NVIDIA H20 96GB HBM3 data center GPU built on Hopper architecture for the China market. 4.0 TB/s memory bandwidth with 296 TFLOPS FP8/INT8 and 148 TFLOPS BF16/FP16 tensor throughput. Optimized for LLM training and inference at scale.
H20 96GB
Hopper
96 GB HBM3
4.0 TB/s
296 TFLOPS
296 TOPS
148 TFLOPS
74 TFLOPS
PCIe Gen5 x16
400W
Dual-slot FHFL
900 GB/s (optional)
NVIDIA H20 is the flagship China-compliant data center GPU, purpose-built to meet export regulations while maximizing AI throughput. With 96GB HBM3 and 4.0 TB/s bandwidth, it powers LLM training, inference serving, and deep-learning recommendation across Chinese cloud and enterprise clusters — the highest-volume Hopper deployment globally.
Need NVIDIA H20?
Enterprise hardware — configured, tested, deployed. Contact us for pricing and availability.
Request Quote