← Anulum Institute
HW & GPU Clusters

Designed, built, and deployed by engineers who use them

Custom GPU compute clusters for AI training, inference, rendering, and HPC. From single workstation to rack-scale. 8+ years of hands-on GPU infrastructure experience since the mining era.

8+ years GPU experience Mining → AI conversion On-site installation CH/LI

The challenge

AI workloads demand serious compute. Cloud GPU costs spiral out of control at scale. You want your own hardware — but designing a reliable, efficient GPU cluster requires experience that most IT departments do not have. Thermal management, power distribution, network fabric, storage topology, driver compatibility — getting any of these wrong means downtime and wasted budget. We have been building and running GPU systems since 2017. We know what works, what fails, and what scales.

What we deliver

Cluster Design
Requirements analysis, GPU selection (NVIDIA/AMD), topology planning, power and cooling assessment. Architecture diagrams and component specifications.
Hardware Procurement
Vendor relationships, component sourcing, Swiss import logistics. We know which vendors deliver and which to avoid.
Build & Integration
Rack assembly, cabling, GPU installation, cooling integration, power distribution. Professional build quality with burn-in testing.
Networking
10GbE, InfiniBand, RoCE. Storage fabric design (NFS, Ceph, ZFS). Low-latency interconnects for multi-node training.
Software Stack
OS installation, CUDA/ROCm setup, container runtime (Docker/Podman), orchestration (Slurm/Kubernetes), monitoring (Prometheus/Grafana).
Mining Rig Conversion
Repurpose existing mining hardware for AI inference and training. New cases, cooling upgrades, OS migration, driver optimisation.

Pricing

Workstation
CHF 3'000 design fee
Single ML workstation, 1-2 GPUs. Optimised for inference or small-scale training.
  • GPU selection advisory
  • Component specification
  • Build + OS configuration
  • Burn-in testing + benchmarks
  • HW budget: CHF 8'000 – 25'000
Enquire
Multi-Node
CHF 8'000 design + build
Rackmount servers, 4-16 GPUs. Enterprise-grade with networking and storage.
  • Architecture design
  • Procurement advisory
  • Rack integration + cabling
  • 10GbE / InfiniBand networking
  • Slurm / Kubernetes setup
  • HW budget: CHF 25'000 – 120'000
Schedule a call
Rack-Scale
CHF 20'000+ design + PM
Full rack or multi-rack GPU cluster, 16-64+ GPUs. High-availability with redundancy.
  • Full architecture + project management
  • InfiniBand / RoCE fabric
  • Distributed storage (Ceph/ZFS)
  • Redundancy + monitoring
  • On-site installation + commissioning
  • HW budget: CHF 120'000+
Contact us

How we work

Step 1
Requirements
Workload analysis, budget, space, power, and cooling assessment at your site.
Step 2
Design
Component selection, architecture diagram, and detailed quote with alternatives.
Step 3
Build
Assemble, cable, and test at our facility. Burn-in and benchmark before delivery.
Step 4
Deploy
Deliver, rack, connect, and configure on-site. Handover documentation and training.

Why ANULUM

We run our own clusters
ML350 Gen8 (128 GB RAM, SAS arrays), multi-GPU mining rigs converted to AI workloads. We use what we build.
GPU experience since 2017
From Ethereum mining rigs to AI training clusters. 8+ years of thermal management, driver debugging, and uptime engineering.
Not a reseller
We design and build. We do not just mark up a Dell quote. Custom configurations that hardware vendors cannot offer.
On-site in CH/LI
Based in Marbach SG. Delivery and installation anywhere in Switzerland and Liechtenstein. We rack it, we cable it, we test it.

Need compute? Let us design it.

Free 30-minute requirements call. We will assess your workload, space, power, and budget — and tell you exactly what hardware you need and what it will cost.

Schedule a requirements call

Frequently asked questions

NVIDIA or AMD?
For AI training: NVIDIA (CUDA ecosystem). For inference: both work well. For rendering: depends on your software. We recommend based on your specific workload, not brand loyalty.
Can you convert our old mining rigs?
Yes. We assess your existing hardware, upgrade where needed (cooling, storage, networking), install a proper OS and AI software stack, and benchmark. Often more cost-effective than buying new.
Do you handle procurement?
We provide detailed component lists with vendor recommendations and pricing. You purchase directly (transparency) or we handle procurement with a disclosed markup. Swiss import logistics included.
What about ongoing support?
Optional maintenance contracts: monitoring, OS/driver updates, hardware replacement coordination. Typically CHF 500-2,000/month depending on cluster size.
Cloud vs on-premise — which is better?
For sustained workloads (>40h/week GPU utilisation), on-premise breaks even within 6-12 months. For burst workloads, cloud may be more economical. We help you calculate the crossover point.