What is STACKIT Compute Engine GPU?
STACKIT Compute Engine GPU provides high-performance GPU-accelerated virtual machines for AI training, ML inference, and high-performance computing. Currently available are NVIDIA H100 (as an HGX variant with up to 8 GPUs for very large models, and an NVL variant with 1-4 GPUs), NVIDIA A100 PCIe (80GB, 1-4 GPUs), and NVIDIA L40S (48GB, 1-4 GPUs) for generative AI, training, inference, and rendering/3D graphics. All GPU instances run exclusively in German data centers.
Core Features
- NVIDIA H100 HGX with 80 GB HBM3 and up to 8 GPUs per instance for models over 175 billion parameters
- NVIDIA H100 NVL with 94 GB HBM3 and 1-4 GPUs for smaller models and inference
- NVIDIA A100 PCIe with 80 GB HBM2e for training, inference, and data analysis
- NVIDIA L40S with 48 GB GDDR6 for generative AI, training, inference, and rendering/3D graphics
- Flexible scaling from 1 to 8 GPUs depending on model and instance type
Typical Use Cases
AI model training: Training and fine-tuning large language models with multi-GPU setups on H100 or A100 instances.
ML inference: Production deployments of inference workloads on H100 NVL or L40S instances.
3D rendering and graphics: GPU-accelerated rendering workloads on L40S instances.
High performance computing: Scientific simulations and data analysis on A100 or H100 instances.
Benefits
- Training data stays GDPR-compliant in Germany
- Current NVIDIA GPU generations (Hopper, Ampere, Ada Lovelace)
- Flexible hourly pay-per-use billing
- Scaling from single GPU to multi-GPU clusters
Integration with innFactory
As an official STACKIT partner, innFactory supports you with GPU computing: architecture, selecting the right GPU class, migration, operations, and cost optimization.
Available Tiers & Options
NVIDIA L40S
- 48 GB GDDR6
- Generative AI, training, inference
- Rendering and 3D graphics
- Less memory than H100/A100 for very large models
NVIDIA A100 PCIe
- 80 GB HBM2e
- Training, inference, data analysis
- 1-4 GPUs per instance
- Lower throughput than H100 for very large models
NVIDIA H100 (HGX/NVL)
- 80 GB (HGX) or 94 GB (NVL) HBM3
- HGX: up to 8 GPUs per instance for models over 175B parameters
- NVL: 1-4 GPUs for smaller models and inference
- Higher cost than A100/L40S
Typical Use Cases
Technical Specifications
Frequently Asked Questions
Which GPU models are currently available?
STACKIT offers NVIDIA H100 (as an HGX variant with up to 8 GPUs, and an NVL variant with 1-4 GPUs), NVIDIA A100 PCIe (80GB, 1-4 GPUs), and NVIDIA L40S (48GB, 1-4 GPUs).
Which GPU is best suited for which use case?
H100 HGX is suited for models over 175 billion parameters, training, and HPC. H100 NVL and A100 are suited for smaller models, training, and inference. L40S is additionally suited for generative AI as well as rendering and 3D graphics.
Can I use multiple GPUs in one instance?
Yes. Depending on the model, multi-GPU instances are available with up to 8 GPUs (H100 HGX) or up to 4 GPUs (H100 NVL, A100, L40S).
What compliance applies to AI training on STACKIT?
The instances run exclusively in German data centers. Training data stays within the EU.
How is billing handled?
Billing is usage-based with hourly billing (pay-per-use). Current prices are available in the STACKIT pricing calculator.
Note: All product information on this page has been compiled with care, but is provided without guarantee and may be outdated or incomplete. Cloud services evolve rapidly — features, pricing, SLAs, and availability change frequently. Authoritative and up-to-date information can only be found on the official product page of STACKIT (official documentation). This page does not represent an offer by STACKIT.
