Phones    |    Mobile SoCs    |    IoT Performance Ranking Desktop GPUs and CPUs
  Below are the preliminary results of the latest flagship chipsets obtained with AI Benchmark V5 Preview.
The scores of all SoCs would be improved greatly over the next few months as all vendors are now actively working on adding support for new NN ops and layers.
Processor CPU Cores AI Accelerator Year Lib CPU-Q
Score
CPU-F
Score
INT8
NNAPI 1.1
INT8
NNAPI 1.3
INT8
Accuracy
FP16
NNAPI 1.1
FP16
NNAPI 1.3
FP16
Accuracy
INT8
Parallel
FP16
Parallel
INT8
NLP
FP16
NLP
AI Score, K
MediaTek Dimensity 90001x3.05 GHz X2 & 3x2.85 GHz A710 & 4x A510APU 5.02021mm18599221605783035790992.33905015312093.718058124571182815463 692.5
Google Tensor2x2.8 GHz X1 & 2x2.25 GHz A76 & 4x1.8 GHz A55Google Tensor TPU2021nn1279712165357136929490.1297515827682.91506811857152306642 256.9
Exynos 21001x2.9 GHz X1 & 3x2.80 GHz A78 & 4x2.2 GHz A55GPU (Mali-G78 MP14)2021gg135481638466905861587.891246490795.91863359931986906 183.5
HiSilicon Kirin 90001x3.13GHz + 3x2.54GHz A77 & 4x2.04GHz A55GPU (Mali-G78 MP24)2020gg145231364155585272187.882396221895.91824350741997522 174.1
Google Tensor2x2.8 GHz X1 & 2x2.25 GHz A76 & 4x1.8 GHz A55GPU (Mali-G78 MP20)2021gg135431372261675547887.891635894695.91796324135346971 172.5
Snapdragon 8881x2.84 GHz & 3x2.42 GHz & 4x1.80 GHz Kryo 680GPU (Adreno 660)2020gg154701532064634698990.795355589989.31763350435277026 164.8
HiSilicon Kirin 90001x3.13GHz + 3x2.54GHz A77 & 4x2.04GHz A55NPU (3 Cores, Da Vinci series)2020nn142761367745063578687.8233982908889.314891027834937347 133.8
Snapdragon 8881x2.84 GHz & 3x2.42 GHz & 4x1.80 GHz Kryo 680DSP (Hex. 780) + GPU (Adreno 660)2020nn1626716043393638674.8114754852081.146365011715223 101.5
MediaTek Dimensity 1000+4x2.6 GHz Cortex A77 & 4x2.0 GHz Cortex A55APU 3.0 (6 cores)2019nn979710887154942768267.3116161686989.74033348519594722 95.3
MediaTek Dimensity 1000+4x2.6 GHz Cortex A77 & 4x2.0 GHz Cortex A55GPU (Mali-G77 MC9)2019gg6204714030752731887.846973182295.91085219611263026 86.6
Exynos 21001x2.9 GHz X1 & 3x2.80 GHz A78 & 4x2.2 GHz A55NPU + GPU (Mali-G78 MP14)2021nn129281465131541203297.95665964584.4672224521215368 68.1

xy  -  This device accelerates quantized [x] and floating-point [y] models using the:


n   -   Android NNAPI

g   -   TensorFlow Lite GPU Delegate  (OpenCL / OpenGL based)

h   -   Qualcomm Hexagon NN Direct Delegate

e   -   Samsung Eden Delegate

m  -   MediaTek Neuron Delegate

c   -   TensorFlow Lite / NNAPI default CPU backend


1  -  This SoC might be using unofficial / prototype hardware or drivers

2  -  These are the results of an early prototype. The results of the commercial SoC might be different

Copyright © 2021 by A.I.

ETH Zurich, Switzerland