20230408_东方证券_计算机行业证券研究报告：AI兴起智能算力浪潮来袭_38页.pdf
|ChatGPT ChatGPT AI Bert GPT4 80%AI 30 100P 60%60%AI AI AI AI AI GPGPU TPU NPU AI AI 80%AI AI AI 5A 昇+AI AI AI AI AI AI AI(688008)(688041)-U(688256)AI AI(000977)(601138)(00992)(603019)(002261)(600839)AI A(000032)-W(688158)AI AI 2.昇.昇.昇.XVCXyRtOsRqRqOsRnQtMpO7N9R6MmOqQsQsRjMpPsOlOpOpP9PmMxOvPpNzQNZnPwP AI 3.AI 4.5 AI,.昇.昇.18 DCU.19 CPU DCU.20 Z100 NV A100 AMD MI100.昇.昇.昇.昇.AI 5.44 X785-G30 HPC/.45 X785-G40 GPU.46 5A.47.昇腾.AI 6 AI 1.1 ChatGPT ChatGPT J.Sevilla Compute Trends Across Three Eras of Machine Learning,2022 International Joint Conference on Neural Networks(IJCNN)2012 20 5-6 2015-2016 10 2 3(OOM)2022 ChatGPT AI Bert GPT4 GB 1 J.Sevilla,L.Heim,A.Ho,T.Besiroglu,M.Hobbhahn and P.Villalobos,Compute Trends Across Three Eras of Machine Learning,2022 International Joint Conference on Neural Networks(IJCNN),Padua,Italy,2022,pp.1-8,doi:10.1109/IJCNN55064.2022.9891914 AI 2022 AI AI IDC 2022-2023 2023 50%AI 2%AI 7 IDC 2026 ZFLOPS OpenAI 2012 AI 3.5 2 IDC 2021 155.2EFLOPS 2026 1,271.4EFLOPS 2021-2026 52.3%3 EFLOPS IDC AI AI 4 AI 0.00%2.00%4.00%6.00%8.00%10.00%12.00%14.00%0102030405060708090 2021 2022 31.775155.2268427640922.81271.402004006008001000120014002019 2020 2021 2022E 2023E 2024E 2025E 2026E(EFLOPS)AI 8 Int8 5 AI,1.2 51%2016 3%2021 51%IDC 2021 2026 52.3%18.5%6 2016-2021 AI 9 2022 45%28%1000GFlops IMB AI 2.1 2.1.1 80%AI AI 2020 4 20 2021 5 2022 2 17 8 10 AI 30 100P 850 ICPA 2022 3 95.00%47%3%51%2%2%0%20%40%60%80%100%2016 2021 AI 10 20 20.LuoJia AI 7 AI CPU GPU FPGA ASIC AI NVLink OAM AI AI 8 2.1.2 AI for Science AI 11 AI AI 60%60%AI 16(FP16)400PFLOPS 32(FP32)200PFLOPS 16(INT16)400POPS 2.2 2.2.1 昇 1000P 2023 2 昇 2023 2 17 2023 昇 昇 AI 昇+9 2023 2 昇 昇 100P 1000P 昇 100P 47 248P 500P 1000P AI 12 昇 MTGFinTech MTGFinTech 10 昇 2.2.2 AIDC AIDC 20 AIDC 13 5000 2.7 GPU 4910 Petaflops 200-300 Petaflops AIDC 20 AI AIDC 3%5%80%PUE 1.28 10%AIDC 11 AIDC AI 13 AI AI AI AI AI GPGPU AISC AI AI 2022()2027 2164 AI 80%AI AI AI AI+昇腾 AI AI AI AI AI AI 5A 昇+AI AI AI AI 3.1 AI AI AI CPU AI CPU+AI GPU、FPGA、ASIC AI CPU AI AI 2 CPU AI AI CPU AI GPGPU TPU NPU AI AI AI AI 2022 AI 385 AI 2027 AI 2164 AI 14 AI GPGPU 92%,AISC FPGA AI AI 2022 AI AI 47.2%52.8%2027 AI 23.7%76.3%AI AI AI 12 AI 13 2021 AI AI IDC 14 2022-2027 AI AI AI AI Nvidia AMD Google AI google TensorFlow TPU 昇 CANN Mindspore AI 80%AI IDC 2021 80 Nvidia 80%AMD Intel 1953273855668271210167521640.00%10.00%20.00%30.00%40.00%50.00%60.00%70.00%80.00%05001000150020002500 AI 91.90%7.80%0.30%GPGPU ASIC FPGA47.2%44.3%40.4%35.7%29.6%23.7%52.8%55.7%59.6%64.3%70.4%76.3%0.0%10.0%20.0%30.0%40.0%50.0%60.0%70.0%80.0%90.0%100.0%2022 2023E 2024E 2025E 2026E 2027EAI AI AI 15 15 2021 AI*IDC 3.1.1 GPU GPU A100 V100 H100 2021 AI 90%IDC 2021 80 80%AI 16 A100 V100 NVIDIA A800 H800 2022 10 A100 H100 12nm V100 GPU A100 A800 A100 A800 600GB/s 400GB/s A800 H100 H800 1 GPU 80%20%NvidiaAMD Intel AI 16/H100 A100 A800 V100 FP32 67 teraFLOPS 19.5 teraFLOPS 19.5 teraFLOPS 8.2 teraFLOPS FP16 Tensor Core 1979 teraFLOPS*624 teraFLOPS 624 teraFLOPS 16.4 teraFLOPS INT8 Tensor Core 3958 TOPS 1248 TOPS 1248 TOPS GPU 80GB 80GB 80GB 32GB GPU 3.35TB/s 2039 GB/s 2039 GB/s 1134 GB/s NVLink 900GB/s PCIe 5.0 128GB/s NVLi 600 GB/s PCIe 4.0 64 GB/s NVLi 400 GB/s PCIe 4.0 64 GB/s NVLi 300 GB/s PCIe 4.0 32 GB/s(TDP)700W 400W 400W 300W 2022.03 2020.03 2022.11 2017.5 NVIDIA CUDA Compute Unified Device Architecture GPU CUDA NVIDIA C/C+NVIDIA GPU CUDA GPU GPU CUDA CUDA 17 CUDA GPU Nvidia 3.1.2 CPU GPGPU CPU GPGPU CPU DCU,GPGPU CPU DCU AMD CPU DCU 18 DCU AI 17 CPU DCU CPU DCU 2018 DCU 2020 1 2022 6 19 CPU DCU DCU A100 AMD MI100 DCU HCZB-2021-ZB0364 DCU Z100 8192 FP64 10.8TFlops 32GB HBM2 AI A100 FP64 9.7 TFlops 40/80GB HBM2 AMD MI100 FP64 11.5 TFlops 32GB HBM2 20 Z100 NV A100 AMD MI100 AI 18 AMD DCU DCU AMD ROCm GPU TensorFlow Pytorch PaddlePaddle ROCm CUDA CUDA ROCm 2022 FP64 CPU DCU AI AI TensorFlow PyTorch Caffe2 AI 1000 3.1.3 AI 2016 AI 2016 AI 1A 2017 970 Mate10 AI MLU100 MLU270 MLU290 21 370 370 7nm Chiplet 256TOPS(INT8)AI 19 270 LPDDR5 270 370 Chiplet MLU370-S4 MLU370-X4 MLU370-X8 590 590 2022 WAIC 590 MLUarch05 590 PCIe 22 370 3.1.4 2011 6 AI 2021 FPGA 2017 FPGA 12000 2018 AI 2020,2021 AI 2022 23 AI 20 AI XPU-R 256TOPSINT8 128 TFLOPSFP16 14 AI 7nm GDDR6 AI 2-3 AI TensorFlow Pytorch PaddlePaddle 2022 AI AI 2024 24,200GB/s 25 AI R200 AI AI CPU AI AI AI AI AI 21 26 3.1.5+昇 昇 2004+昇+x86+GPU 昇 AI Atlas 昇腾 310 和昇腾 910 昇 310 NPU AI 16TOPSINT8,8TOPSFP16 AI 8W AI AI 昇 910 3D Cube AI FP16 320 TFLOPS INT8 640 TOPS AI Altas300 AI Atlas 300I Duo Atlas 300I Pro Atlas 300V Pro Atlas 300T Pro AI 27 昇 HUAWEI Ascend 310 28 昇 HUAWEI Ascend 910 AI 22 CANN AI 昇 AI CANN 2018 6.0 AI CANN 1400 AI 昇 AI CANN AI 昇 MindSpore PaddlePaddle PyTorch TensorFlow AI 900 29 CANN 昇 MindSpore AI ChatGPT 昇 MindSpore AI 2000.LuoJia 昇 AI 昇 MindSpore AI 30 昇 MindSpore 31 昇 MindSpore AI 23 3.2 AI GPU GPU IDC,2021 53.9 68.6%GPU 90%NPU ASIC FPGA GPU 43.8%11.6%6.3 IDC 2026 103.4 AI 2022 AI Meta 66%AI AI AI 6%2.3%1.5%1.5%0204060801001202021 2026E 88.40%11.60%GPU ASIC FPGA GPU AI 24 3.2.1 AI IT 2023 3 6 1036 2023 3 6 1768 B B302 36 37 19.0%17.0%16.0%14.0%6.0%2.3%1.5%1.5%22.7%Meta AI 25 AI AI IDC 2021H2 2021H1 AI 156 AI 20.9%3.6 pct 68.3%IDC 2021 H2 AI 52.5%5 AI 50%38 2021H1 AI 39 2021H2 AI IDC IDC 3.2.2 ODM 2022 X86 ARM 2022 ARM HPC HPC 40 20.9%13.0%9.8%6.1%4.8%3.9%3.9%2.6%1.2%1.0%32.6%HPE IBM Oracle 52.5%8.5%7.6%7.4%6.9%6.0%11.1%AI 26 NVIDIA HGX OVX CGX NVIDIA Grace CPU NVIDIA Grace Hopper Superchip NVIDIA Grace CPU 2 Grace Hopper Superchip NVIDIA Grace CPU NVIDIA Hopper GPU AI 10 NVIDIA Grace CPU NVLink-C2C LPDDRX CPU DIMM PCIe DC-SCM 41 NVIDIA Grace CPU Grace Hopper Superchip CSP CSP CSP AI 27 3.2.3 HPC TOP500 ISG 50PFLOPs 5 Maru Guru 2021 6 TOP500 Maru Guru 23 24 ThinkSystem SD650-V2 42 ThinkSystem 1 100 ThinkSystem SD630 V2 50 ThinkSystem SR670 V2 GPU 4 NVIDIA A100 Tensor Core GPU 2 IBM Spectrum Scale 10 PB Mellanox InfiniBand HDR 43 AI 28 3.2.4 2006 2014 X785-G30 X785-G40 44 X785-G30 HPC/45 X785-G40 GPU 5 7 50-5A 5A 46 5A 47 GPU 50 GPU 4 NVIDIA A100 GPU 100 CPU Platinum 8358 32C 2.6GHz I/O 10PB I/O 50GB/s 45GB/s HDR Infiniband 100Gb HDR Infiniband AI 29 3.2.5 AI AICC AI 昇 AICC 昇 910 AI Atlas 900 AI 昇 910 50 PC 256P 1024P FLOPS FP16 ResNet-50ImageNet 50KW 95%79%昇 AICC CANN AI MindSpore MindStudio 昇 MindX AI 48 昇腾 Atlas 900 AI AI AI 昇 昇 Atlas 900+昇 AI AI 30 I AI II II 1000P ops AI 64PB 200PB GB AI NLP III 49 50 II AIPerf500 3.2.6 昇 1996 2008 31 10+1500 2017 51 昇+AI AI 昇 AI AI AI AI AI 昇 昇 AI 31 昇 AI+8 33 52 3.2.7 2020 6+昇 昇 AI 300P AI 15 昇 100 200 53 54 AI AT800 Model 9000 AT800 Model 9000+昇 AI AT800 Model 9000 AI 32 ChatGPT AI AT800 2.56 PFLOPS FP16 8*100G RoCE v2 1070%1.3 2.56 PFLOPS/5.6 kW 3.3 AI DGX DGX NVIDIA DGX AI AI AI DGX A100 H100 80G Tensor Core GPU Azure Google GCP Oracle OCI 55 DGX AI AI AI AI IDC,2022 AI 74.6 2021 AI IDC 2022H1 AI AI AI 3.3.1 AI 2012 IaaS PaaS 31 11 CPU 56 AI 33 AI UCloud 2018 GPU 4.4-8.8kW GPU GPU A100 V100S MI100 AI 57 58 3.3.2 PKS 2021 PKS IaaS PaaS SaaS PKS P CPU K S CPU 60%AI 34 PKS CPU 74%87%59 PKS 2021 802.6 2023 1203.9+CPU AI 3.3.3 GPU Caffe TensorFlow Kubernetes Docker/AI 60 AI 35 AI AI AI(688008)(688041)-U(688256)AI AI(000977)(601138)(00992)(603019)(002261)(600839)AI A(000032)-W(688158)AI CPU GPU AI AI ChatGPT AI 36 12/15%5%15%-5%+5%-5%5%-5%+5%-5%前沿报告库是中国新经济产业咨询报告共享平台。行业范围涵盖新一代信息技术、5G、物联网、新能源、新材料、新消费、大健康、大数据、智能制造等新兴领域。为企事业单位、科研院所、投融资机构等提供研究和决策参考。扫一扫免费获取海量报告
