Netcup vGPU

Netcup just launched a new RS line with NVIDIA H200 vGPU!

Early access requires registration here:
https://www.netcup.com/en/vgpu-earlyaccess

Here is the currently listed offers:
RS 2000 vGPU 7:
€137.31/month

• GPU: NVIDIA H200
• VRAM: 7 GB (dedicated)
• CPU: AMD EPYC 9535
• CPU Cores: 8 vCore (dedicated)
• RAM: 16 GB DDR5 RAM (ECC)
• Storage: 512 GB NVMe SSD
• Location: Nuremberg, Germany

RS 4000 vGPU 14:
€261.39/month

• GPU: NVIDIA H200
• VRAM: 14 GB (dedicated)
• CPU: AMD EPYC 9535
• CPU Cores: 12 vCore (dedicated)
• RAM: 32 GB DDR5 RAM (ECC)
• Storage: 1 TB NVMe SSD
• Location: Nuremberg, Germany

Analysis (generated by Gemini 2.5):

Hourly Cost Comparison:

Assuming ~730 hours per month:

RS 2000 vGPU 7: €137.31 / 730 ≈ €0.188 per hour

RS 4000 vGPU 14: €261.39 / 730 ≈ €0.358 per hour

These hourly rates cover the entire server (GPU fraction, CPU, RAM, SSD).

Cost per GB of H200 VRAM:

RS 2000 vGPU 7: €0.188 / 7 GB ≈ €0.0269 per GB/hour

RS 4000 vGPU 14: €0.358 / 14 GB ≈ €0.0256 per GB/hour

Market Comparison (Full H200): As established previously, renting a full 141GB H200 GPU often costs $2.50-$4.00+ per hour (€2.30-€3.68+). This translates to roughly €0.016 to €0.026 per GB/hour.

Value Interpretation:

GPU Cost: Netcup's price per GB of H200 VRAM (€0.0256 - €0.0269 /GB/hr) is at the upper end or slightly above the estimated market rate per GB when renting a full H200 card. The larger RS 4000 plan offers slightly better value per GB of VRAM than the RS 2000 plan.

Included Server Resources: Crucially, the price includes not just the GPU fraction but also powerful dedicated AMD EPYC CPU cores, substantial ECC DDR5 RAM, and fast NVMe storage. These components have significant value on their own. Renting a comparable non-GPU root server would still incur costs.

Accessibility & Fractional Use: The main appeal is accessing the cutting-edge H200 architecture and performance characteristics without needing to rent (and pay for) the full 141GB VRAM. This is ideal for workloads that benefit from the H200's capabilities but don't require its maximum memory capacity (e.g., certain AI inference tasks, video encoding, rendering smaller scenes).

Root Server Control: Unlike some cloud GPU instances, these are root servers, offering full control over the environment.