GPUSPECINDEX

NVIDIA · Hopper · 2022

H100 GPU accelerator (PCIe card)

H100 GPU accelerator (PCIe card): 80 GB HBM2E, 350 W, 14,592 CUDA cores. Launched March 22, 2022.

Memory
80 GB
Power
350 W
Bus width
5,120 bit
Memory80 GB5 of 32 in Data Center
Power350 W11 of 40 by higher draw in Data Center
CUDA cores14,5928 of 47 in Data Center

Affiliate links — we may earn a commission.

Memory

Capacity
80 GB
Memory type
HBM2E
Bus width
5,120 bit
Bandwidth
2,039 GB/s

Compute

CUDA cores
14,592
SM count
114
TMUs
456
ROPs
24
FP32
51,200 GFLOPS
FP64
25,600 GFLOPS

Clocks

Base clock
1,065 MHz

Board

TDP
350 W
Form factor
datacenter

Physical

Architecture
Hopper
Series
Data Center
Launch date
March 22, 2022

Not recorded by the source: generation, codename, process and transistors, or 8 further fields.

Silicon, made visible

Die area is drawn against the largest recorded die in the catalog. Shader dots use a labeled, card-specific ratio so both early and modern GPUs remain legible.

Die size not recorded · 14,592 shadersDie size not recorded; shader grid shown without scale. Each dot = 128 shaders.14,592 shadersEach dot = 128 shadersDie size not recorded; shader grid shown without scale

Within its series, the H100 GPU accelerator (PCIe card) ranks 7th of 36 for FP32 throughput and 5th of 32 for memory capacity. Its 350 W TDP is above the 75 W PCIe slot-power ceiling.

Source: Wikipedia, revision 1369488585 · catalog loaded 2026-08-16