NVIDIA DGX
The NVIDIA DGX H100 represents the pinnacle of computing platforms for artificial intelligence, machine learning, and data analytics.
Built on the NVIDIA Hopper architecture, this system delivers extraordinary performance for complex AI models and simulations that would take hours or even days on conventional hardware. Thanks to its integrated optimization for deep neural networks and massive computational parallelism, it is ready to power projects in medicine, Industry 4.0, mobility, energy, and robotics.
At TUKE, the NVIDIA DGX H100 is not only a technological showcase — it is an open research platform that enables students, PhD candidates, and research teams to experiment, innovate, and publish at a world-class level. It promotes interdisciplinary collaboration and significantly shortens the time between concept and result.
Technical Specifications
| Specification | Details |
|---|---|
| GPU | 8× NVIDIA A100-SXM4-40GB |
| Performance (FP8) | 32 petaFLOPS |
| NVIDIA NVSwitch | 4× (4th generation NVLink) |
| NVLink Interconnect | 18× per GPU, 900 GB/s bidirectional GPU-to-GPU bandwidth |
| CPU | 2× AMD EPYC 7742 – 64 cores – 2.25 GHz |
| System Memory | 1 TB |
| Network | 2× 100 Gbit |
| Network Management | 10 Gb/s onboard NIC (RJ45), host baseboard management controller (BMC) with RJ45 |
| Storage (OS) | 2× 1.92 TB NVMe M.2 |
| Internal Storage (Cache) | 5× 3.84 TB NVMe U.2 |
| Power Consumption (max) | ~10.2–11.3 kW |
| Operating System | DGX OS |
For more information or technical inquiries, please contact: