GPU Wiki

Compare GPUs

Select up to four GPU Wiki entries to compare matching specifications row by row. The comparison uses the site database and keeps generation-specific specs visibly separate.

Popular comparisons

Add a third or fourth GPU
Spec H100 CNX
Overview
Product name NVIDIA H100 CNX
Architecture Hopper
Generation / family NVIDIA H-Series
GPU codename GH100
Launch date 2022-03-22
Market segment Data Center in NVIDIA H-Series
Board type / entry type Data Center GPU
GPU
Process node TSMC 5 nm
Die size 814 mm²
Transistor count 80 billion
GPU chip GH100
SM / multiprocessor count 114
CUDA Cores 14592
Tensor Cores 456
ROPs 24
TMUs 456
Clocks
Base / Core clock 690 MHz
Boost clock 1,845 MHz
Memory clock 1,593 MHz
Memory
Memory size 80 GB
Memory type HBM2e
Memory bus width 5,120-bit
Memory bandwidth 2,040 GB/s
Bus / Interface
Bus interface PCIe 5.0 x16
AGP / PCI / PCIe distinction PCIe
Power
TDP / TBP / board power 350 W
Recommended PSU 750 W
Power connectors 8-pin EPS
Slot width Dual-slot, 267 mm length
Features
CUDA support 9.0
Performance / Benchmarks
FP32 throughput 53.84 TFLOPS
FP16 throughput 215.4 TFLOPS
Muted cells indicate unavailable, not applicable, or not directly comparable specifications.

Frame rates compare runs of the same game at the same quality preset on one test system. DLAA counts as native resolution, because it applies anti-aliasing at full render resolution rather than upscaling; DLSS and FSR upscaling are never counted as native. Where a capture named ray tracing without a readable on or off value, it is read as on.