Core details
- Release
- 2022
- Brand
- NVIDIA
- Platform
- Desktop
- Architecture
- Hopper
NVIDIA H100 PCIe (80GB)
A flagship data-center GPU designed for large-scale AI training and inference. It features 4th-gen Tensor Cores and a Transformer Engine for unprecedented performance in LLM workloads, though it lacks display outputs...
Compare
No curated comparison for this item yet
Choose another item to check whether a ready comparison exists. If this item has not been matched yet, CompareAll will show nearby ready paths instead.
Specs
Exported fields available now
Source-backed spec set18 fields shown1 source link
ManufacturerNVIDIA
ArchitectureHopper
VRAM80
MemoryHBM3
TDP350
Process4
Use-case note
With a massive 80GB of memory, this elite professional accelerator is engineered specifically for the most demanding enterprise workloads, including large-scale AI training and high-performance computing. It stands as a premier solution for organizations requiring top-tier performance for LLM inference and complex data processing at scale.
Size and build
- Length (mm)
- 268
- Height (mm)
- 111
Performance and memory
- Process (nm)
- 4
- VRAM (GB)
- 80
- Memory type
- HBM3
- Bus width (bits)
- 5120
- TDP (W)
- 350
- FP32 compute
- 51.22
- GPI index
- 29.5
- Target Resolution En
- 1080p / 60 fps
More exported fields
- Is Estimated Score
- Yes
- Target Resolution Zh
- 1080p / 60 fps
- Gpi Method
- estimated_non_time_spy_datacenter_accelerator
- Gpi Note
- Datacenter/HPC/AI accelerator without a direct consumer 3DMark Time Spy graphics score in DB; existing GPI is treated as estimated pending a formal non-Time-Spy methodology.
This page is using the richer exported specs and source links available for this entity.
Core fields for this category are present in the current export.
Similar
Other GPUs to explore
AMD Instinct MI100 (32GB)
AMD Radeon AMD Radeon AMD Instinct MI100 (32GB)
The first data center GPU based on the CDNA architecture, featuring 120 Compute Units and Matrix Core technology to deliver high performance in HPC and AI training workloads.
- Manufacturer
- AMD
- Architecture
- CDNA 1.0
- VRAM
- 32
AMD Instinct MI210 (64GB)
AMD Radeon AMD Radeon AMD Instinct MI210 (64GB)
A high-performance data center GPU accelerator designed for HPC and AI training. It features 64GB of HBM2e memory and lacks display outputs or DirectX support, resulting in a 0 score for 3DMark gaming benchmarks.
- Manufacturer
- AMD
- Architecture
- CDNA 2
- VRAM
- 64
AMD Instinct MI25 (16GB)
AMD Radeon AMD Radeon AMD Instinct MI25 (16GB)
A server-grade accelerator designed for deep learning, high-performance computing (HPC), and virtualization. Based on the Vega 10 architecture, it features high-bandwidth HBM2 memory and is optimized for large-scale A...
- Manufacturer
- AMD
- Architecture
- Vega (GCN 5.0)
- VRAM
- 16
AMD Instinct MI250 (128GB)
AMD Radeon AMD Radeon AMD Instinct MI250 (128GB)
The AMD Instinct MI250 is a professional-grade data center accelerator based on the CDNA 2 architecture, specifically designed for exascale HPC and AI workloads. It features a massive 128GB HBM2e memory capacity with...
- Manufacturer
- AMD
- Architecture
- CDNA 2
- VRAM
- 128