NVIDIA DGX-1 / DGX-2 Maintenance — bridge service for older DGX generations
We service older NVIDIA DGX platforms (DGX-1 with Pascal P100 or Volta V100, DGX-2 with 16x V100 and NVSwitch) as bridge service until refresh strategy. With refurbishing pools for main components (V100 GPU boards, power supplies, NVSwitch modules) and SLA up to 24×7×4. Important honesty: DGX hardware is proprietary, component availability for DGX-1/-2 is structurally weak today. For very old fleets, refresh to DGX A100/H100 is often more economical than TPM bridge — we make this transparent in risk assessment before contract conclusion.
Which DGX Legacy models we service
DGX-1 was NVIDIA's first generation of complete AI training servers (released April 2016 with Pascal P100, refresh 2017-2019 with Volta V100). DGX-2 extended the concept (released 2018) with 16 V100 GPUs and the first generation NVSwitch interconnect. Both lines are technically still usable for AI inference and smaller training workloads, but no longer powerful enough for modern LLM training or MoE models.
Why (or why not) TPM for DGX-1 / DGX-2
DGX-1/-2 maintenance is not trivial. Unlike standard server EOSL (Supermicro X9, F5 s-series, etc.), DGX hardware is proprietary — NVLink, NVSwitch, P100/V100 GPU boards with NVIDIA-specific firmware. Refurbishing pools for DGX-1/-2 are structurally thinner than for standard servers, and some critical components (NVSwitch Gen 1, early V100 boards) are scarcely available on the open market. NVIDIA offers standard support for DGX-1/-2 only limited or not at all — those who stay here often have no OEM service option. TPM is a bridge service solution with clear risk disclosures.
We recommend DGX-1/-2 TPM in two scenarios: (1) AI inference workloads that don't strictly need state-of-the-art hardware — DGX-1/-2 are technically still usable for many inference use cases, TPM extends productive lifespan economically. (2) Bridge service between DGX-1/-2 EOSL and a planned refresh phase to DGX H100/H200 or GB200 — TPM typically bridges 6-18 months until refresh hardware is available and installed. For very critical productive AI training workloads on DGX-1/-2 we recommend refresh instead of TPM bridge — modern DGX generations are 10-50x more powerful and energy efficient.
Generations timeline & TPM coverage
Per hardware generation: vendor phase (slate) and TechCare coverage window (teal) up to ~5 years post-OEM EOSL.
- DGX-1 (Pascal, 8x P100)2016–2023TPM until2028+Recommended
- DGX-1 (Volta, 8x V100 16GB)2017–2024TPM until2029+Recommended
- DGX-1 (Volta, 8x V100 32GB)2018–2024TPM until2029+Recommended
- DGX-2 (16x V100, NVSwitch Gen 1)2018–2025TPM until2030+Recommended
EOSL status of DGX Legacy generations
DGX-1 and DGX-2 are completely EOSL. NVIDIA factory warranty has expired for all configurations, standard support only limited available.
| Model family | Released | OEM support ends | TPM status |
|---|---|---|---|
| DGX-1 (Pascal, 8x P100) | 2016 | 2022–2023 | Recommended |
| DGX-1 (Volta, 8x V100 16GB) | 2017 | 2023–2024 | Recommended |
| DGX-1 (Volta, 8x V100 32GB) | 2018 | 2024 | Recommended |
| DGX-2 (16x V100, NVSwitch Gen 1) | 2018 | 2024–2025 | Recommended |
As of 2026. EOSL data based on official vendor roadmaps and subject to change. Binding case-by-case information available on request.
What we deliver
OEM components
Our warehouse and certified refurbishing sources for DGX and Mellanox.
DGX specialist engineer
German-speaking technicians with NVLink/NVSwitch training, 4-hour response time guaranteed.
Flexible SLA per system
Parts Only, 5×9 NBD or 24×7×4 — freely combinable by location and criticality.
Multi-vendor contract
One contract for DGX, Mellanox and all other vendors. AI cluster stack consolidation.
Risk assessment
Component pool status per model before contract conclusion — honest disclosure.
CUDA & AI software stay
CUDA, AI Enterprise, Base Command Manager independent of hardware maintenance.