is NVIDIA’s flagship AI appliance server family built as complete systems around NVIDIA SXM GPUs with an AI-ready software stack.

Context: Core concept: DGX identity and appliance model

DGX is less customizable than , because DGX is an NVIDIA-manufactured appliance with no option to customize configurations.

Context: Cause→effect relationship: DGX customization tradeoff vs HGX

servers are NVIDIA-certified platforms that use the same GPU class as DGX but are offered in multiple configurations and built by various companies.

Context: Core concept: HGX definition and certification vs manufacturing

HGX supports 4 or 8 GPU configurations and can use either CPUs or CPUs.

Context: Meaningful detail: HGX CPU options

is a modular, superdense AI server design intended for maximum flexibility and expansion, featuring the Grace Hopper GH200 superchip.

Context: Core concept: MGX definition

MGX connects the Grace Hopper GH200 superchip components with coherent CPU/GPU memory using .

Context: Core concept: NVLink-C2C role in MGX

MGX uses the Grace Hopper , which combines GPU and CPU functionality in one module.

Context: Core concept: Grace Hopper GH200

MGX uses the Grace Hopper GH200 and NVLink-C2C with coherent CPU/GPU memory, which causes higher interconnect efficiency than (as claimed).

Context: Cause→effect relationship: MGX interconnect design vs PCIe Gen 5

EGX is based on GPUs and is fully customizable by chassis, supporting a wide range of GPU counts and system components.

Context: Core concept: EGX PCIe GPU basis

EGX is fully customizable by chassis, which causes EGX to provide the greatest configuration flexibility but with less NVIDIA and packaging.

Context: Cause→effect relationship: EGX flexibility tradeoff

DGX includes a comprehensive hardware-software-support package, which causes DGX to be positioned for the most demanding workloads with less integration burden for users.

Context: Cause→effect relationship: DGX support package vs AI workload fit

DGX H100 uses GPUs, while DGX B200 is announced to use GPUs expected in late 2024.

Context: GPU architecture transition: Hopper to Blackwell across DGX

EGX offers the greatest configuration flexibility but lacks GPUs, because EGX uses PCIe GPUs rather than the more powerful SXM approach.

Context: Meaningful distinction: SXM vs PCIe GPU form factors

MGX is compatible with NVIDIA , HPC SDK, and Omniverse.

Context: Core concept: MGX software stack compatibility