英伟达数据中心产品看板

Blackwell 系列产品参数

Platform	GB200	B200	B100	HGX B200	HGX B100
Configuration	2x B200 GPU, 1x Grace CPU	Blackwell GPU	Blackwell GPU	8x B200 GPU	8x B100 GPU
FP4 Tensor Dense/Sparse	20/40 petaflops	9/18 petaflops	7/14 petaflops	72/144 petaflops	56/112 petaflops
FP6/FP8 Tensor Dense/Sparse	10/20 petaflops	4.5/9 petaflops	3.5/7 petaflops	36/72 petaflops	28/56 petaflops
INT8 Tensor Dense/Sparse	10/20 petaops	4.5/9 petaops	3.5/7 petaops	36/72 petaops	28/56 petaops
FP16/BF16 Tensor Dense/Sparse	5/10 petaflops	2.25/4.5 petaflops	1.8/3.5 petaflops	18/36 petaflops	14/28 petaflops
TF32 Tensor Dense/Sparse	2.5/5 petaflops	1.12/2.25 petaflops	0.9/1.8 petaflops	9/18 petaflops	7/14 petaflops
FP64 Tensor Dense	90 teraflops	40 teraflops	30 teraflops	320 teraflops	240 teraflops
Memory	384GB (2x8x24GB)	192GB (8x24GB)	192GB (8x24GB)	1536GB (8x8x24GB)	1536GB (8x8x24GB)
Bandwidth	16 TB/s	8 TB/s	8 TB/s	64 TB/s	64 TB/s
NVLink Bandwidth	2x 1.8 TB/s	1.8 TB/s	1.8 TB/s	14.4 TB/s	14.4 TB/s
Power	Up to 2700W	1000W	700W	8000W?	5600W?

通过chatgpt 生成了tesla 主要产品线产品的参数：

Product	Release Year	Tensor Core FLOPS	CUDA Core FLOPS	Memory Capacity	Interconnect	Key Features
Tesla K80	2014	N/A	8.74 TFLOPS	24 GB GDDR5	PCIe Gen3 x16	Kepler Architecture, Dual-GPU
Tesla P100	2016	N/A	Up to 10.6 TFLOPS	12 GB - 16 GB HBM2	NVLink, PCIe Gen3 x16	Pascal Architecture, NVLink and NVLink Bridge
Tesla V100	2017	Up to 125 TFLOPS	Up to 14.1 TFLOPS	16 GB - 32 GB HBM2	NVLink, PCIe Gen3 x16	Tensor Core, NVLink and NVLink Bridge, Volta Architecture
Tesla T4	2018	65.6 TFLOPS	8.1 TFLOPS	16 GB GDDR6	PCIe Gen3 x16	Turing Tensor Core, Power Efficiency
Tesla A10	2020	624 TFLOPS	77.4 TFLOPS	N/A	PCIe Gen4 x16	Ampere Architecture, AI Inference
Tesla A30	2022	777 TFLOPS	156 TFLOPS	N/A	PCIe Gen4 x16	Ampere Architecture, AI Inference
RTX 3090	2020	285.1 TFLOPS	10,496 CUDA cores	24 GB GDDR6X	PCIe Gen4 x16	Ampere Architecture, Gaming & Professional GPU
RTX 4090	N/A	N/A	N/A	N/A	N/A	Unreleased

Reference

• https://en.wikipedia.org/wiki/Nvidia_Tesla