Dedicated NVIDIA L40S, RTX 5000 PRO, RTX 6000 PRO and H200 GPUs — single-tenant or cloud-mode, with up to 1.1 TB VRAM aggregate. Train large language models, run inference at scale, render frames, ship faster.
L40S (Ada Lovelace), H200 (Hopper) and RTX PRO 5000/6000 — best-in-class accelerators for inference, training, rendering and simulation.
Pin a card to your tenancy, or share elastically with cloud-mode for cost-efficient burst capacity. Same GPUs, two billing models.
Dedicated GPU plans give you the entire card — no shared VRAM, no scheduling jitter. Predictable performance for production workloads.
Up to 7.68 TB of CPU RAM and 1.92 TB NVMe storage per node, with 15 TB of egress per month included.
Servers operated in our European data center with hardware-grade tenancy isolation and strict access controls.
All prices in EUR per month. One-time setup fee covers provisioning, OS install and CUDA toolkit configuration. Activated within one business day.
Sorted by monthly price. Swipe horizontally on mobile.
| GPU | GPU RAM | CPUs | RAM | Storage | Price | Taxa única de configuração | |
|---|---|---|---|---|---|---|---|
| NVIDIA L40S | 48 GB | 32 | 234 GB | 1.75 TB | $2.070/mês | $690 | |
| NVIDIA L40S | 48 GB | 32 | 234 GB | 900 GB | $2.364/mês | $788 | |
| RTX 5000 PRO | 48 GB | 32 | 234 GB | 1.75 TB | $2.790/mês | $930 | |
| RTX 5000 PRO | 48 GB | 32 | 234 GB | 900 GB | $3.090/mês | $1.030 | |
| RTX 6000 PRO | 96 GB | 32 | 234 GB | 1.92 TB | $4.200/mês | $1.400 | |
| RTX 6000 PRO | 96 GB | 32 | 234 GB | 1.92 TB NVMe | $4.500/mês | $1.500 | |
| NVIDIA H200 | 141 GB | 32 | 234 GB | 1.9 TB | $5.514/mês | $1.838 | |
| NVIDIA H200 | 141 GB | 32 | 234 GB | 1.9 TB | $6.882/mês | $2.294 | |
| NVIDIA 8x H200 | 1128 GB | 192 | 7.68 TB | 1.9 TB | $48.000/mês | $16.000 |
Multi-billion parameter language models with mixed-precision training on H200 clusters.
Serve Llama 3, Mistral, Qwen and custom fine-tunes on dedicated GPUs with predictable tail latency.
Stable Diffusion, Flux, ComfyUI pipelines with high-VRAM RTX PRO cards.
Blender Cycles, Octane, Redshift render farms on Ada Lovelace cards.
NVENC-accelerated transcoding pipelines for streaming platforms and OTT workflows.
Molecular dynamics, computational fluid dynamics, weather modeling on H200.
Dedicated GPUs reserve the entire physical card for your tenancy — no shared VRAM, no scheduler jitter, full PCIe bandwidth. Cloud-mode GPUs are virtualized for elastic billing and burst capacity at slightly lower cost. Same hardware, different tenancy model.
Yes. You get full root access. Install any CUDA toolkit, PyTorch, TensorFlow, JAX or container runtime. We pre-install a modern stack but you're free to replace it.
NVIDIA driver and CUDA upgrades are handled on request through a maintenance window. We can also leave it fully under your control if you prefer.
Yes — the 8× H200 plan ships with 8 GPUs in a single node, NVLink-connected for high-bandwidth peer-to-peer transfers. Ideal for distributed training and large-context inference.
15 TB of monthly egress is included per plan. Beyond that, overage is metered. Inbound traffic is unmetered.
Pre-orders are activated within one business day. For larger fleets or NVLink-connected clusters, we schedule a call to validate the configuration first.
Tell us your workload and our engineers will recommend the right GPU and tenancy model.