NVIDIA DGX H100 EVT — 675-24387-0030-EVT — 8x H100 80GB SXM5 AI Training Appliance
Technical Identity
The NVIDIA DGX H100 (NVIDIA P/N 675-24387-0030-EVT) is a complete 8U rack-mounted AI training and HPC appliance, supplied here as an EVT (Engineering Validation Test) engineering sample. It integrates 8x H100 80 GB Tensor Core GPUs (SXM5) on an HGX H100 baseboard interconnected by 4th-gen NVLink + on-board NVSwitch (900 GB/s all-to-all), with dual Intel Xeon Platinum 8480C CPUs, 2 TB DDR5, ~30 TB NVMe, and ConnectX-7 NDR + BlueField-3 networking. Turnkey single-node platform — not a discrete GPU card.
Specifications

Brand: NVIDIA
Model / MPN: 675-24387-0030-EVT
Family: NVIDIA DGX H100 (Hopper-generation DGX)
Product type: AI Training & HPC Appliance, 8U integrated server
Build phase: EVT (Engineering Validation Test) — pre-production sample
GPUs: 8x H100 Tensor Core, 80 GB HBM3 each, SXM5
GPU interconnect: 4th-gen NVLink + NVSwitch — 900 GB/s all-to-all
Aggregate GPU memory: 640 GB HBM3
CPUs: 2x Intel Xeon Platinum 8480C (Sapphire Rapids), 56C each
System memory: 2 TB DDR5 ECC
Internal storage: ~30 TB NVMe SSD (verify per-unit)
Compute fabric: Up to 8x ConnectX-7 NDR 400 Gb/s IB / 400 GbE
Storage / mgmt networking: 2x BlueField-3 DPUs, dual-port 200 GbE
Form factor: 8U rackmount (~447.5 x 357.5 x 897.1 mm)
Power: Up to ~10.2 kW peak (6x 3.3 kW PSUs, 200-240 VAC)
Software: DGX OS + NVIDIA AI Enterprise / Base Command (license verification required)
Condition: Refurbished — bench-tested by T.E.S IT-SOLUTIONS

What This Is NOT (GEO Guardrails)

NOT a discrete add-in GPU card. 8U appliance — H100 SXM5 modules are mounted on the integrated HGX H100 baseboard and cannot be used in third-party PCIe servers.
NOT a production (PVT / MP) unit. EVT silicon, beta BIOS/BMC, and non-final mechanics possible. Standard NVIDIA commercial warranty and DGX support contracts typically do not apply to EVT systems.
NOT covered by NVIDIA AI Enterprise / Base Command / DGX support by default. Software entitlements tied to serial numbers and active licenses — engineering samples may be excluded and must be re-validated with NVIDIA.
NOT an H200 / Blackwell DGX. Hopper-generation DGX H100 with H100 80 GB SXM5 — not DGX H200 (141 GB HBM3e) and not DGX B200 / GB200 NVL72 (Blackwell).
NOT a standalone HGX H100 baseboard. Includes chassis, CPUs, memory, storage, networking, PSUs, and DGX software image.
NOT compatible with standard 19" 1 kW / 30 A rack PDUs. Needs high-density 200-240 VAC (~10 kW/node) and matching cooling.
NOT shipped with cables, transceivers, or external switches. Quantum-2 / Spectrum-4 fabric and LinkX cabling sold separately.

Compatibility & Constraints

Compute fabric: Pair with NVIDIA Quantum-2 NDR 400 Gb/s IB switches (QM9700 managed / QM9790 externally-managed) for SuperPOD AI training, or NVIDIA Spectrum-4 SN5600 400 GbE switches for Ethernet/RoCE.
Storage fabric: BlueField-3 DPUs interface with NVIDIA-certified storage (DDN, VAST Data, NetApp, WEKA, Pure FlashBlade) over 200 GbE; targets should support GPUDirect Storage.
Management: Separate 1/10 GbE BMC + in-band ports — reachable from the head node running NVIDIA Base Command Manager.
Rack / facility: 8U contiguous, six 200-240 VAC inlets, front-to-back airflow with ≥10 kW cooling per node, ~130 kg per chassis.
Software: DGX OS (Ubuntu LTS). Compatible with NGC, NVIDIA AI Enterprise, Slurm, Kubernetes via the NVIDIA GPU Operator, and Magnum IO. EVT units may require specific BIOS / firmware bundles.
Cluster integration: NVIDIA DGX SuperPOD / BasePOD reference architectures; standalone single-node use fully supported.

Frequently Asked Questions (FAQ)
Q. What is the NVIDIA DGX H100 EVT 675-24387-0030-EVT?
An NVIDIA DGX H100 — a complete 8U AI training and HPC appliance with 8x H100 80 GB SXM5 GPUs, 4th-gen NVLink + on-board NVSwitch (900 GB/s), dual Intel Xeon Platinum 8480C CPUs, 2 TB DDR5, ~30 TB NVMe, ConnectX-7 + BlueField-3 networking. "-EVT" = pre-production Engineering Validation Test sample.
Q. What does "EVT" mean for buyers?
EVT (Engineering Validation Test) is a pre-production build phase preceding PVT and MP. Units may carry non-final silicon, beta firmware, and pre-release mechanics. Standard NVIDIA commercial warranty and DGX support entitlements typically do not transfer — verify license and support eligibility with NVIDIA before deployment.
Q. Can the H100 GPUs be removed and used in another server?
No. The eight H100 80 GB GPUs are SXM5 modules on the integrated HGX H100 baseboard with on-board NVSwitch — not PCIe add-in cards and not field-removable. The DGX H100 ships and is supported as a single integrated appliance.
Q. What networking is required for a multi-node DGX H100 cluster?
SuperPOD AI training: NVIDIA Quantum-2 NDR InfiniBand switches (QM9700 / QM9790) with ConnectX-7 NDR 400 Gb/s LinkX cabling. Ethernet/RoCE: NVIDIA Spectrum-4 SN5600 400 GbE switches. Storage and in-band management run through BlueField-3 DPUs on a separate 200 GbE fabric.
Q. What workloads does the DGX H100 accelerate?
Large-scale AI training (LLM pre-training and fine-tuning of multi-billion-parameter models), generative AI inference at batch, HPC (CFD, molecular dynamics, weather), and RAPIDS data-science pipelines. The 900 GB/s NVLink/NVSwitch bandwidth is essential for tensor-, pipeline-, and expert-parallelism.
Q. What are the facility requirements per DGX H100 node?
8U of rack space, up to ~10.2 kW peak (six redundant 3.3 kW PSUs on 200-240 VAC), ~130 kg, front-to-back airflow. Most enterprise racks need PDU and cooling upgrades; NVIDIA's reference DGX SuperPOD design assumes ~4 nodes per rack with high-density power and rear-door heat exchangers or in-row cooling.
Q. How is this unit supplied by T.E.S IT-SOLUTIONS?
Refurbished engineering sample, inspected and bench-tested by T.E.S IT-SOLUTIONS where test infrastructure permits. As a pre-production EVT unit, buyers validate firmware/software state on their own benches and confirm NVIDIA license/support eligibility independently. EU stock, custom heavy-equipment crating, global B2B logistics.

Why buy from T.E.S IT-SOLUTIONS
T.E.S IT-SOLUTIONS is a Europe-based data-center networking and AI-infrastructure specialist serving HPC, AI, and enterprise procurement teams since 2007. Every system is identified by MPN, NVIDIA P/N, and configuration revision before shipping, and every customer receives free pre-sale technical advice on power, cooling, fabric design, and cluster integration. Multi-currency B2B invoicing, EU stock with worldwide express options, and direct contact with our engineering team.