Ranked by tensor FP16/BF16 throughput, the precision most training actually runs at.
NVIDIA Vera Rubin Superchip leads with 4 x 10^15 FLOP/s, ahead of AMD Instinct MI355X (2.5 x 10^15 FLOP/s) and NVIDIA GB300 (Blackwell Ultra) (2.5 x 10^15 FLOP/s).
Peak throughput is a ceiling, not a promise: real training runs reach some fraction of it, and that utilization figure is tracked per model rather than per chip. Memory bandwidth is often the tighter constraint in practice.
| # | Name | Manufacturer | Released | FP16 throughput |
|---|---|---|---|---|
| 01 | NVIDIA Vera Rubin Superchip | NVIDIA | — | 4 x 10^15 |
| 02 | AMD Instinct MI355X | AMD | Jun 12, 2025 | 2.5 x 10^15 |
| 03 | NVIDIA GB300 (Blackwell Ultra) | NVIDIA | Aug 22, 2025 | 2.5 x 10^15 |
| 04 | NVIDIA GB200 | NVIDIA | Feb 15, 2025 | 2.5 x 10^15 |
| 05 | AMD Instinct MI350X | AMD | Jun 12, 2025 | 2.3 x 10^15 |
| 06 | Google TPU v7 Ironwood | Nov 6, 2025 | 2.3 x 10^15 | |
| 07 | NVIDIA B200 | NVIDIA | Nov 15, 2024 | 2.3 x 10^15 |
| 08 | NVIDIA B300 (Blackwell Ultra) | NVIDIA | Aug 22, 2025 | 2.3 x 10^15 |
| 09 | Intel Habana Gaudi3 | Intel | Sep 24, 2024 | 1.7 x 10^15 |
| 10 | AMD Instinct MI325X | AMD | Oct 10, 2024 | 1.3 x 10^15 |
| 11 | AMD Instinct MI300X | AMD | Dec 6, 2023 | 1.3 x 10^15 |
| 12 | Microsoft Maia 200 | Microsoft | Jan 26, 2026 | 1.3 x 10^15 |
| 13 | Biren BR100 | Biren | Aug 22, 2022 | 10^15 |
| 14 | NVIDIA GH100 | NVIDIA | Mar 21, 2023 | 9.9 x 10^14 |
| 15 | NVIDIA H200 SXM | NVIDIA | Nov 18, 2024 | 9.9 x 10^14 |
| 16 | NVIDIA GH200 | NVIDIA | Aug 8, 2023 | 9.9 x 10^14 |
| 17 | NVIDIA H100 SXM5 80GB | NVIDIA | Sep 20, 2022 | 9.9 x 10^14 |
| 18 | NVIDIA H800 SXM5 | NVIDIA | Mar 21, 2023 | 9.9 x 10^14 |
| 19 | AMD Instinct MI300A | AMD | Jan 4, 2023 | 9.8 x 10^14 |
| 20 | Google TPU v6e Trillium | May 14, 2024 | 9.2 x 10^14 | |
| 21 | Huawei Ascend 920 | Huawei | Oct 1, 2025 | 9 x 10^14 |
| 22 | Maia 100 (M100) | Microsoft | Aug 27, 2024 | 8 x 10^14 |
| 23 | Huawei Ascend 910C | Huawei | Oct 15, 2024 | 8 x 10^14 |
| 24 | NVIDIA H100 PCIe | NVIDIA | Sep 20, 2022 | 7.6 x 10^14 |
| 25 | Amazon Trainium3 | Amazon AWS | Dec 2, 2025 | 6.7 x 10^14 |
| 26 | Amazon Trainium2 | Amazon AWS | Dec 3, 2024 | 6.7 x 10^14 |
| 27 | AMD Radeon Instinct MI308X | AMD | Dec 6, 2023 | 6.5 x 10^14 |
| 28 | Meta MTIA 300 | Meta | Mar 11, 2026 | 6 x 10^14 |
| 29 | NVIDIA RTX Pro 6000 Blackwell | NVIDIA | Mar 18, 2025 | 5 x 10^14 |
| 30 | Google TPU v5p | Dec 6, 2023 | 4.6 x 10^14 | |
| 31 | Intel Habana Gaudi2 | Intel | May 10, 2022 | 4.5 x 10^14 |
| 32 | AMD Radeon Instinct MI250X | AMD | Nov 8, 2021 | 3.8 x 10^14 |
| 33 | Tesla D1 Dojo | Tesla | Aug 21, 2021 | 3.6 x 10^14 |
| 34 | NVIDIA GeForce RTX 4090 | NVIDIA | Sep 20, 2022 | 3.3 x 10^14 |
| 35 | NVIDIA A800 PCIe 80 GB | NVIDIA | Nov 3, 2022 | 3.1 x 10^14 |
| 36 | NVIDIA A100 PCIe | NVIDIA | Jun 22, 2020 | 3.1 x 10^14 |
| 37 | NVIDIA A100 SXM4 80 GB | NVIDIA | Nov 16, 2020 | 3.1 x 10^14 |
| 38 | NVIDIA A100 SXM4 40 GB | NVIDIA | May 14, 2020 | 3.1 x 10^14 |
| 39 | NVIDIA A100 | NVIDIA | Mar 1, 2020 | 3.1 x 10^14 |
| 40 | NVIDIA A800 PCIe 40 GB | NVIDIA | Nov 8, 2022 | 3.1 x 10^14 |
| 41 | NVIDIA A800 SXM | NVIDIA | Aug 11, 2022 | 3.1 x 10^14 |
| 42 | Google TPU v4 | May 20, 2021 | 2.8 x 10^14 | |
| 43 | Google TPU v5e | Aug 29, 2023 | 2 x 10^14 | |
| 44 | Amazon Trainium1 | Amazon AWS | Oct 15, 2022 | 1.9 x 10^14 |
| 45 | AWS Inferentia2 | Amazon AWS | Apr 13, 2023 | 1.9 x 10^14 |
| 46 | NVIDIA L40 | NVIDIA | Oct 13, 2022 | 1.8 x 10^14 |
| 47 | Meta MTIA v2 (MTIA 200) | Meta | Apr 10, 2024 | 1.8 x 10^14 |
| 48 | NVIDIA GeForce RTX 3090 Ti | NVIDIA | Mar 29, 2022 | 1.6 x 10^14 |
| 49 | NVIDIA A40 PCIe | NVIDIA | Oct 5, 2020 | 1.5 x 10^14 |
| 50 | NVIDIA HGX H20 | NVIDIA | Nov 9, 2023 | 1.5 x 10^14 |
| 51 | Google TPU v4i | Jan 1, 2020 | 1.4 x 10^14 | |
| 52 | NVIDIA Tesla V100S PCIe 32 GB | NVIDIA | Nov 26, 2019 | 1.3 x 10^14 |
| 53 | NVIDIA Tesla V100 SXM2 16 GB | NVIDIA | Jun 21, 2017 | 1.3 x 10^14 |
| 54 | NVIDIA Tesla V100 SXM2 32 GB | NVIDIA | Mar 27, 2018 | 1.3 x 10^14 |
| 55 | NVIDIA Tesla V100 DGXS 16 GB | NVIDIA | Mar 27, 2018 | 1.3 x 10^14 |
| 56 | NVIDIA V100 | NVIDIA | Jun 21, 2017 | 1.3 x 10^14 |
| 57 | NVIDIA Tesla V100 SXM2 | NVIDIA | Jul 1, 2017 | 1.3 x 10^14 |
| 58 | NVIDIA Tesla V100 SXM3 32 GB | NVIDIA | Mar 27, 2018 | 1.3 x 10^14 |
| 59 | NVIDIA Tesla V100 DGXS 32 GB | NVIDIA | Mar 27, 2018 | 1.3 x 10^14 |
| 60 | Google TPU v3 | May 18, 2018 | 1.2 x 10^14 |
