NVIDIA H100 SXM
H100 chip, self-developed module, paired with New H3C head unit; equivalent to branded all-in-one systems.
Product Specifications
Parameter Item | Parameter Value |
GPU Architecture | NVIDIA Hopper |
Manufacturing Process | TSMC 4N |
Transistor count | 800 billion |
FP64 | 34 TFLOPS |
FP64 Tensor Core | 67 TFLOPS |
FP32 | 67 TFLOPS |
TF32 Tensor Core (with sparsity) | 989 TFLOPS |
BFLOAT16 Tensor Core (with Sparsity) | 1,979 TFLOPS |
FP16 Tensor Core (with sparsity) | 1,979 TFLOPS |
FP8 Tensor Core (with Sparsity) | 3,958 TFLOPS |
INT8 Tensor Core (with sparsity) | 3,958 TOPS |
GPU VRAM | 80 GB HBM3 |
VRAM bandwidth | 3.35 TB/s |
NVLink Interconnect | 4-generation NVLink, 900 GB/s |
PCIe | PCIe Gen5, 128 GB/s |
Tensor Core | 4th Generation |
Transformer Engine | 2th Generation |
Maximum Power Dissipation (TDP) | Up to 700W |
Multi-Instance GPU (MIG) | Up to 7 instances, each with 10GB |
Dimensions | SXM |
Video Decoder | 7 NVDEC / 7 JPEG |
Confidential Computing | Support |
Typical Applications | LLM training and inference, HPC, data analytics |
Pricing Details
Product price is negotiable and determined by order quantity, configuration requirements, and delivery terms.
Submit an inquiry using the form below. Our expert team will provide a detailed quote within 24 hours.
Product Details
Can't find the right product?
Contact our expert team for customized solutions tailored to your specific needs.
Contact Us