NVIDIA GPU Generations

Year	Architecture	Example GPUs	CUDA Cores (range)	Memory (type / size)	Memory Speed / BW	Supported Data Types
2024	Blackwell	B100, B200, GB200, RTX 50xx	~20k–30k+	HBM3e, 96–288 GB	~4–8 TB/s	FP64, FP32, TF32, FP16, BF16, FP8, FP6, FP4, INT8/4
2022	Hopper	H100, H200, GH200	~14k–18k+	HBM3, 80–94 GB	~3–3.9 TB/s	FP64, FP32, TF32, FP16, BF16, FP8, INT8
2022	Ada Lovelace	RTX 4090, RTX 4080, RTX 4070, RTX 6000 Ada	~5k–16k	GDDR6X, 12–24 GB	~21 Gbps (~1 TB/s)	FP32, FP16, BF16, TF32, INT8, limited FP8
2020	Ampere	RTX 3090, RTX 3080, RTX 3070, A100, A30	~3k–10k+	GDDR6X / HBM2e, 8–80 GB	~760 GB/s–1.6 TB/s	FP64, FP32, TF32, FP16, BF16, INT8/4
2018	Turing	RTX 2080 Ti, RTX 2080, RTX 2070, T4	~2300–4300	GDDR6, 6–11 GB	~14 Gbps	FP32, FP16, INT8, INT4
2017	Volta	V100, Titan V	5120	HBM2, 16–32 GB	~900 GB/s	FP64, FP32, FP16, INT8
2016	Pascal	GTX 1080 Ti, GTX 1080, Tesla P100, P40	~2560–3584	GDDR5X / HBM2, 8–16 GB	~10–16 Gbps	FP64, FP32, FP16 (limited)

More information