GPUs, TPUs and custom accelerators with their throughput at each precision, memory bandwidth, power draw, process node and price.
| Meta MTIA 300 | Meta | Mar 11, 2026 | 6×10¹⁴ FLOP/s | 201 GB | 800 W |
| Microsoft Maia 200 | Microsoft | Jan 26, 2026 | 1.3×10¹⁵ FLOP/s | 201 GB | 750 W |
| Amazon Trainium3 | Amazon AWS | Dec 2, 2025 | 6.7×10¹⁴ FLOP/s | — | 700 W |
| Google TPU v7 Ironwood | Nov 6, 2025 | 2.3×10¹⁵ FLOP/s | 179 GB | 960 W | |
| Huawei Ascend 920 | Huawei | Oct 1, 2025 | 9×10¹⁴ FLOP/s | — | — |
| NVIDIA B300 (Blackwell Ultra) | NVIDIA | Aug 22, 2025 | 2.3×10¹⁵ FLOP/s | 251 GB | 1.1 kW |
| NVIDIA GB300 (Blackwell Ultra) | NVIDIA | Aug 22, 2025 | 2.5×10¹⁵ FLOP/s | 268 GB | 1.4 kW |
| AMD Instinct MI355X | AMD | Jun 12, 2025 | 2.5×10¹⁵ FLOP/s | 268 GB | 1.4 kW |
| AMD Instinct MI350X | AMD | Jun 12, 2025 | 2.3×10¹⁵ FLOP/s | 268 GB | 1.0 kW |
| NVIDIA RTX Pro 6000 Blackwell | NVIDIA | Mar 18, 2025 | 5×10¹⁴ FLOP/s | 89.4 GB | 600 W |
| NVIDIA GB200 | NVIDIA | Feb 15, 2025 | 2.5×10¹⁵ FLOP/s | 173 GB | 1.2 kW |
| Amazon Trainium2 | Amazon AWS | Dec 3, 2024 | 6.7×10¹⁴ FLOP/s | 89.4 GB | 500 W |
| NVIDIA H200 SXM | NVIDIA | Nov 18, 2024 | 9.9×10¹⁴ FLOP/s | 131 GB | 700 W |
| NVIDIA B200 | NVIDIA | Nov 15, 2024 | 2.3×10¹⁵ FLOP/s | 168 GB | 1.0 kW |
| Cambricon Siyuan 590 | Cambricon | Oct 31, 2024 | — | — | 200 W |
| Huawei Ascend 910C | Huawei | Oct 15, 2024 | 8×10¹⁴ FLOP/s | 119 GB | 700 W |
| AMD Instinct MI325X | AMD | Oct 10, 2024 | 1.3×10¹⁵ FLOP/s | 238 GB | 1.0 kW |
| Intel Habana Gaudi3 | Intel | Sep 24, 2024 | 1.7×10¹⁵ FLOP/s | 119 GB | 900 W |
| Maia 100 (M100) | Microsoft | Aug 27, 2024 | 8×10¹⁴ FLOP/s | — | 500 W |
| Google TPU v6e Trillium | May 14, 2024 | 9.2×10¹⁴ FLOP/s | 29.8 GB | 380 W | |
| Meta MTIA v2 (MTIA 200) | Meta | Apr 10, 2024 | 1.8×10¹⁴ FLOP/s | 244 MB | 90 W |
| NVIDIA H100 NVL | NVIDIA | Mar 14, 2024 | — | 87.5 GB | 400 W |
| MTT S4000 | Moore Threads (Tencent) | Dec 19, 2023 | 10¹⁴ FLOP/s | 44.7 GB | 450 W |
| Google TPU v5p | Dec 6, 2023 | 4.6×10¹⁴ FLOP/s | 88.5 GB | 540 W | |
| AMD Radeon Instinct MI308X | AMD | Dec 6, 2023 | 6.5×10¹⁴ FLOP/s | 179 GB | 750 W |
| AMD Instinct MI300X | AMD | Dec 6, 2023 | 1.3×10¹⁵ FLOP/s | 179 GB | 750 W |
| Sunway SW26010-Pro | Sunway | Nov 26, 2023 | — | 89.4 GB | — |
| NVIDIA HGX H20 | NVIDIA | Nov 9, 2023 | 1.5×10¹⁴ FLOP/s | 89.4 GB | 400 W |
| NVIDIA L2 PCle | NVIDIA | Nov 9, 2023 | 9.7×10¹³ FLOP/s | 22.4 GB | — |
| NVIDIA L20 PCle | NVIDIA | Sep 16, 2023 | 1.2×10¹⁴ FLOP/s | 44.7 GB | 275 W |
| Google TPU v5e | Aug 29, 2023 | 2×10¹⁴ FLOP/s | 14.9 GB | 225 W | |
| NVIDIA GH200 | NVIDIA | Aug 8, 2023 | 9.9×10¹⁴ FLOP/s | — | 700 W |
| MetaX MXC500 (PCIe) | MetaX | Jun 13, 2023 | — | 59.6 GB | 350 W |
| MetaX MXN100 | MetaX | Jun 7, 2023 | — | — | — |
| Meta MTIA v1 (MTIA 100) | Meta | May 18, 2023 | 5.1×10¹³ FLOP/s | 122 MB | 25 W |
| AWS Inferentia2 | Amazon AWS | Apr 13, 2023 | 1.9×10¹⁴ FLOP/s | 29.8 GB | — |
| NVIDIA GH100 | NVIDIA | Mar 21, 2023 | 9.9×10¹⁴ FLOP/s | — | 700 W |
| NVIDIA H800 SXM5 | NVIDIA | Mar 21, 2023 | 9.9×10¹⁴ FLOP/s | 74.5 GB | 700 W |
| NVIDIA L4 | NVIDIA | Mar 21, 2023 | 1.2×10¹⁴ FLOP/s | 22.4 GB | 72 W |
| NVIDIA GeForce RTX 4070 | NVIDIA | Feb 8, 2023 | — | 11.2 GB | 285 W |
| Intel Data Center GPU Max 1100 | Intel | Jan 10, 2023 | — | 44.7 GB | 300 W |
| Intel Data Center GPU Max 1350 | Intel | Jan 10, 2023 | — | 89.4 GB | 450 W |
| Intel Data Center GPU Max 1550 | Intel | Jan 10, 2023 | — | 119 GB | 600 W |
| AMD Instinct MI300A | AMD | Jan 4, 2023 | 9.8×10¹⁴ FLOP/s | 119 GB | 760 W |
| AMD Radeon Instinct MI300 | AMD | Jan 4, 2023 | — | 119 GB | 600 W |
| Baidu Kunlun RG800 | Kunlunxin Baidu | Jan 1, 2023 | — | 29.8 GB | 130 W |
| MetaX MXC500 (OAM) | MetaX | Jan 1, 2023 | — | 59.6 GB | 450 W |
| Tianshu Zhixin Zhikai 100 | Iluvatar CoreX | Dec 20, 2022 | — | — | — |
| NVIDIA RTX 6000 Ada Generation | NVIDIA | Dec 3, 2022 | — | — | — |
| NVIDIA A800 PCIe 40 GB | NVIDIA | Nov 8, 2022 | 3.1×10¹⁴ FLOP/s | — | 250 W |
| AMD Radeon RX 7900 XTX | AMD | Nov 3, 2022 | — | 22.4 GB | 335 W |
| NVIDIA A800 PCIe 80 GB | NVIDIA | Nov 3, 2022 | 3.1×10¹⁴ FLOP/s | — | 250 W |
| Amazon Trainium1 | Amazon AWS | Oct 15, 2022 | 1.9×10¹⁴ FLOP/s | 29.8 GB | — |
| GroqChip LPU v1 | — | Oct 15, 2022 | — | 219 MB | 215 W |
| NVIDIA L40 | NVIDIA | Oct 13, 2022 | 1.8×10¹⁴ FLOP/s | 44.7 GB | 300 W |
| NVIDIA H100 PCIe | NVIDIA | Sep 20, 2022 | 7.6×10¹⁴ FLOP/s | 74.5 GB | 350 W |
| NVIDIA GeForce RTX 4090 | NVIDIA | Sep 20, 2022 | 3.3×10¹⁴ FLOP/s | 22.4 GB | 450 W |
| NVIDIA H100 SXM5 80GB | NVIDIA | Sep 20, 2022 | 9.9×10¹⁴ FLOP/s | 74.5 GB | 700 W |
| NVIDIA GeForce RTX 4080 | NVIDIA | Sep 20, 2022 | — | 14.9 GB | 320 W |
| Biren BR100 | Biren | Aug 22, 2022 | 10¹⁵ FLOP/s | 59.6 GB | 550 W |
| NVIDIA A800 SXM | NVIDIA | Aug 11, 2022 | 3.1×10¹⁴ FLOP/s | 74.5 GB | 400 W |
| MT-3000 | National University of Defense Technology | Jul 1, 2022 | — | 119 GB | 256 W |
| Intel Habana Gaudi2 | Intel | May 10, 2022 | 4.5×10¹⁴ FLOP/s | 89.4 GB | 600 W |
| NVIDIA GeForce RTX 3090 Ti | NVIDIA | Mar 29, 2022 | 1.6×10¹⁴ FLOP/s | 22.4 GB | 450 W |
| Cambricon MLU370-X8 | Cambricon | Mar 21, 2022 | — | 44.7 GB | 250 W |
| Huawei Ascend 910B | Huawei | Mar 17, 2022 | — | 29.8 GB | 400 W |
| Baidu Kunlun R200 (150W) | Kunlunxin Baidu | Jan 1, 2022 | — | 14.9 GB | 150 W |
| Baidu Kunlun R100 (75W) | Kunlunxin Baidu | Jan 1, 2022 | — | 11.2 GB | 75 W |
| Baidu Kunlun R100 (100W) | Kunlunxin Baidu | Jan 1, 2022 | — | 11.2 GB | 100 W |
| Baidu Kunlun R200-8F (160W) | Kunlunxin Baidu | Jan 1, 2022 | — | 29.8 GB | 160 W |
| AMD Radeon Instinct MI250X | AMD | Nov 8, 2021 | 3.8×10¹⁴ FLOP/s | 119 GB | 500 W |
| Cambricon MLU370-S4/S8 | Cambricon | Nov 3, 2021 | — | 22.4 GB | 75 W |
| Cambricon MLU370-S8 | Cambricon | Nov 3, 2021 | — | 44.7 GB | 75 W |
| Cambricon MLU370-X4 | Cambricon | Nov 3, 2021 | — | 22.4 GB | 150 W |
| Cambricon MLU370-S4 | Cambricon | Nov 3, 2021 | — | 22.4 GB | 75 W |
| Tesla D1 Dojo | Tesla | Aug 21, 2021 | 3.6×10¹⁴ FLOP/s | — | 400 W |
| Baidu Kunlun II | Kunlunxin Baidu | Aug 18, 2021 | — | 29.8 GB | 150 W |
| Shensuan-1 | Hygon | Jul 1, 2021 | — | 29.8 GB | 350 W |
| Google TPU v4 | May 20, 2021 | 2.8×10¹⁴ FLOP/s | 29.8 GB | 340 W | |
| NVIDIA RTX A4000 | NVIDIA | Apr 12, 2021 | — | 14.9 GB | 140 W |
| NVIDIA A10G | NVIDIA | Apr 12, 2021 | — | 22.4 GB | 150 W |
| NVIDIA A10 PCIe | NVIDIA | Apr 12, 2021 | — | 22.4 GB | 150 W |
| NVIDIA RTX A5000 | NVIDIA | Apr 12, 2021 | — | 22.4 GB | 230 W |
| Tianshu Zhixin Tiangai 100 (BI-V100) | Iluvatar CoreX | Mar 31, 2021 | — | 29.8 GB | 250 W |
| Cambricon MLU290-M5 | Cambricon | Jan 21, 2021 | — | 29.8 GB | 350 W |
| NVIDIA A100 SXM4 80 GB | NVIDIA | Nov 16, 2020 | 3.1×10¹⁴ FLOP/s | 74.5 GB | 400 W |
| NVIDIA RTX A6000 | NVIDIA | Oct 5, 2020 | — | 44.7 GB | 300 W |
| NVIDIA A40 PCIe | NVIDIA | Oct 5, 2020 | 1.5×10¹⁴ FLOP/s | 44.7 GB | 300 W |
| NVIDIA GeForce RTX 3080 | NVIDIA | Sep 1, 2020 | — | 9.3 GB | 320 W |
| NVIDIA GeForce RTX 3090 | NVIDIA | Sep 1, 2020 | — | 22.4 GB | 350 W |
| NVIDIA A100 PCIe | NVIDIA | Jun 22, 2020 | 3.1×10¹⁴ FLOP/s | 37.3 GB | 300 W |
| NVIDIA A100 SXM4 40 GB | NVIDIA | May 14, 2020 | 3.1×10¹⁴ FLOP/s | 37.3 GB | 400 W |
| NVIDIA A100 | NVIDIA | Mar 1, 2020 | 3.1×10¹⁴ FLOP/s | 37.3 GB | 400 W |
| Baidu Kunlun K200 | Kunlunxin Baidu | Jan 1, 2020 | — | 14.9 GB | 150 W |
| Google TPU v4i | Jan 1, 2020 | 1.4×10¹⁴ FLOP/s | 7.5 GB | 175 W | |
| NVIDIA Tesla V100S PCIe 32 GB | NVIDIA | Nov 26, 2019 | 1.3×10¹⁴ FLOP/s | 29.8 GB | 250 W |
| T-Head Hanguang 800 | Alibaba | Sep 25, 2019 | — | — | 276 W |
| Huawei Ascend 910A Premium | Huawei | Aug 23, 2019 | — | 29.8 GB | — |
| Huawei Ascend 910 (320 TFLOPS) | Huawei | Aug 23, 2019 | — | — | 310 W |
| Huawei Ascend 910A Pro | Huawei | Aug 23, 2019 | — | 29.8 GB | — |
| Huawei Ascend 910A | Huawei | Aug 23, 2019 | — | 29.8 GB | 310 W |
| Huawei Ascend 910 | Huawei | Aug 23, 2019 | — | — | 310 W |
| Cambricon MLU270-F4 | Cambricon | Jun 20, 2019 | — | 14.9 GB | 150 W |
| Cambricon MLU270-S4 | Cambricon | Jun 20, 2019 | — | 14.9 GB | 70 W |
| Intel Habana GAUDI HL-205 | Intel | Jun 18, 2019 | — | — | 300 W |
| Baidu Kunlun I | Kunlunxin Baidu | Jan 1, 2019 | — | — | 150 W |
| NVIDIA Quadro RTX 4000 | NVIDIA | Nov 13, 2018 | — | 7.5 GB | 160 W |
| Huawei Ascend 310 | Huawei | Oct 10, 2018 | — | — | 8 W |
| NVIDIA GeForce RTX 2080 Ti 11GB | NVIDIA | Sep 20, 2018 | 1.1×10¹⁴ FLOP/s | 10.2 GB | 250 W |
| NVIDIA Tesla T4 | NVIDIA | Sep 13, 2018 | — | 14.9 GB | 70 W |
| NVIDIA Quadro RTX 6000 | NVIDIA | Aug 13, 2018 | — | 22.4 GB | 260 W |
| NVIDIA Quadro RTX 8000 | NVIDIA | Aug 13, 2018 | — | 44.7 GB | 260 W |
| NVIDIA Quadro RTX 5000 | NVIDIA | Aug 13, 2018 | — | 14.9 GB | 230 W |
| Google TPU v3 | May 18, 2018 | 1.2×10¹⁴ FLOP/s | 29.8 GB | 450 W | |
| NVIDIA Tesla V100 PCIe 32 GB | NVIDIA | Mar 27, 2018 | 1.1×10¹⁴ FLOP/s | 29.8 GB | 250 W |
| NVIDIA Tesla V100 DGXS 32 GB | NVIDIA | Mar 27, 2018 | 1.3×10¹⁴ FLOP/s | 29.8 GB | 250 W |
| NVIDIA Tesla V100 SXM3 32 GB | NVIDIA | Mar 27, 2018 | 1.3×10¹⁴ FLOP/s | — | 300 W |
| NVIDIA Tesla V100 SXM2 32 GB | NVIDIA | Mar 27, 2018 | 1.3×10¹⁴ FLOP/s | 29.8 GB | 300 W |
| NVIDIA Tesla V100 DGXS 16 GB | NVIDIA | Mar 27, 2018 | 1.3×10¹⁴ FLOP/s | 14.9 GB | 250 W |
| NVIDIA Titan V | NVIDIA | Dec 1, 2017 | — | 11.2 GB | 250 W |
| NVIDIA Tesla V100 SXM2 | NVIDIA | Jul 1, 2017 | 1.3×10¹⁴ FLOP/s | 14.9 GB | 300 W |
| NVIDIA Tesla V100 SXM2 16 GB | NVIDIA | Jun 21, 2017 | 1.3×10¹⁴ FLOP/s | — | 300 W |
| NVIDIA V100 | NVIDIA | Jun 21, 2017 | 1.3×10¹⁴ FLOP/s | — | 300 W |
| NVIDIA Tesla V100 PCIe 16 GB | NVIDIA | Jun 21, 2017 | 1.1×10¹⁴ FLOP/s | 14.9 GB | 250 W |
| PEZY-SC2 | PEZY | Jun 1, 2017 | — | — | 180 W |
| Google TPU v2 | May 1, 2017 | 4.6×10¹³ FLOP/s | 14.9 GB | 280 W | |
| NVIDIA TITAN Xp | NVIDIA | Apr 6, 2017 | — | 11.2 GB | 250 W |
| NVIDIA GeForce GTX 1080 Ti | NVIDIA | Mar 10, 2017 | — | 10.2 GB | 250 W |
| NVIDIA Quadro P600 | NVIDIA | Feb 7, 2017 | — | 1.9 GB | 40 W |
| NVIDIA Quadro P4000 | NVIDIA | Feb 6, 2017 | — | 7.5 GB | 105 W |
| MT-2000 | National University of Defense Technology | Jan 1, 2017 | — | — | 240 W |
| NVIDIA Quadro P5000 | NVIDIA | Oct 1, 2016 | — | 14.9 GB | 180 W |
| NVIDIA Quadro P6000 | NVIDIA | Oct 1, 2016 | — | 22.4 GB | 250 W |
| NVIDIA Tesla P4 | NVIDIA | Sep 13, 2016 | — | 7.5 GB | 75 W |
| NVIDIA P40 | NVIDIA | Sep 13, 2016 | — | 22.4 GB | 250 W |
| NVIDIA P100 PCIe 16GB | NVIDIA | Jun 20, 2016 | — | 14.9 GB | 250 W |
| NVIDIA Tesla P100 PCIe 12GB | NVIDIA | Jun 20, 2016 | — | — | 250 W |
| NVIDIA Tesla P100 PCIe 16GB | NVIDIA | Jun 20, 2016 | — | — | 250 W |
| NVIDIA GeForce GTX 1080 | NVIDIA | May 27, 2016 | — | 7.5 GB | 180 W |
| NVIDIA Tesla P100 SXM2 | NVIDIA | Apr 5, 2016 | — | — | 300 W |
| NVIDIA Tesla P100 DGXS | NVIDIA | Apr 5, 2016 | — | — | 300 W |
| NVIDIA P100 | NVIDIA | Apr 5, 2016 | — | 14.9 GB | 250 W |
| NVIDIA M40 | NVIDIA | Nov 10, 2015 | — | 11.2 GB | 250 W |
| NVIDIA TESLA M60 | NVIDIA | Aug 30, 2015 | — | 14.9 GB | 300 W |
| NVIDIA Quadro M4000 | NVIDIA | Jun 29, 2015 | — | 7.5 GB | 120 W |
| Google TPU v1 | May 15, 2015 | — | 8 GB | 75 W | |
| NVIDIA GeForce GTX TITAN X | NVIDIA | Mar 17, 2015 | — | 11.2 GB | 250 W |
| NVIDIA Quadro K1200 | NVIDIA | Jan 28, 2015 | — | 3.7 GB | 45 W |
| NVIDIA Tesla K80 | NVIDIA | Nov 17, 2014 | — | 22.4 GB | 300 W |
| NVIDIA GeForce GTX 980 | NVIDIA | Sep 19, 2014 | — | 3.7 GB | 165 W |
| NVIDIA GeForce GTX Titan Black | NVIDIA | Feb 18, 2014 | — | 5.6 GB | 250 W |
| NVIDIA Tesla K40s | NVIDIA | Nov 22, 2013 | — | 11.2 GB | 245 W |
| NVIDIA Tesla K40t | NVIDIA | Nov 22, 2013 | — | 11.2 GB | 245 W |
| NVIDIA Quadro K6000 | NVIDIA | Jul 23, 2013 | — | 11.2 GB | 225 W |
| NVIDIA GeForce GTX 780 | NVIDIA | May 23, 2013 | — | 2.8 GB | 250 W |
| NVIDIA Quadro K4000 | NVIDIA | Mar 1, 2013 | — | 2.8 GB | 80 W |
| NVIDIA GeForce GTX TITAN | NVIDIA | Feb 19, 2013 | — | 5.6 GB | 250 W |
| NVIDIA Tesla K20c | NVIDIA | Nov 12, 2012 | — | 4.7 GB | 225 W |
| NVIDIA Tesla K20X | NVIDIA | Nov 12, 2012 | — | — | 235 W |
| NVIDIA Tesla C2050 | NVIDIA | Jul 25, 2011 | — | — | 238 W |
| NVIDIA GeForce GTX 580 | NVIDIA | Nov 9, 2010 | — | 1.4 GB | 244 W |
| NVIDIA GeForce GTX 280 | NVIDIA | Jun 16, 2008 | — | 977 MB | 236 W |
| NVIDIA VR300 (Rubin Ultra) | NVIDIA | — | — | — | — |
| NVIDIA Vera Rubin Superchip | NVIDIA | — | 4×10¹⁵ FLOP/s | 268 GB | — |
| Huawei Ascend 910B3 | Huawei | — | — | 59.6 GB | — |
| Huawei Ascend 910B1 (v1) | Huawei | — | — | 59.6 GB | — |
| Huawei Ascend 970 | Huawei | — | — | — | — |
| Huawei Ascend 910B1 (v2) | Huawei | — | — | 59.6 GB | — |
| Huawei Ascend 800T A2 | Huawei | — | — | — | 310 W |
| Huawei Ascend 910B2 | Huawei | — | — | 59.6 GB | — |
| Huawei Ascend 910B Pro | Huawei | — | — | 29.8 GB | — |
| Cambricon Siyuan 690 | Cambricon | — | — | — | — |
| Huawei Ascend 960 | Huawei | — | — | — | — |
| Huawei Ascend 950DT | Huawei | — | — | — | — |
| NVIDIA Rubin CPX | NVIDIA | — | — | 119 GB | — |
| Huawei Ascend 910B4 | Huawei | — | — | 29.8 GB | — |
