Skip to content

Chapter 0: How We Got Here -- The Hardware Story · Part 10 of 15

Act 8 -- The server that emerged: DGX H100 anatomy

Now that you understand why the hardware evolved this way, you can look at the DGX H100 and understand every design decision:

DGX H100 — anatomy and alternatives

Official NVIDIA DGX H100 system image.

Figure: Official NVIDIA DGX H100 system image. Seeing the chassis itself helps ground the anatomy discussion in the physical box operators actually deploy.

Every choice in the DGX design is an answer to a specific constraint:

Design decisionWhy
8 GPUs per chassisMaximum NVLink bandwidth per GPU within practical power and thermal limits
4 NVSwitch chipsNon-blocking all-to-all between 8 GPUs at 900 GB/s per GPU
1 ConnectX-7 per GPUNo shared uplinks -- each GPU gets its own 400G path to the fabric
8 NICs on 8 different switchesRail topology -- fault isolation and maximum AllReduce performance
2 Intel Xeon CPUsManagement only -- training is GPU + NIC, CPUs are mostly idle
8x 3.84 TB NVMe U.2 (~30 TB)Local OS, checkpointing -- training data lives on shared parallel storage (BeeGFS, WEKA)