英伟达GPU参数速查表

NVIDIA GPU参数速查表 | 更新日期:20251129
GPU型号产品架构产品形态算力情况,挂*为稀疏值显存显存带宽GPU互联详细参数
GB300Blackwell
Ultra
多芯组合,2个B300和1个Grace CPUFP4:30/38.9P*
FP8:10/20P*
FP16:5/10P*
TF32:2.5/5P*
576G HBM3e16TB/sNvlink 2x1.8TB/s查看
B300Blackwell
Ultra
SXMFP4:15/17.5P?*
FP8:4.5/9P*
FP16:2.25/4.5P*
TF32:1.12/2.25P*
FP64:40T
288G HBM3e8TB/sNvlink 1.8TB/s查看
GB200Blackwell多芯组合,2个B200和1个Grace CPUFP4:20/40P*
FP8:10/20P*
FP16:5/10P*
TF32:2.5/5P*
FP64:90T
384G HBM3e16TB/sNvlink 2x1.8TB/s查看
B200BlackwellSXMFP4:9/18P*
FP8:4.5/9P*
FP16:2.25/4.5P*
TF32:1.12/2.25P*
FP64:40T
192G HBM3e8TB/sNvlink 1.8TB/s查看
B100BlackwellSXMFP16:1.8/3.5P*
TF32:0.9/1.8P*
FP64:30T
192G HBM3e8TB/sNvlink 1.8TB/s查看
H20标准款HopperSXMINT8和FP8:296T
FP16:148T
TF32:59.8T
96G HBM34TB/sNVlink 900GB/s查看
H20大显存HopperSXMINT8和FP8:296T
FP16:148T
TF32:59.8T
141G HBM3e4.8TB/sNVlink 900GB/s
H200HopperSXMFP8和INT8:1/2P*
FP16:1/2P*
TF32:495/989T*
FP64:67T
141G HBM3e4.8TB/sNvlink 900GB/s查看
RTX 4090AdaPCIeFP16:165T/330T*
TF32:82.6/165.2T*
源于AutoDL官网
24G GDDR6x
新增48G
1TB/s64GB/s查看
RTX 4090DAdaPCIeFP16:147/294T*
TF32:73.54/147T*
源于AutoDL官网
24G GDDR6x1TB/s64GB/s查看
RTX 5090BlackwellPCIe 5.0FP16:210T/420T?
TF32:108T
源自网络,供参考
32G GDDR71.8TB/sPCIe 128GB/s查看
RTX 5090DBlackwellPCIe 5.0FP16:150/297T*?
TF32:未找到
源自网络,供参考
32G GDDR71.8TB/sPCIe 128GB/s查看
H100HopperSXMFP8和INT8:1/2P*
FP16:1/2P*
TF32:495/989T*
FP64:67T
80G HBM33.35TB/sNvlink 900GB/s查看
H100HopperPCIeFP16:0.8/1.6P*
TF32:378/756T*
FP64:51T
80G HBM32TB/sPCIe 128GB/s查看
H800HopperSXMFP16:1/2P*
TF32:495/989T*
FP64:1T
80G HBM33.35TB/sNvlink 400GB/s查看
H800HopperPCIeFP16:0.8/1.6P*
TF32:378/756T*
FP64:0.8T
80G HBM32TB/sPCIe 128GB/s查看
A100AmpereSXMFP16:0.3/0.6P*
TF32:156/312T*
FP64:19.5T
80GHBM2e2TB/sNvlink 600GB/s查看
A100AmperePCIeFP16:0.3/0.6P*
TF32:156/312T*
FP64:19.5T
80GB HBM2e2TB/sPCIe 64GB/s查看
A800AmpereSXMFP16:0.3/0.6P*
TF32:156/312T*
FP64:19.5T
80GHBM2e2TB/sNvlink 400GB/s查看
A800AmperePCIeFP16:0.3/0.6P*
TF32:156/312T*
FP64:19.5T
80GB HBM2e2TB/sPCIe 64GB/s查看
L40SAdaPCIeFP16:366/733T*
TF32:181/366*
48G GDDR6x864GB/s64GB/s查看
L40AdaPCIeFP16:181/362T*
TF32:90.5/181*
48G GDDR6x864GB/s64GB/s查看
L20AdaPCIeFP16:119.5T
TF32:59.8T
48G GDDR6x864GB/s64GB/s查看
A40AmperePCIe

FP16:149.7/299.4*

TF32:74.8/149.6*

48G GDDR6x696GB/s

Nvlink

112GB/s

PCIe

64GB/s

查看
L4AdaPCIeFP16:121/242T*
TF32:60.5/121*
24G GDDR6x300GB/s64GB/s查看
L2AdaPCIeFP16:96.5T
TF32:48.3T
24G GDDR6x300GB/s64GB/s查看
V100VoltaPCIe

FP16:32.71T
FP32:16.35T
FP64:8.177T

16/32G HBM21.13TB/s

Nvlink

300GB/s

PCIe

32GB/s

查看

T4

TuringPCIe

FP16:65.13T
FP32:8.141T
FP64:254.4G

16G GDDR6x320GB/s32GB/s查看

请注意:以上内容根据公开数据整理,建议在决策前访问NVIDIA官方网站获取最权威和最新的数据!


参考:

anzhihe 安志合个人博客,版权所有 丨 如未注明,均为原创 丨 转载请注明转自:https://chegva.com/6562.html | ☆★★每天进步一点点,加油!★★☆ | 

您可能还感兴趣的文章!

发表评论

电子邮件地址不会被公开。 必填项已用*标注