NVIDIA DGX
NVIDIA DGX
Also known as: DGX A100, DGX H100, DGX H200, DGX B200, DGX SuperPOD
DGX is NVIDIA's purpose-built training appliance: 8 Hopper or Blackwell GPUs in one chassis with NVLink fabric, fast InfiniBand, and an integrated AI software stack. A single DGX H200 delivers ~32 PFLOPS FP8. FI Tech uses partner DGX clusters in Riyadh and through the Saudi SDAIA / NEOM compute pool…
Definition
DGX is NVIDIA's purpose-built training appliance: 8 Hopper or Blackwell GPUs in one chassis with NVLink fabric, fast InfiniBand, and an integrated AI software stack. A single DGX H200 delivers ~32 PFLOPS FP8. FI Tech uses partner DGX clusters in Riyadh and through the Saudi SDAIA / NEOM compute pool for training large vision foundation models on local data — fine-tuning Grounding DINO, RF-DETR, or DINOv3 on Saudi-construction, Aramco-yard, and SABIC-plant footage that cannot leave the kingdom under PDPL. Inference workloads stay on Jetson and Hailo-8 at the edge.
