NVIDIA H100 NVL
GPU architecture NVIDIA Hopper
GPU accelerator optimised for AI, LLM and data centre computing
14592 NVIDIA CUDA cores for large-scale parallel computing
456 NVIDIA Tensor cores to train and infer AI models
94 GB of HBM2e memory with ECC to support very large artificial intelligence models
Memory bandwidth up to 3,9 TB/s
Interface PCIe 5.0 x16 for maximum data throughput
Passive cooling designed for servers and GPU clusters
Maximum power consumption: 400 W
Free shipping from €300
Promocja cenowa na model HDR-15-5
Product intended for professional use only
NVIDIA H100 NVL
Description
NVIDIA H100 NVL - GPU accelerator for large AI models
NVIDIA H100 NVL is an advanced GPU accelerator designed specifically to support the largest artificial intelligence models and AI generative infrastructure. Based on the NVIDIA Hopper architecture, the chip provides exceptional performance for training and inference of AI models used in modern data centres.
The H100 NVL GPU is designed to work with massive language models and artificial intelligence systems used in enterprises. Thanks to its very large memory capacity and extreme data throughput, it enables efficient processing of models with billions of parameters.
NVIDIA Hopper Architecture and Tensor Cores
The H100 NVLGPU uses the NVIDIA Hopper architecture, which is designed to accelerate AI computing and analyze massive data sets. The accelerator features 14592 CUDA cores and 456 Tensor cores to accelerate matrix operations used in deep learning.
The card also offers 94GB of HBM2e memory with ECC error correction and a bandwidth of up to 3.9 TB/s to support the largest artificial intelligence models on the market today.
CUDA cores
Tensor cores
of HBM2e memory
memory bandwidth
GPU for generative artificial intelligence
The NVIDIA H100 NVL is used in the computing infrastructure for the most advanced AI systems. The accelerator enables both the training of huge models and their deployment in production environments.
LLM training and inference
models combining text, image and video
AI with enterprise knowledge base
AI-as-a-Service platforms
big data and enterprise AI
academic computing and simulation
GPU infrastructure for data centers
The NVIDIA H100 NVL is designed to run in modern server systems and GPU clusters supporting the most demanding computing workloads. The PCIe 5.0 x16 interface provides very high bandwidth communication with the server system.
With passive cooling and a maximum power consumption of 400 W, the card is designed for installation in professional GPU platforms used in data centres and AI infrastructures.
GPUs for LLM and AI enterprise infrastructures
The increasing complexity of artificial intelligence models is driving the need for more and more processing power and memory bandwidth. The NVIDIA H100 NVL is designed with these requirements in mind and is one of the key solutions used to build AI infrastructure in enterprises and data centres.
Technical Specification
| H100 SXM | H100 NVL | |
|---|---|---|
| FP64 | 34 teraFLOPS | 30 teraFLOPS |
| FP64 Tensor Core | 67 teraFLOPS | 60 teraFLOPS |
| FP32 | 67 teraFLOPS | 60 teraFLOPS |
| TF32 Tensor Core* | 989 teraFLOPS | 835 teraFLOPS |
| BFLOAT16 Tensor Core* | 1,979 teraFLOPS | 1,671 teraFLOPS |
| FP16 Tensor Core* | 1,979 teraFLOPS | 1,671 teraFLOPS |
| FP8 Tensor Core* | 3,958 teraFLOPS | 3,341 teraFLOPS |
| INT8 Tensor Core* | 3,958 TOPS | 3,341 TOPS |
| GPU Memory | 80 GB | 94 GB |
| GPU Memory Bandwidth | 3.35 TB/s | 3.9 TB/s |
| Decoders | 7 NVDEC, 7 JPEG | 7 NVDEC, 7 JPEG |
| Max Thermal Design Power (TDP) | Up to 700 W (configurable) | 350-400 W (configurable) |
| Multi-Instance GPUs | Up to 7 MIGs @ 10GB each | Up to 7 MIGs @ 12GB each |
| Form Factor | SXM | PCIe dual-slot air-cooled |
| Interconnect | NVIDIA NVLink™: 900 GB/s, PCIe Gen5: 128 GB/s | NVIDIA NVLink: 600 GB/s, PCIe Gen5: 128 GB/s |
| Server Options | NVIDIA HGX H100, Partner and NVIDIA-Certified Systems™ with 4 or 8 GPUs | NVIDIA DGX H100 with 8 GPUs, Partner and NVIDIA-Certified Systems with 1–8 GPUs |
| NVIDIA Enterprise Add-on | Included | Included |
| Note | *With sparsity | |
