资源描述
|2023 04 20 AI TMT/AI AI AI AI AI GPT-1 1.17 GPT-3 1750 AI GPT-3 1 3 A100 5 AI GPU 1000 AI 2000 AI 70%AI 1300 2022 AI 1%GPU CPU AI AI GPU CPU 4 CPU+AI 2 CPU+8 GPU 4U 16 GPU AI GPU+CPU+80%AI GPU/CPU/DPU/DGX H100 GPU Grace CPU CPU BlueField DPU DGX AI DGX Cloud AI Foundations AI LLM AI CUDA PyTorch AI AI GPU CPU PCB/IC RAID GPU AI GPU GPU%456 9.2 81406 9.4 60808 8.3%1m 6m 12m 9.3 16.1 26.9 4.1 7.7 27.0 1 2023-04-04 2 PCB AI 2023-03-23 3 GTC 2023 AI iPhone AI 2023-03-23 S1090511060002 S1090521060001-20-10010203040Apr/22 Aug/22 Nov/22 Mar/23(%)电子 沪深300 AI 2.4 1 AI GPT.4 2.5 3.6 AI.10 1 AI 1%.10 2 CPU GPU+.12 AI.16 1 GPU CPU.16 2 CUDA AI.20.22 1.22 2 GPU.23 1 AI.4 2.4 3 A100 GPT-3 GPU.6 4 A100 GPT-3 GPU.6 5 A30 GPU.7 6 AI GPU AI.8 7 2020-2026.9 8.10 9.11 10.11 11.11 12 AI CAGR.11 13 22Q4.12 14 AI 21H1.12 ZXAZzQtOoNqRqOmPsPtMnM7NcM8OoMpPnPoNiNqQnRkPsQnN6MrQtPwMmRtMwMqRtN 3 15 CPU GPU.12 16 CPU GPU.12 17 K1 Power E950.13 18 Taishan 200 Pro.13 19 AI NF5468A5 GPU.14 20 AI NF5468A5 GPU.15 21 AI NF5468A5 GPU.15 22 GPU.16 23.16 24.17 25 H100 AI.17 26 Grace CPU.18 27 Grace CPU.18 28 Grace Hopper.18 29 Grace Hopper.18 30 BlueField-3 DPU.19 31 H100 GPU NVLink.19 32 NVSwitch DGX 8 GPU.19 33 DGX.20 34 AI Foundations.20 35.21 36 FY20Q1-23Q4 YOY.21 37 FY20Q1-23Q4.21 38 A.22 39 GPU.24 1.5 4 1 AI GPT AI AI 1 AI AI AI 4 NLP ELMo 9400 Megatron-Turing NLG 5300 5600 GPT-1 GPT-3 2018 6 GPT-1 GPT-1 BooksCorpus 1.17 5GB GPT-2 2019 2 15 40GB GPT-3 2020 5 1750 45TB AI AI GPU 2 AI 5 2 AI OpenAI GPT-3 GPT-4 GPT-4 GPT-3 M6 10000B GLaM 1000B 2023 3 AI ChatGPT AI ChatGPT 1 Google 137B LaMDA 2021 5 57.7d 2.81 T tokens 1024 TPU V3 260B ERNIE 3.0 Titan 2021 7 28 d 300B tokens 2048 V100 A21 Labs 137B Jurassic-I 2021 8-300B tokens 800 GPU 245B Yuan 1.0 2021 10-180B tokens 2128 GPU 530B MT-NLG 2021 10-270B tokens 4480 80G A100 Open AI 175B WebGPT 2021 12-Open AI 175B InstructGPT 2022 1-Salesforce 13B CodeGen 2022 3-577B tokens-Amazon 20B AlexaTM 2022 8 120 d 1.3T tokens 128 A100 130B GLM 2022 8-DeepMind 70B Sparrow 2022 9-64 TPUv3 Meta AI Papers with Code 120B Galactica 2022 11-106b tokens-Meta AI 65B LLaMA 2023 2 21d 1.4T tokens 2048 80G A100 360 1B 2023 3-Open AI-GPT-4 2023 3-INTERN 2.5 2023 3-A100 1085B 2023 3 100d 329B tokens 512 Ascend 910-K-PLUG-11B-Open AI 175B GPT-3 2022 5-300B tokens-20B UniLM-2 2022 4-1T tokens 512 TPU V4-2023-DA-Transformer 2023-HunYuan 2023 4-arxiv 6 3 AI AI 1 2 A100 A30 GPU GPU GPU GPU 3 A100 GPT-3 GPU GPT-3 GPT-3 Language Models are Few-Shot Learners GPT-3 175B 1746 3000 tokens 3.14*1023 Flops floating point operations A100 FP16 312 TFLOPS 100%GPU 11667 30%GPT-3 GPU 3 4 A100 GPT-3 GPU Language Models are Few-Shot Learners 7 A30 Tensor Core GPU FP16 Tensor Core 165 TFLOPS 1 10 10 token 1000 token 750 token Flops 2 GPT-3 1746 3.49*1023 A30 A100 GPU A100 GPU 24495 5 A30 GPU 5 AI 2027 AI GPU 1000 AI 2000 GPU GPU A100 2027 14 10 2023 GPU GPU ChatGPT AI Office 2027 14 2023 5 GPU 8 6 AI GPU AI GPU 100%100%GPU AI IDC 2022 58.5%42.4%+2023 2024 2025 2026 2027 10 15 20 25 30 1746 2000 2500 3000 3500 3000 3500 4000 4500 5000 6 6 6 6 6 3.14E+24 6.30E+24 1.20E+25 2.03E+25 3.15E+25 36375 72917 138889 234375 364583 312 312 312 312 312 116587 233707 445157 751202 1168536 14573 29213 55645 93900 1460672023 2024 2025 2026 2027 1 3 6 9 12 10 12 15 17 20 10 36 90 153 240 1000 1000 1000 1000 1000 2 2 2 2 2 1746 2000 2500 3000 3500 3.49E+23 1.44E+24 4.50E+24 9.18E+24 1.68E+25 165 165 165 165 165 24495 101010 315657 643939 1178451 6124 25253 78914 160985 294613 141081 334717 760813 1395141 2346988 20697 54466 134559 254885 440680 82 164 312 526 818 7 30 95 193 354 89 194 406 719 1172 146 292 556 939 1461 15 63 197 402 737 161 355 754 1341 2197 9 7 2020-2026 IDC 49%42%42%41%39%38%38%52%58%59%60%61%62%62%0%20%40%60%80%100%2020 2021 2022 2023 2024 2025 2026训练 推理 10 AI 1 AI 1%70%70%20%10%8 M5 AI OA ERP AI 70%20%10%11 9 10 M5 1300 AI 2022 1%AI MIC 2022 1362 TrendForce 2022 GPGPU General Purpose GPU AI 1%2023 ChatGPT AI 2023+8%20222026 10.8%AI TrendForce 11 12 AI CAGR MIC TrendForce AI IDC 22Q4 ODM 28.2%5%SuperMicro AI 21H1 AI 20%14%10%AI 0%1%2%3%4%5%6%7%020040060080010001200140016001800出货量（万台）yoy05101520252022 2023E 2024F 2025F 2026FCAGR=10.8%12 13 22Q4 14 AI 21H1 IDC IDC GPU CPU GPU 40%GPU CPU GPU V100 NIC CPU 2 2000 CPU 4 8500 GPU CPU 9000 44000 CPU 1000 900 GPU 300 CPU 35 Tesla V100 GPU 420 CPU 1350 GPU 810 CPU 40%15 CPU GPU 16 CPU GPU 2 CPU GPU+4 K1 Power E950 E950 4 scale-up POWER9 IO FSP PSU IO PCIe Gen4 FSP PSU 4+GPU+IO CPU GPU 28.2%14.5%11.8%7.0%7.0%5.0%4.7%4.1%2.9%2.2%12.7%ODM HPE SuperMicro IBM Cisco 20%14%10%6%5%4%4%2%1%1%33%HPE IBM Oracle 2000 2000-8500 4000 4000 1000 1000 9000 44000-13 17 K1 Power E950 CPU CPU CPU Taishan 200 Pro 2U4 CPU Taishan 200 Pro 920 8 DDR4 DIMM 4TB 4 CPU Hydra Fullmesh 18 Taishan 200 Pro AI GPU AI NF5468A5 GPU NF5468A5 AI AI GPU 4U 2 AMD EPYC 8 NVIDIA AMD Intel AI NF5468A5 CPU+GPU AI GPU GPU CPU 14 19 AI NF5468A5 GPU AI GPU CPU PCB NF5468A5 GPU 2 CPU 8 GPU AMD 7543 CPU 2.5 A100 40GB GPU 6 ZOL 32GB DDR4 RDIMM 2000 1.92TB NVMe 3000 2 CPU GPU CPU GPU 1 2 40GB A100 GPU 2 AMD 7502 CPU 2 30 GPU CPU 53%2*6+2*2=16 GPU AI GPU+CPU+80%15 20 AI NF5468A5 GPU AI GPU FPGA FPGA FPGA FPGA AI GPU FPGA GPU SIMD Single Instruction Multiple Data GPU CNN DNN RNN LSTM GPU FPGA FPGA AIHPC 95%5%FPGA FPGA 21 AI NF5468A5 GPU 16 AI 1 GPU CPU 1 GPU GPU Tesla GeForce RTX Quadro Tesla GPU Techpowerup GeForce RTX 4090 13000-16000 24GB GDDR6X 1008GB/s FP64 1290GFLOPS H100 PCIe 33000 80GB 2TB/s FP64 26TFLOPS NVLink 600GB/s GPU 2023 3 Ada Lovelace TSMC 4nm 22 GPU Techpowerup 23 Tesla Hopper H100 2022.3 Hopper TSMC 4nmTesla Ada L4,L40 2022.10 Ada Lovelace TSMC 4nmTesla Ampere A100,A800 2020.5-2022.11 AmpereTSMC 7nm;Samsung8nm A100 A800 TSMC 7nm Tesla Turing T4 T102018.9-2020 Turing TSMC 12nmTesla Volta V1002017.6-2019.11 Volta TSMC 12nmTesla Pascal P1002016.4-2017.3 Pascal TSMC 16nmGeForce RTX 40 2022.9-2023 Ada Lovelace TSMC 4nmGeForce RTX 30 2020.9-2022.12 Ampere Samsung 8nmGeForce RTX 20 2018.9-2021.12 Turing TSMC 12nmGeForce GTX 16 2019.2-2022.6 Turing TSMC 12nmQuadro Ada RTX 6000 Ada 2022.12 Ada Lovelace TSMC 4nmQuadro Ampere 2020.10-2022.3 Ampere Samsung 8nmQuadro Turing 2018.8-2021.5 Turing TSMC 12nmQuadro Volta 2018.3 Volta TSMC 12nmRTX Titan TITAN RTX 2018.12 Turing TSMC 12nm Data CenterGeForce RTX/QuadroHopper Ampere Turing Volta Tensor CoreFP64 TF32 bfloat16 FP16 FP8 INT8FP64 TF32 bfloat16 FP16 INT8 INT4 INT1FP16 INT8 INT4 INT1FP16 CUDA CoreFP64 TF32 FP16 bfloat16 INT8FP64 TF32 FP16 bfloat16 INT8FP64 FP32 FP16 INT8FP64 FP32 FP16 INT8 17 2016/2018/2021/2022 AI P100/V100/A100/H100 P100 H100 16GB 80GB 732GB/s 2TB/s FP64 FP32 FP16 2022 11 A800 A100 A800 GPU 400GB/s A100 NVLink GPU 600GB/s H100 Transformer Tensor Core AI A100 9 InfiniBand H100 A100 AI 30 24*25 H100 AI 2 CPU Grace CPU Grace CPU Superchip Grace Hopper Superchip Grace CPU NVLink-C2C Grace CPU Grace CPU 72 Arm Neoverse V2 Neoverse V2 Arm v9.0-A LPDDR5X ECC LPDDR5X DDR5 50%GB Grace CPU 144 Arm Neoverse V2 960GB LPDDR5X 500W 1TB/s H100 PCIe A100 80GB PCIe A800 80GB PCIe V100 PCIe P100 PCIe 16GB 2022.3 2021.6 2022.11 2018.5 2016.6 80GB HBM2e 80GB HBM2e 80GB HBM2e 32GB HBM2 16GB HBM2 2TB/s 1935GB/s 1935GB/s 900GB/s 732GB/sFP64 26 TFLOPS 9.7 TFLOPS 9.7 TFLOPS 7 TFLOPS 4.7 TFLOPSFP64 Tensor Core 51 TFLOPS 19.5 TFLOPS 19.5 TFLOPS-FP32 51 TFLOPS 19.5 TFLOPS 19.5 TFLOPS 14 TFLOPS 9.3 TFLOPSTensor Float 32 756 TFLOPS*156 TFLOPS|312 TFLOPS*156 TFLOPS|312 TFLOPS*-BFLOAT16 Tensor Core 1513 TFLOPS*312 TFLOPS|624TFLOPS*312 TFLOPS|624TFLOPS*-FP16 Tensor Core 1513 TFLOPS*312 TFLOPS|624 TFLOPS*312 TFLOPS|624 TFLOPS*-FP8 Tensor Core 3026 TFLOPS*-INT8 Tensor Core 3026 TOPS*624 TOPS|1248 TOPS*624 TOPS|1248 TOPS*-300-350W 300W 300W 250W 250WCUDA Core 14592 6912 6912 5120 3584Tensor Core 456 432 432 640-18 DNA Grace CPU AMD 2P Milan 7763 CPU 26 Grace CPU 27 Grace CPU H100 GPU Grace CPU NVLink Grace Hopper Grace Hopper GPU CPU AI Grace Hopper NVLink-C2C 900 GB/s PCIe Gen5 7 GPU Grace-Hopper NVIDIA HPC SDK NVIDIA AI NVIDIA Omniverse 28 Grace Hopper 29 Grace Hopper 3 DGX AI GPU DPU NVLink AI BlueField DPU CPU DPU Arm CPU BlueField DPU CPU DPU CPU BlueField-3 AI DPU BlueField-3 DPU 300 CPU CPU 19 30 BlueField-3 DPU NVLInk NVSwitch GPU NVLink GPU GPU H100 GPU 18 NVLInk 900GB/s PCIe 5.0 7 DXG H100 NVLInk NVSwitch NVLink GPU NVSwitch TSMC 4nm NVSwitch 64 NVLink NVIDIA SHARP 31 H100 GPU NVLink 32 NVSwitch DGX 8 GPU 8 H100 GPU 2 BlueField-3 DPU DGX H100 DGX AI DGX H100 8 H100 GPU GPU 640GB NVLink NVSwitches DGX H100 GPU 900GB/s x86 CPU AI 2 BlueField-3 DPU 8 ConnectX-7 InfiniBand 400GB/s DGX H100 DGX AI AI NVSwitch NVLink 256 GPU 57.6TB/s AI 20 33 DGX GTC 4 DGX Cloud AI Foundations DGX Cloud AI Foundations 2023GTC cuQuantum cuOpt CV-CUDA Parabricks cuLitho DGX Cloud AI NVIDIA DGX Cloud NVIDIA Enterprise AI AI Foundations LLM AI NVIDIA AI Foundations 34 AI Foundations GTC 2 CUDA AI GPU/CPU/DPU/DGX AI Tesla GPU 2023 3 H100 GPU Grace CPU CPU BlueField DPU H100 GPU BlueField-3 DPU NVLink DGX DGX AI AI DGX Cloud AI Foundations AI LLM AI CUDA 21 GPU CPU DPU CUDA CUDA TensorFlow PyTorch CUDA GPU CUDA CUDA 35 AIGC GPU H100 AI CSP 2022 2023 OEM 36 FY20Q1-23Q4 YOY 37 FY20Q1-23Q4-40%-20%0%20%40%60%80%100%02040608010020Q120Q220Q320Q421Q121Q221Q321Q422Q122Q222Q322Q423Q123Q223Q323Q4营收（亿美元）YoY0%20%40%60%80%20Q120Q220Q320Q421Q121Q221Q321Q422Q122Q222Q322Q423Q123Q223Q323Q4毛利率 净利率 22 1 OEM/ODM PCB GPU CPU PCB/BMC RAID GPU OEM/ODM OEM/ODM JDM OEM/ODM OEM/ODM JDM B G 38 A 23 2 GPU AI GPU GPU CPU GPU GPU GPU GPU&IP 370 290 270 370-X8 7nm FP16 96 TFLOPS 48GB 1000 2U 4 290 INT16 1 PetaOPS 128GB 2021 2022&CPU DCU DCU CPU DCU CPU 7000 5000 3000 7000 5000 3000 2023 DCU 8000 CUDA GPU GPU JM5 JM7 JM9 GPU CAD 22M6 JM9 X86 ARM MIPS Linux CPU 2001 863 973 2010 1 2 3 CPU 24 39 GPU CPU/GPU AI/ML AI AI DRAM 8 NAND 3 AI DDR5 HBM AI-GPU DRAM/NAND/HBM HBM FP32 FP16 INT16 INT8 MLU370-S4 7 18 72 96 192 24 307.2 75MLU370-X4 7 24 96 128 256 24 307.2 150MLU370-X8 7 24 96 128 256 48 614.4 250MLU290-M5 7-256 512 32 1228 350MLU270-S4 16-64 128 16 102 70MLU270-F4 16-64 128 16 102 150 7 32 1TB/s 260-350JM7200 28-10-40JM9-1.5-8 128-512GFLOPS-8 25.6 7 37 147 147 295 32 1228.8 250 7 24 96-384 32 800-7 240-1920 64 128 450-550 7 1024 32 819 300CAISA 28-10.9-28-10.9 8-56 28-43.6 32-230 MXN 7-80-160 16-12 32 128-256 32 1.6TB/s 300 12 32 128-256 32 1.6TB/s 300 12 20 80-16 512 225 12 22 86-16 560 300 12 32 128-256 16 819 150 12 17.6 70.4-70.4 16 512 150 7-128-256 32 512-7 32 128 128 256 16 512 150 7 32 128 128 256 32 512 160 7 32 128 128 256 32 512-14-64-256-512-14-128-512-14-256-512-7 14.4-16 448 255MTT S30-2.6-4-40MTT S10-2-2-MTT S50 12-8-75 7 15.2-32 448 250 12 10.6-42.4 32-150 25 DC/DC LDO DrMOS CPU GPU 5-6 8-12 DrMOS CPU GPU DrMOS SPD PCIe PCIe Retimer Switch USB PCB/AI PCIe5.0 HDI PCB CPU GPU AI PCB/AI I/O 400G/800G&DGX A100 640GB 6.5kw 10kw AI 20-50kW/AI AI A100 16 1+16 DrMOS AI A100 AI 1 AI 2 AI 3 4 AI 5 GPU CPU 26 1 2023-04-04 2 GTC 2023 AI iPhone AI 2023-03-23 3 PCB AI 2023-03-23 4 688183 2023-04-06 5 002938 2023-03-29 6 688486 2023-03-07 7 688141 2022 2023-02-27 8 FY23Q4 2023-02-24 9 002384 T 2023-02-01 10 002475 2022-11-17 11 300476 H1&2022-08-22 12 688041 x86 CPU DCU 2022-08-09 13 603228 Q2 H2 2022-08-08 14 002463+2022-03-24 27 14 08-11 11 TMT 11/12/14/15/16/17/19/20/21/22 2/5/2/2/4/3/3/4/3/5 11/12/14/15/16/17/18/19/20 2/4/1/2/3/3/2/3/3 10/14/15/16/17/18/19/20 TMT/1/2/3/3/3/3/2/2/1 2018/2019 2019/2020/2021 2020 2021 2021 2022 6-12 A 300 500 20%5-20%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
展开阅读全文