This page has a list of some of the data columns/fields that are included in the 2019 IEEE-HPEC paper:
A. Reuther, P. Michaleas, M. Jones, V. Gadepally, S. Samsi and J. Kepner, "Survey and Benchmarking of Machine Learning Accelerators," 2019 IEEE High Performance Extreme Computing Conference (HPEC), 2019, pp. 1-9, [IEEE Xplore doi:: 10.1109/HPEC.2019.8916327] [ArXiv.org/abs/1908.11348].
For the full dataset in CSV format, please download [peak_accelerators_ieee_hpec_2019.csv].
Company | Product | Label | Peak Perf. (GOPs/ GFLOPs) | Peak Power (W) | Precision | Form Factor | References |
---|---|---|---|---|---|---|---|
NVIDIA | Tesla K80 | K80 | 8740 | 300 | float | Card | [www.anandtech.com] |
NVIDIA | Pascal NVMe | P100 | 21200 | 300 | half | Card | [www.nvidia.com] [www.anandtech.com] |
NVIDIA | Volta NVMe | V100 | 125000 | 300 | AI | Card | [www.nvidia.com] [www.anandtech.com] |
NVIDIA | DGX-1 | DGX-1 | 960000 | 3500 | AI | System | [www.anandtech.com] |
NVIDIA | DGX-2 | DGX-2 | 1.92e+06 | 10000 | AI | System | [www.anandtech.com] |
NVIDIA | DGX Station | DGX-Station | 480000 | 1500 | AI | System | [clustervision.com] |
NVIDIA | Jetson1 | Jetson1 | 408 | 11.7 | AI | System | [devblogs.nvidia.com] |
NVIDIA | Jetson2 | Jetson2 | 580 | 12.8 | AI | System | [devblogs.nvidia.com] |
NVIDIA | Jetson Xavier | Xavier | 10000 | 30 | half | System | [www.extremetech.com] |
NVIDIA | Turing Quadro RTX 6000 | Turing | 131000 | 260 | int8 | Card | [devblogs.nvidia.com] |
GraphCore | C2 | GraphCoreC2 | 5000 | 300 | AI | Card | [www.graphcore.ai] [github.com] |
GraphCore | C2 | GraphCoreNode | 32000 | 2400 | AI | System | [www.graphcore.ai] [github.com] |
Wave Computing | DPU | Wave DPU | 180000 | 300 | AI | Card | [www.nextplatform.com] [github.com] |
Wave Computing | DPU | Wave System | 2.9e+06 | 2700 | AI | System | [www.top500.org] [github.com] |
Intel | Xeon Phi 7210F | Phi7210F | 2660 | 230 | double | Chip | [en.wikipedia.org] |
Intel | Xeon Phi 7290F | Phi7290F | 3460 | 260 | double | Chip | [en.wikipedia.org] |
Intel | Xeon Skylake SP (Scalable) | 2xSkylakeSP | 2620 | 410 | float | Chip | [software.intel.com] [ark.intel.com] |
Intel | Nervana Lake Crest | Nervana1 | 38000 | 210 | float | Card | [newsroom.intel.com] [www.hpcwire.com] [www.top500.org] |
Intel | Nervana NNP L-1000 (Spring Crest) | Nervana2 | 120000 | 210 | AI | Card | [newsroom.intel.com] [www.hpcwire.com] [www.top500.org] |
Intel | Movidius Myriad X | MovidiusX | 1000 | 2 | int16 | Chip | [www.extremetech.com] [thenewstack.io] |
Intel | Arria 10 1150 | Arria | 283000 | 85 | AI | Chip | [arxiv.org] [www.nextplatform.com] |
Habana | Goya HL-1000 | Goya | 57000 | 100 | AI | Chip | [www.top500.org] [www.cv-foundation.org] |
TPU1 | TPU1 | 23000 | 40 | int16 | Chip | [www.nextplatform.com] | |
TPU2 | TPU2 | 45000 | 250 | AI | Chip | [www.nextplatform.com] | |
TPU3 | TPU3 | 90000 | 200 | AI | Chip | [www.nextplatform.com] | |
TPU Edge | TPUedge | 58.5 | 2 | AI | System | [aiyprojects.withgoogle.com] | |
MIT | Eyeriss | Eyeriss | 32.7 | 0.45 | AI | Chip | [ieeexplore.ieee.org] [github.com] |
IBM | TrueNorth | TrueNorth | 1890 | 0.5 | int8 | System | [www.top500.org] [github.com] |
IBM | TrueNorth | TrueNorth | 1890 | 44 | int8 | System | [www.top500.org] [github.com] |
Rockchip | RK3399Pro | RK3399Pro | 2400 | 3 | 8bit | Chip | [www.rock-chips.com] [www.cnx-software.com] |
Baidu | Baidu Kunlun 818-300 | Baidu | 260000 | 100 | AI | Chip | [www.eetimes.com] [www.zdnet.com] |
AIStorm | AIStorm | AIStorm | 2500 | 0.227 | int8 | Chip | [www.eetimes.com] |
Xilinx / Tokyo Tech | Zynq XC7Z020 | Zynq-020 | 330 | 2.3 | 1-bit | System | [arxiv.org] [ieeexplore.ieee.org] |
Intel / Univ Sydney | GX1155 | GX1155 | 40800 | 48 | 1-bit | System | [arxiv.org] [ieeexplore.ieee.org] |
Fudan Univ | Zynq XC7Z020 | Zynq-020 | 410 | 2.26 | 2-bit | System | [arxiv.org] [ieeexplore.ieee.org] |
Xilinx / Tsinghua U | Zynq XC7Z020 | Zynq-020 | 84.3 | 3.5 | int8 | System | [arxiv.org] [ieeexplore.ieee.org] |
ASU / Intel | Intel Arria 10 GX1150 | Arria 10 | 645 | 21.2 | int16/int8 | System | [arxiv.org] [dl.acm.org] |
DeePhi / Stanford / Xilinx | Xilinx XCKU060 | Zynq-060 | 2520 | 41 | int16/int12 | System | [arxiv.org] [dl.acm.org] |
Peking Univ / UCLA / Xilinx | Xilinx XC7Z020 + XCV7VX620Tx6 | XilinxCluster | 1280 | 160 | int16 | System | [arxiv.org] [dl.acm.org] |
Univ Wisconsin / Intel | Intel Arria 10 GX1150 | Arria 10 | 1790 | 37.46 | int16 | System | [arxiv.org] [dl.acm.org] |
Peking Univ / Xilinx | Xilinx ZCU102 | ZCU102 | 2940 | 23.6 | int16 | System | [arxiv.org] [ieeexplore.ieee.org] |
Intel | Intel Arria 10 GX1150 | Arria 10 | 1380 | 45 | fp16 | System | [arxiv.org] [dl.acm.org] |
USC / Intel | Stratix-V | Stratix-V | 229 | 8.04 | int32 | System | [arxiv.org] [ieeexplore.ieee.org] |
Univ Wisconsin / Intel | Intel Arria 10 GX1150 | Arria 10 | 866 | 41.73 | float | System | [arxiv.org] [dl.acm.org] |
Cambricon | Cambricon MLU-100 | Cambricon | 166000 | 110 | int8 | Chip | [www.anandtech.com] [ieeexplore.ieee.org] |
Cambricon | Cambricon MLU-100 | Cambricon | 83200 | 110 | fp16 | Chip | [www.anandtech.com] [ieeexplore.ieee.org] |
DianNao | DianNao | DianNao | 452 | 0.485 | int16 | Chip | [cacm.acm.org] |
DianNao | DaDianNao | DaDianNao | 5590 | 15.97 | int16 | Chip | [cacm.acm.org] |
DianNao | ShiDianNao | ShiDianNao | 194 | 0.32 | int16 | Chip | [cacm.acm.org] |
DianNao | PuDianNao | PuDianNao | 1060 | 0.596 | int16 | Chip | [cacm.acm.org] |
Apple | Apple A12 Neural Engine | A12 | 624 | 5.5 | int8 | System | [www.anandtech.com] [medium.com] |
Qualcomm | Snapdragon 845 | S845 | 200 | 5.01 | int8 | System | [www.anandtech.com] |
QualComm | Snapdragon 835 | S835 | 130 | 3.79 | int8 | System | [www.anandtech.com] |
Huawei | Kirin 970 (Mali--75) | Mali-75 | 261 | 6.33 | int8 | System | [www.anandtech.com] |
Huawei | Kirin 980 (Mali-76) | Mali-76 | 468 | 5 | int8 | System | [www.anandtech.com] |
Groq | Groq | Groq | 400000 | 300 | int16 | Card | [www.electronicdesign.com] |
AMD | Radeon Instinct MI60 | AMD-MI60 | 29500 | 300 | fp16 | Card | [www.amd.com] |
AMD | Radeon Instinct MI6 | AMD-MI6 | 5730 | 150 | fp16 | Card | [www.amd.com] |
Tesla | Tesla Full Self-Driving Computer | Tesla | 72000 | 72 | int8 | [www.youtube.com] | |
ASU / Intel | Stratix-V 5SGSD8 | Stratix-V | 118 | 19.1 | int16/int8 | System | [arxiv.org] [dl.acm.org] |
NUDT / Xilinx | Virtex7 XC7VX690T | Virtex7 | 222 | 24.8 | int16/int8 | System | [arxiv.org] [ieeexplore.ieee.org] |
Imperial College | Xilinx XC7Z020 | Zynq-020 | 12.7 | 1.75 | int16 | System | [arxiv.org] [dl.acm.org] |
Tsinghua / Xilinx | Xilinx Zynq XC7Z045 | Zynq-045 | 137 | 9.63 | int16 | System | [arxiv.org] [dl.acm.org] |
Peking Univ / Xilinx | Xilinx XC7Z045 | Zynq-045 | 230 | 9.4 | int16 | System | [arxiv.org] [dl.acm.org] |
Peking Univ / UCLA / Xilinx | Xilinx XC7VX690T | Virtex7 | 354 | 26 | int16 | System | [arxiv.org] [ieeexplore.ieee.org] |
Peking Univ / UCLA / Intel | Stratix-V 5SGSMD5 | Stratix-V | 364 | 25 | int16 | System | [arxiv.org] [ieeexplore.ieee.org] |
Fudan Univ / Xilinx | Xilinx XC7VX690T | Virtex7 | 566 | 30.2 | int16 | System | [arxiv.org] [ieeexplore.ieee.org] |
NUDT / Xilinx | Xilinx XC7VX690T | Virtex7 | 431 | 25 | int16 | System | [arxiv.org] [dl.acm.org] |
NUDT / Xilinx | Xilinx XCVU440 | Xilinx-440 | 785 | 26 | int16 | System | [arxiv.org] [dl.acm.org] |
Peking Univ / UCLA / Xilinx | Xilinx XC7VX485T | XC-485T | 7.26 | 19.63 | float | System | [arxiv.org] [ieeexplore.ieee.org] |
Peking Univ / UCLA / Xilinx | Xilinx XC7VX485T | XC-485T | 61.6 | 18.61 | float | System | [arxiv.org] [dl.acm.org] |
USC / Intel | Stratix V | Stratix-V | 124 | 13.18 | float | System | [arxiv.org] [dl.acm.org] |
Copyright 2021 MIT, Albert I. Reuther