Back to GPUs
H100 SGX
Data Center
NVIDIA - Hopper Architecture
HPC
AI
data-center
enterprise
LLM
NVIDIA's data center GPU designed for high-performance computing and AI workloads, featuring exceptional performance for large language models and deep learning applications.
Launch Date
March 21, 2023
Launch MSRP
$30,000
Performance
Delivers up to 3X higher performance for AI training and up to 30X higher performance for AI inference compared to the previous generation.
Technical Specifications
CUDA Cores:
14,592Tensor Cores:
456Ray Tracing Cores:
N/ABase Clock:
1.095 GHzBoost Clock:
1.755 GHzMemory:
80 GB HBM2eMemory Bus:
5120-bitMemory Bandwidth:
2,040 GB/sTransistors:
80 billionDie Size:
814 mm²TDP:
350 WKey Features
- HBM2e memory
- PCIe 5.0 interface
- NVLink support
- 4th generation Tensor Cores
- Transformer Engine with FP8 precision