Prototype, 2017
11 water-cooled GPUs in one chassis
A dual-CPU server platform in a single CAD-designed case, cooled by one water loop.
GRUENCY / AI workstations
AI workstations for sustained load
Model training and local LLM inference keep a machine at full power for hours or days. GRUENCY sizes GPUs, power delivery and cooling for that continuous load. Each AI workstation is tested under burn-in before delivery.
What is decided before ordering
Compute, memory, PCIe lanes, power and noise are specified together, starting from the models you run.
- GPU class and VRAM sized to the models: quantisation, context length and batch size
- PCIe topology planned slot by slot, with the link width of each GPU stated
- Workstation or server platform, with ECC memory where the work requires it
- Power delivery with headroom over measured sustained draw
- Water-cooling or high-pressure air cooling, designed to an agreed noise target
- Storage layout for datasets, checkpoints and scratch space
Testing before delivery
Each AI workstation leaves Warsaw with its own test record. It includes a sustained full-load burn-in, a thermal profile for every GPU and a noise check against the agreed target. On arrival, Sebastian commissions the machine remotely into your environment. Speed is measured as well: GPU and memory performance, and tokens per second on language models.
Previous work
NVIDIA vendor work
AI compute and visual production systems
Vendor work for NVIDIA on AI, rendering and visualisation systems. WS.COMPUTER clients are based in over 30 countries.
Since 2011
Multi-GPU builds
Machines across successive GPU generations, each specified for its workload.
Questions about AI workstations
How do you size a machine for local LLM inference?
From the models you run. Parameter count, quantisation, context length and batch size determine the VRAM and memory bandwidth needed. This is the first question in the consultation.
Air or water cooling for a training machine?
It depends on sustained power draw and the noise target. Two GPUs can often be air-cooled. Four or more GPUs under continuous load usually need a custom water loop.
Do you deliver and commission outside Poland?
Yes. Machines are delivered across the EU and worldwide in custom transport packaging. Remote commissioning into your environment is part of every delivery.
Request a consultation
Describe the models, the data and where the machine will be used.
Or email info@gruency.com · Configure at WS.COMPUTER