Humanity Statsv1
Ao vivo
176 registos

O hardware que torna a IA possivel

GPUs, TPUs and custom accelerators with their throughput at each precision, memory bandwidth, power draw, process node and price.

176 shown
Meta MTIA 300MetaMar 11, 20266×10¹⁴ FLOP/s201 GB800 W
Microsoft Maia 200MicrosoftJan 26, 20261.3×10¹⁵ FLOP/s201 GB750 W
Amazon Trainium3Amazon AWSDec 2, 20256.7×10¹⁴ FLOP/s—700 W
Google TPU v7 IronwoodGoogleNov 6, 20252.3×10¹⁵ FLOP/s179 GB960 W
Huawei Ascend 920HuaweiOct 1, 20259×10¹⁴ FLOP/s——
NVIDIA GB300 (Blackwell Ultra)NVIDIAAug 22, 20252.5×10¹⁵ FLOP/s268 GB1.4 kW
NVIDIA B300 (Blackwell Ultra)NVIDIAAug 22, 20252.3×10¹⁵ FLOP/s251 GB1.1 kW
AMD Instinct MI355XAMDJun 12, 20252.5×10¹⁵ FLOP/s268 GB1.4 kW
AMD Instinct MI350XAMDJun 12, 20252.3×10¹⁵ FLOP/s268 GB1.0 kW
NVIDIA RTX Pro 6000 BlackwellNVIDIAMar 18, 20255×10¹⁴ FLOP/s89.4 GB600 W
NVIDIA GB200NVIDIAFeb 15, 20252.5×10¹⁵ FLOP/s173 GB1.2 kW
Amazon Trainium2Amazon AWSDec 3, 20246.7×10¹⁴ FLOP/s89.4 GB500 W
NVIDIA H200 SXMNVIDIANov 18, 20249.9×10¹⁴ FLOP/s131 GB700 W
NVIDIA B200NVIDIANov 15, 20242.3×10¹⁵ FLOP/s168 GB1.0 kW
Cambricon Siyuan 590CambriconOct 31, 2024——200 W
Huawei Ascend 910CHuaweiOct 15, 20248×10¹⁴ FLOP/s119 GB700 W
AMD Instinct MI325XAMDOct 10, 20241.3×10¹⁵ FLOP/s238 GB1.0 kW
Intel Habana Gaudi3IntelSep 24, 20241.7×10¹⁵ FLOP/s119 GB900 W
Maia 100 (M100)MicrosoftAug 27, 20248×10¹⁴ FLOP/s—500 W
Google TPU v6e TrilliumGoogleMay 14, 20249.2×10¹⁴ FLOP/s29.8 GB380 W
Meta MTIA v2 (MTIA 200)MetaApr 10, 20241.8×10¹⁴ FLOP/s244 MB90 W
NVIDIA H100 NVLNVIDIAMar 14, 2024—87.5 GB400 W
MTT S4000Moore Threads (Tencent)Dec 19, 202310¹⁴ FLOP/s44.7 GB450 W
Google TPU v5pGoogleDec 6, 20234.6×10¹⁴ FLOP/s88.5 GB540 W
AMD Instinct MI300XAMDDec 6, 20231.3×10¹⁵ FLOP/s179 GB750 W
AMD Radeon Instinct MI308XAMDDec 6, 20236.5×10¹⁴ FLOP/s179 GB750 W
Sunway SW26010-ProSunwayNov 26, 2023—89.4 GB—
NVIDIA HGX H20NVIDIANov 9, 20231.5×10¹⁴ FLOP/s89.4 GB400 W
NVIDIA L2 PCleNVIDIANov 9, 20239.7×10¹³ FLOP/s22.4 GB—
NVIDIA L20 PCleNVIDIASep 16, 20231.2×10¹⁴ FLOP/s44.7 GB275 W
Google TPU v5eGoogleAug 29, 20232×10¹⁴ FLOP/s14.9 GB225 W
NVIDIA GH200NVIDIAAug 8, 20239.9×10¹⁴ FLOP/s—700 W
MetaX MXC500 (PCIe)MetaXJun 13, 2023—59.6 GB350 W
MetaX MXN100MetaXJun 7, 2023———
Meta MTIA v1 (MTIA 100)MetaMay 18, 20235.1×10¹³ FLOP/s122 MB25 W
AWS Inferentia2Amazon AWSApr 13, 20231.9×10¹⁴ FLOP/s29.8 GB—
NVIDIA H800 SXM5NVIDIAMar 21, 20239.9×10¹⁴ FLOP/s74.5 GB700 W
NVIDIA GH100NVIDIAMar 21, 20239.9×10¹⁴ FLOP/s—700 W
NVIDIA L4NVIDIAMar 21, 20231.2×10¹⁴ FLOP/s22.4 GB72 W
NVIDIA GeForce RTX 4070NVIDIAFeb 8, 2023—11.2 GB285 W
Intel Data Center GPU Max 1100IntelJan 10, 2023—44.7 GB300 W
Intel Data Center GPU Max 1350IntelJan 10, 2023—89.4 GB450 W
Intel Data Center GPU Max 1550IntelJan 10, 2023—119 GB600 W
AMD Radeon Instinct MI300AMDJan 4, 2023—119 GB600 W
AMD Instinct MI300AAMDJan 4, 20239.8×10¹⁴ FLOP/s119 GB760 W
Baidu Kunlun RG800Kunlunxin BaiduJan 1, 2023—29.8 GB130 W
MetaX MXC500 (OAM)MetaXJan 1, 2023—59.6 GB450 W
Tianshu Zhixin Zhikai 100Iluvatar CoreXDec 20, 2022———
NVIDIA RTX 6000 Ada GenerationNVIDIADec 3, 2022———
NVIDIA A800 PCIe 40 GBNVIDIANov 8, 20223.1×10¹⁴ FLOP/s—250 W
AMD Radeon RX 7900 XTXAMDNov 3, 2022—22.4 GB335 W
NVIDIA A800 PCIe 80 GBNVIDIANov 3, 20223.1×10¹⁴ FLOP/s—250 W
Amazon Trainium1Amazon AWSOct 15, 20221.9×10¹⁴ FLOP/s29.8 GB—
GroqChip LPU v1—Oct 15, 2022—219 MB215 W
NVIDIA L40NVIDIAOct 13, 20221.8×10¹⁴ FLOP/s44.7 GB300 W
NVIDIA H100 PCIeNVIDIASep 20, 20227.6×10¹⁴ FLOP/s74.5 GB350 W
NVIDIA GeForce RTX 4090NVIDIASep 20, 20223.3×10¹⁴ FLOP/s22.4 GB450 W
NVIDIA H100 SXM5 80GBNVIDIASep 20, 20229.9×10¹⁴ FLOP/s74.5 GB700 W
NVIDIA GeForce RTX 4080NVIDIASep 20, 2022—14.9 GB320 W
Biren BR100BirenAug 22, 202210¹⁵ FLOP/s59.6 GB550 W
NVIDIA A800 SXMNVIDIAAug 11, 20223.1×10¹⁴ FLOP/s74.5 GB400 W
MT-3000National University of Defense TechnologyJul 1, 2022—119 GB256 W
Intel Habana Gaudi2IntelMay 10, 20224.5×10¹⁴ FLOP/s89.4 GB600 W
NVIDIA GeForce RTX 3090 TiNVIDIAMar 29, 20221.6×10¹⁴ FLOP/s22.4 GB450 W
Cambricon MLU370-X8CambriconMar 21, 2022—44.7 GB250 W
Huawei Ascend 910BHuaweiMar 17, 2022—29.8 GB400 W
Baidu Kunlun R100 (75W)Kunlunxin BaiduJan 1, 2022—11.2 GB75 W
Baidu Kunlun R100 (100W)Kunlunxin BaiduJan 1, 2022—11.2 GB100 W
Baidu Kunlun R200 (150W)Kunlunxin BaiduJan 1, 2022—14.9 GB150 W
Baidu Kunlun R200-8F (160W)Kunlunxin BaiduJan 1, 2022—29.8 GB160 W
AMD Radeon Instinct MI250XAMDNov 8, 20213.8×10¹⁴ FLOP/s119 GB500 W
Cambricon MLU370-S4/S8CambriconNov 3, 2021—22.4 GB75 W
Cambricon MLU370-X4CambriconNov 3, 2021—22.4 GB150 W
Cambricon MLU370-S8CambriconNov 3, 2021—44.7 GB75 W
Cambricon MLU370-S4CambriconNov 3, 2021—22.4 GB75 W
Tesla D1 DojoTeslaAug 21, 20213.6×10¹⁴ FLOP/s—400 W
Baidu Kunlun IIKunlunxin BaiduAug 18, 2021—29.8 GB150 W
Shensuan-1HygonJul 1, 2021—29.8 GB350 W
Google TPU v4GoogleMay 20, 20212.8×10¹⁴ FLOP/s29.8 GB340 W
NVIDIA RTX A4000NVIDIAApr 12, 2021—14.9 GB140 W
NVIDIA A10GNVIDIAApr 12, 2021—22.4 GB150 W
NVIDIA RTX A5000NVIDIAApr 12, 2021—22.4 GB230 W
NVIDIA A10 PCIeNVIDIAApr 12, 2021—22.4 GB150 W
Tianshu Zhixin Tiangai 100 (BI-V100)Iluvatar CoreXMar 31, 2021—29.8 GB250 W
Cambricon MLU290-M5CambriconJan 21, 2021—29.8 GB350 W
NVIDIA A100 SXM4 80 GBNVIDIANov 16, 20203.1×10¹⁴ FLOP/s74.5 GB400 W
NVIDIA RTX A6000NVIDIAOct 5, 2020—44.7 GB300 W
NVIDIA A40 PCIeNVIDIAOct 5, 20201.5×10¹⁴ FLOP/s44.7 GB300 W
NVIDIA GeForce RTX 3080NVIDIASep 1, 2020—9.3 GB320 W
NVIDIA GeForce RTX 3090NVIDIASep 1, 2020—22.4 GB350 W
NVIDIA A100 PCIeNVIDIAJun 22, 20203.1×10¹⁴ FLOP/s37.3 GB300 W
NVIDIA A100 SXM4 40 GBNVIDIAMay 14, 20203.1×10¹⁴ FLOP/s37.3 GB400 W
NVIDIA A100NVIDIAMar 1, 20203.1×10¹⁴ FLOP/s37.3 GB400 W
Google TPU v4iGoogleJan 1, 20201.4×10¹⁴ FLOP/s7.5 GB175 W
Baidu Kunlun K200Kunlunxin BaiduJan 1, 2020—14.9 GB150 W
NVIDIA Tesla V100S PCIe 32 GBNVIDIANov 26, 20191.3×10¹⁴ FLOP/s29.8 GB250 W
T-Head Hanguang 800AlibabaSep 25, 2019——276 W
Huawei Ascend 910A PremiumHuaweiAug 23, 2019—29.8 GB—
Huawei Ascend 910AHuaweiAug 23, 2019—29.8 GB310 W
Huawei Ascend 910A ProHuaweiAug 23, 2019—29.8 GB—
Huawei Ascend 910HuaweiAug 23, 2019——310 W
Huawei Ascend 910 (320 TFLOPS)HuaweiAug 23, 2019——310 W
Cambricon MLU270-F4CambriconJun 20, 2019—14.9 GB150 W
Cambricon MLU270-S4CambriconJun 20, 2019—14.9 GB70 W
Intel Habana GAUDI HL-205IntelJun 18, 2019——300 W
Baidu Kunlun IKunlunxin BaiduJan 1, 2019——150 W
NVIDIA Quadro RTX 4000NVIDIANov 13, 2018—7.5 GB160 W
Huawei Ascend 310HuaweiOct 10, 2018——8 W
NVIDIA GeForce RTX 2080 Ti 11GBNVIDIASep 20, 20181.1×10¹⁴ FLOP/s10.2 GB250 W
NVIDIA Tesla T4NVIDIASep 13, 2018—14.9 GB70 W
NVIDIA Quadro RTX 5000NVIDIAAug 13, 2018—14.9 GB230 W
NVIDIA Quadro RTX 6000NVIDIAAug 13, 2018—22.4 GB260 W
NVIDIA Quadro RTX 8000NVIDIAAug 13, 2018—44.7 GB260 W
Google TPU v3GoogleMay 18, 20181.2×10¹⁴ FLOP/s29.8 GB450 W
NVIDIA Tesla V100 DGXS 16 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s14.9 GB250 W
NVIDIA Tesla V100 PCIe 32 GBNVIDIAMar 27, 20181.1×10¹⁴ FLOP/s29.8 GB250 W
NVIDIA Tesla V100 DGXS 32 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s29.8 GB250 W
NVIDIA Tesla V100 SXM3 32 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s—300 W
NVIDIA Tesla V100 SXM2 32 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s29.8 GB300 W
NVIDIA Titan VNVIDIADec 1, 2017—11.2 GB250 W
NVIDIA Tesla V100 SXM2NVIDIAJul 1, 20171.3×10¹⁴ FLOP/s14.9 GB300 W
NVIDIA Tesla V100 SXM2 16 GBNVIDIAJun 21, 20171.3×10¹⁴ FLOP/s—300 W
NVIDIA Tesla V100 PCIe 16 GBNVIDIAJun 21, 20171.1×10¹⁴ FLOP/s14.9 GB250 W
NVIDIA V100NVIDIAJun 21, 20171.3×10¹⁴ FLOP/s—300 W
PEZY-SC2PEZYJun 1, 2017——180 W
Google TPU v2GoogleMay 1, 20174.6×10¹³ FLOP/s14.9 GB280 W
NVIDIA TITAN XpNVIDIAApr 6, 2017—11.2 GB250 W
NVIDIA GeForce GTX 1080 TiNVIDIAMar 10, 2017—10.2 GB250 W
NVIDIA Quadro P600NVIDIAFeb 7, 2017—1.9 GB40 W
NVIDIA Quadro P4000NVIDIAFeb 6, 2017—7.5 GB105 W
MT-2000National University of Defense TechnologyJan 1, 2017——240 W
NVIDIA Quadro P5000NVIDIAOct 1, 2016—14.9 GB180 W
NVIDIA Quadro P6000NVIDIAOct 1, 2016—22.4 GB250 W
NVIDIA P40NVIDIASep 13, 2016—22.4 GB250 W
NVIDIA Tesla P4NVIDIASep 13, 2016—7.5 GB75 W
NVIDIA Tesla P100 PCIe 16GBNVIDIAJun 20, 2016——250 W
NVIDIA Tesla P100 PCIe 12GBNVIDIAJun 20, 2016——250 W
NVIDIA P100 PCIe 16GBNVIDIAJun 20, 2016—14.9 GB250 W
NVIDIA GeForce GTX 1080NVIDIAMay 27, 2016—7.5 GB180 W
NVIDIA Tesla P100 SXM2NVIDIAApr 5, 2016——300 W
NVIDIA P100NVIDIAApr 5, 2016—14.9 GB250 W
NVIDIA Tesla P100 DGXSNVIDIAApr 5, 2016——300 W
NVIDIA M40NVIDIANov 10, 2015—11.2 GB250 W
NVIDIA TESLA M60NVIDIAAug 30, 2015—14.9 GB300 W
NVIDIA Quadro M4000NVIDIAJun 29, 2015—7.5 GB120 W
Google TPU v1GoogleMay 15, 2015—8 GB75 W
NVIDIA GeForce GTX TITAN XNVIDIAMar 17, 2015—11.2 GB250 W
NVIDIA Quadro K1200NVIDIAJan 28, 2015—3.7 GB45 W
NVIDIA Tesla K80NVIDIANov 17, 2014—22.4 GB300 W
NVIDIA GeForce GTX 980NVIDIASep 19, 2014—3.7 GB165 W
NVIDIA GeForce GTX Titan BlackNVIDIAFeb 18, 2014—5.6 GB250 W
NVIDIA Tesla K40sNVIDIANov 22, 2013—11.2 GB245 W
NVIDIA Tesla K40tNVIDIANov 22, 2013—11.2 GB245 W
NVIDIA Quadro K6000NVIDIAJul 23, 2013—11.2 GB225 W
NVIDIA GeForce GTX 780NVIDIAMay 23, 2013—2.8 GB250 W
NVIDIA Quadro K4000NVIDIAMar 1, 2013—2.8 GB80 W
NVIDIA GeForce GTX TITANNVIDIAFeb 19, 2013—5.6 GB250 W
NVIDIA Tesla K20cNVIDIANov 12, 2012—4.7 GB225 W
NVIDIA Tesla K20XNVIDIANov 12, 2012——235 W
NVIDIA Tesla C2050NVIDIAJul 25, 2011——238 W
NVIDIA GeForce GTX 580NVIDIANov 9, 2010—1.4 GB244 W
NVIDIA GeForce GTX 280NVIDIAJun 16, 2008—977 MB236 W
Huawei Ascend 910B ProHuawei——29.8 GB—
Huawei Ascend 910B1 (v1)Huawei——59.6 GB—
Huawei Ascend 910B1 (v2)Huawei——59.6 GB—
Huawei Ascend 910B2Huawei——59.6 GB—
Huawei Ascend 910B3Huawei——59.6 GB—
Huawei Ascend 910B4Huawei——29.8 GB—
Huawei Ascend 800T A2Huawei———310 W
NVIDIA Rubin CPXNVIDIA——119 GB—
Huawei Ascend 950DTHuawei————
Huawei Ascend 970Huawei————
Huawei Ascend 960Huawei————
Cambricon Siyuan 690Cambricon————
NVIDIA Vera Rubin SuperchipNVIDIA—4×10¹⁵ FLOP/s268 GB—
NVIDIA VR300 (Rubin Ultra)NVIDIA————
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
Classificado
Os chips de IA mais rapidosOs chips de IA mais eficientes energeticamente
Mais em IA
Modelos de IAData centers de IAClusters GPUChips de IAEmpresas de IABenchmarks