom hardware for AI |
gptkbp:openSource |
gptkb:Cloud_TPU_tools
|
gptkbp:operator |
matrix multiplication
quantized arithmetic |
gptkbp:platform |
gptkb:TensorFlow
|
gptkbp:relatedTo |
gptkb:microprocessor
gptkb:Google_Coral gptkb:TPU gptkb:Edge_TPU FPGA ASIC AI chip Tensor processing unit |
gptkbp:supports |
gptkb:Recurrent_neural_networks
Natural language processing convolutional neural networks Computer vision recurrent neural networks Deep learning transformer models Convolutional neural networks |
gptkbp:usedBy |
gptkb:Google_Assistant
gptkb:Google_Photos gptkb:Google_Search gptkb:Google_Translate |
gptkbp:usedFor |
augmented reality
natural language processing speech recognition image recognition face recognition object detection video analysis neural network training neural network inference |
gptkbp:usedIn |
gptkb:Google_data_centers
autonomous vehicles data centers smartphones Smartphones Data centers edge devices Edge devices |
gptkbp:website |
https://cloud.google.com/tpu
|
gptkbp:bfsParent |
gptkb:NPU
gptkb:Apple_A-series gptkb:Qualcomm_Snapdragon gptkb:Cyc gptkb:Apple_Silicon gptkb:Norman_Jouppi gptkb:Apple_silicon_(ARM) gptkb:Apple_M1 gptkb:Apple_M1_chip_(2020) gptkb:Apple_M3_chip_(2023) gptkb:Apple_M2_Ultra gptkb:Apple_M3 |
gptkbp:bfsLayer |
5
|