Skip to main content

Engineering workstations

Simulation is patient work done by an impatient person. The machine is specified to shorten the part where you are waiting.

Where the budget goes first

  1. Core count

    Solvers and analysis scale across cores; this is where wall-clock time is won.

  2. Memory capacity and bandwidth

    A run that exceeds memory does not slow down, it fails.

  3. Sustained cooling

    A multi-hour solve is thermally nothing like a benchmark burst.

  4. Fast local storage

    Result files are large and written continuously.

Thermals are a specification, not an afterthought

A machine that holds full speed for thirty seconds and a machine that holds it for six hours can carry identical component lists. The difference is whether the cooling was specified for sustained load or for a burst.

Engineering work is the case where this matters most, because the runs are long and a machine that throttles quietly turns a four-hour job into a seven-hour one without ever reporting a fault.

  • Cooling specified for continuous load, not peak
  • Case airflow planned as part of the build rather than assumed
  • Noise expectations agreed up front — sustained cooling is rarely silent

Memory that fails, rather than slows

When a solve exceeds available memory the usual outcome is not a slower solve. It is an error, hours in, with the work lost. That asymmetry is why memory on an engineering machine is specified with margin rather than to the expected requirement.

What we will not estimate

We will not tell you how long your solve will take, or how much faster it will be than your current machine. Those depend on your model, your solver and your settings, and a number invented here would be a number you planned around.

What we can do is explain which component governs which part of the run, so the money goes where your particular work is bounded.

What each choice costs elsewhere

Build choices, what they gain and what they cost
ChoiceWhat it buysWhat it costs
More coresShorter solves for parallel workloadsLower per-core speed for interactive modelling
Cooling for sustained loadFull speed for the whole runA larger case and more audible fans
Memory with marginRuns that complete rather than failCapacity you may not use every day

Questions

How many cores should an engineering workstation have?

As many as your solver can genuinely use, which is a question about your software rather than about hardware. Some scale nearly linearly; others stop improving well before the core count runs out. Check what your vendor documents before buying for parallelism.

Will this be faster than our current machine?

Almost certainly, but we will not put a multiple on it. That number depends on your model and solver, and a figure invented here is one you would plan around. We would rather explain which component governs your particular bottleneck.

Next

CAD workstations

When modelling dominates the day

Cooling and thermal planning

Why sustained load is the real test

High-performance desktops

General-purpose fast machines

Specify for your solver

Tell us the software and the size of a typical run.

Start a specification