NVIDIA V100
芯片
GPU
历史
The 2017 Volta GPU that made tensor cores — and modern AI training — mainstream.
您正在查看一条芯片记录：该处理器是什么、其关键规格、谁设计它、谁制造它、谁供应其零部件，以及它被用在哪里。
Tesla V100 introduced tensor cores, dedicated matrix-math units that became the defining feature of AI silicon. It powered the first large transformer training runs, including the original GPT models at OpenAI, and established the SXM module + NVLink pattern every later NVIDIA data-center GPU follows.
规格与关键参数
每个数据均标注来源依据：厂商规格、厂商声明、公开报道、估算或官方数据。"未公开披露"即字面含义——我们从不捏造数值。
Overview
Designer
NVIDIA
Architecture
Volta
Announced
May 2017 (GTC)
Status
Historical — superseded
Manufacturing
Foundry / process
TSMC 12 nm FFNVendor spec
Transistors
21.1 billionVendor spec
Die size
815 mm²Vendor spec
内存
Memory
16 or 32 GB HBM2Vendor spec
Memory bandwidth
900 GB/sVendor spec
算力
FP32 (vector)
15.7 TFLOPSVendor spec
FP64 (vector)
7.8 TFLOPSVendor spec
FP16 tensor
125 TFLOPSVendor spec
first-generation tensor cores
电力
TDP (SXM2)
300 WVendor spec
Physical
Form factors
SXM2, PCIe
Interconnect
NVLink 2 — 300 GB/s; PCIe 3.0Vendor spec
它如何与整个技术栈相连
本记录的关联链接一览——点击分组可直接跳转。
设计方
由……制造
内存技术
用于训练
后继产品
来源
记录审阅截至 2026-01-15 — 查看 方法论.