GRUENCY  /  Workstation architecture

Workstation architecture and system design

Workstation architecture means planning PCIe lanes, power, noise and memory together, so the machine suits its workload on delivery and in the years after. GRUENCY applied this method to the 11-GPU build in 2017 and applies it to every WS.COMPUTER machine.

The reference build

Four resources to plan

Each workstation balances four limited resources. The architecture sets the trade-offs between them in writing.

  • PCIe lanes: topology planned for each slot
  • Power: sustained draw measured, power delivery with headroom
  • Noise: a maximum level agreed as part of the specification
  • Memory and storage: memory channels, VRAM and storage bandwidth matched to the workload

From workload to design

  • Workload analysis: what runs, for how long and at what precision
  • Topology plan and component list, with written reasoning for each part
  • Thermal and acoustic design, with water or air cooling
  • Chassis: CAD design, laser-cut alloy, planned for transport and servicing
  • Delivered as a WS.COMPUTER build, or as a written design for another builder
The 2017 11-GPU prototype: eleven red PCIe latches in a row under the glass top, from the original build documentation

The reference build

Prototype, 2017

11 water-cooled GPUs in one chassis

A dual-CPU server platform with a single custom water loop in a CAD-designed chassis.

OctaneRender community

Multi-GPU render configurations

Render throughput planned through PCIe topology and cooling.

Since 2011

Workstation design

Workstation designs across successive hardware generations, with the same method.

Questions about workstation architecture

Is this consulting or building?

Both. The architecture comes first and the build follows. The architecture can also be delivered on its own, as a written design for another party to build.

Can you architect a system someone else builds?

Yes. The topology, thermal design and component reasoning are delivered in writing, and the build can take place anywhere. Reviews of existing designs are covered under Technical advisory.

How does this differ from a configurator?

A configurator checks that parts are compatible. Architecture plans PCIe lanes, power, noise and memory for a specific workload and for later upgrades.

Request a consultation

Describe the workload and where the machine will be used.

Or email info@gruency.com