Humanity Statsv1
实时
Top 60

速度最快的 AI 芯片

按原始算力吞吐排序。

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 GB200 (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.

#NameManufacturerReleasedFP16 throughput
01NVIDIA Vera Rubin SuperchipNVIDIA—4×10¹⁵ FLOP/s
02AMD Instinct MI355XAMDJun 12, 20252.5×10¹⁵ FLOP/s
03NVIDIA GB200NVIDIAFeb 15, 20252.5×10¹⁵ FLOP/s
04NVIDIA GB300 (Blackwell Ultra)NVIDIAAug 22, 20252.5×10¹⁵ FLOP/s
05AMD Instinct MI350XAMDJun 12, 20252.3×10¹⁵ FLOP/s
06Google TPU v7 IronwoodGoogleNov 6, 20252.3×10¹⁵ FLOP/s
07NVIDIA B300 (Blackwell Ultra)NVIDIAAug 22, 20252.3×10¹⁵ FLOP/s
08NVIDIA B200NVIDIANov 15, 20242.3×10¹⁵ FLOP/s
09Intel Habana Gaudi3IntelSep 24, 20241.7×10¹⁵ FLOP/s
10AMD Instinct MI300XAMDDec 6, 20231.3×10¹⁵ FLOP/s
11AMD Instinct MI325XAMDOct 10, 20241.3×10¹⁵ FLOP/s
12Microsoft Maia 200MicrosoftJan 26, 20261.3×10¹⁵ FLOP/s
13Biren BR100BirenAug 22, 202210¹⁵ FLOP/s
14NVIDIA GH200NVIDIAAug 8, 20239.9×10¹⁴ FLOP/s
15NVIDIA GH100NVIDIAMar 21, 20239.9×10¹⁴ FLOP/s
16NVIDIA H200 SXMNVIDIANov 18, 20249.9×10¹⁴ FLOP/s
17NVIDIA H100 SXM5 80GBNVIDIASep 20, 20229.9×10¹⁴ FLOP/s
18NVIDIA H800 SXM5NVIDIAMar 21, 20239.9×10¹⁴ FLOP/s
19AMD Instinct MI300AAMDJan 4, 20239.8×10¹⁴ FLOP/s
20Google TPU v6e TrilliumGoogleMay 14, 20249.2×10¹⁴ FLOP/s
21Huawei Ascend 920HuaweiOct 1, 20259×10¹⁴ FLOP/s
22Huawei Ascend 910CHuaweiOct 15, 20248×10¹⁴ FLOP/s
23Maia 100 (M100)MicrosoftAug 27, 20248×10¹⁴ FLOP/s
24NVIDIA H100 PCIeNVIDIASep 20, 20227.6×10¹⁴ FLOP/s
25Amazon Trainium3Amazon AWSDec 2, 20256.7×10¹⁴ FLOP/s
26Amazon Trainium2Amazon AWSDec 3, 20246.7×10¹⁴ FLOP/s
27AMD Radeon Instinct MI308XAMDDec 6, 20236.5×10¹⁴ FLOP/s
28Meta MTIA 300MetaMar 11, 20266×10¹⁴ FLOP/s
29NVIDIA RTX Pro 6000 BlackwellNVIDIAMar 18, 20255×10¹⁴ FLOP/s
30Google TPU v5pGoogleDec 6, 20234.6×10¹⁴ FLOP/s
31Intel Habana Gaudi2IntelMay 10, 20224.5×10¹⁴ FLOP/s
32AMD Radeon Instinct MI250XAMDNov 8, 20213.8×10¹⁴ FLOP/s
33Tesla D1 DojoTeslaAug 21, 20213.6×10¹⁴ FLOP/s
34NVIDIA GeForce RTX 4090NVIDIASep 20, 20223.3×10¹⁴ FLOP/s
35NVIDIA A800 PCIe 80 GBNVIDIANov 3, 20223.1×10¹⁴ FLOP/s
36NVIDIA A100 PCIeNVIDIAJun 22, 20203.1×10¹⁴ FLOP/s
37NVIDIA A100 SXM4 80 GBNVIDIANov 16, 20203.1×10¹⁴ FLOP/s
38NVIDIA A100NVIDIAMar 1, 20203.1×10¹⁴ FLOP/s
39NVIDIA A100 SXM4 40 GBNVIDIAMay 14, 20203.1×10¹⁴ FLOP/s
40NVIDIA A800 SXMNVIDIAAug 11, 20223.1×10¹⁴ FLOP/s
41NVIDIA A800 PCIe 40 GBNVIDIANov 8, 20223.1×10¹⁴ FLOP/s
42Google TPU v4GoogleMay 20, 20212.8×10¹⁴ FLOP/s
43Google TPU v5eGoogleAug 29, 20232×10¹⁴ FLOP/s
44AWS Inferentia2Amazon AWSApr 13, 20231.9×10¹⁴ FLOP/s
45Amazon Trainium1Amazon AWSOct 15, 20221.9×10¹⁴ FLOP/s
46NVIDIA L40NVIDIAOct 13, 20221.8×10¹⁴ FLOP/s
47Meta MTIA v2 (MTIA 200)MetaApr 10, 20241.8×10¹⁴ FLOP/s
48NVIDIA GeForce RTX 3090 TiNVIDIAMar 29, 20221.6×10¹⁴ FLOP/s
49NVIDIA A40 PCIeNVIDIAOct 5, 20201.5×10¹⁴ FLOP/s
50NVIDIA HGX H20NVIDIANov 9, 20231.5×10¹⁴ FLOP/s
51Google TPU v4iGoogleJan 1, 20201.4×10¹⁴ FLOP/s
52NVIDIA Tesla V100S PCIe 32 GBNVIDIANov 26, 20191.3×10¹⁴ FLOP/s
53NVIDIA Tesla V100 DGXS 16 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s
54NVIDIA Tesla V100 DGXS 32 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s
55NVIDIA Tesla V100 SXM3 32 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s
56NVIDIA Tesla V100 SXM2 32 GBNVIDIAMar 27, 20181.3×10¹⁴ FLOP/s
57NVIDIA Tesla V100 SXM2NVIDIAJul 1, 20171.3×10¹⁴ FLOP/s
58NVIDIA Tesla V100 SXM2 16 GBNVIDIAJun 21, 20171.3×10¹⁴ FLOP/s
59NVIDIA V100NVIDIAJun 21, 20171.3×10¹⁴ FLOP/s
60Google TPU v3GoogleMay 18, 20181.2×10¹⁴ FLOP/s
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.
Browse all 硬件 →
Other rankings of 硬件
能效最高的 AI 芯片