Specification

Architecture: Hopper
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 6 h agomedium
Compute bf16 tflops: 1,671 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
FP16 compute: 1,671 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute fp32 tflops: 60 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute fp64 tensor tflops: 60 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute fp64 tflops: 30 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute fp8 tflops: 3,341 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute int8 tops: 3,341
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute sparsity: Yes
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Compute tf32 tflops: 835 TFLOPS
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Form factor: PCIe dual-slot air-cooled
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Kind: gpu
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 6 h agomedium
Manufacturer: NVIDIA
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 6 h agomedium
Memory bandwidth: 3,900 GB/s
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Memory: 94 GB
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Mig support: Up to 7 MIGS @ 12GB each
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Nvlink bandwidth gbs: 600 GB/s
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
Runtimes: cuda, tensorrt-llm, vllm, sglang, llama.cpp
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 6 h agomedium
Spec sheet: nvidia.com/en-us/data-center/h100
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 6 h agomedium
Tdp note: 350-400W (configurable)
Source:NVIDIA — data center product pagesT1observed 4 h agohigh
TDP: 400 W
Source:NVIDIA — data center product pagesT1observed 4 h agohigh

Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →

Provenance

Attributed facts

21

Source tiers

T1T216 / 5

Freshest observation

4 h ago

Conflicts

None

Estimated

Models whose estimated memory footprint fits this device — see estimate method.

No models recorded

Manufactured by: NVIDIA

As of

Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.

Claim history

5 claims · 5 properties

Architecturearchitecture1

Claim history for Architecture
Value	Valid from → to	Status	Source	Confidence	Extractor
Hopper	11 Sept 2026 20:50 UTC → current	current	AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2	medium	curated

Kindkind1

Claim history for Kind
Value	Valid from → to	Status	Source	Confidence	Extractor
gpu	11 Sept 2026 20:50 UTC → current	current	AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2	medium	curated

Manufacturermanufacturer1

Claim history for Manufacturer
Value	Valid from → to	Status	Source	Confidence	Extractor
NVIDIA	11 Sept 2026 20:50 UTC → current	current	AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2	medium	curated

Runtimesruntimes1

Claim history for Runtimes
Value	Valid from → to	Status	Source	Confidence	Extractor
cuda, tensorrt-llm, vllm, sglang, llama.cpp	11 Sept 2026 20:50 UTC → current	current	AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2	medium	curated

Spec sheetspec_url1

Claim history for Spec sheet
Value	Valid from → to	Status	Source	Confidence	Extractor
https://www.nvidia.com/en-us/data-center/h100/	11 Sept 2026 20:50 UTC → current	current	AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2	medium	curated

Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →

11 Sept 2026
New hardwareHardwareNVIDIA H100 NVLNVIDIA
New hardware: NVIDIA H100 NVL (NVIDIA)
registry
nvidia.com

Full timeline →

Source documents
Source	Document	Type	Tier	Last observed	Snapshots
NVIDIA — data center product pages	nvidia.com/en-us/data-center/h100	spec_page	T1· Official	4 h ago	1

Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.