This is a classic example of an administrator or IT support provider specifying hardware for designers based on workstation-class branding rather than the way SketchUp actually uses the hardware.
I have seen this several times: an older Xeon-based workstation is presented as a powerful solution because it has a high core count and was originally very expensive. Unfortunately, this Broadwell-era Xeon prioritises additional cores over modern single-core performance. SketchUp’s modelling environment still benefits heavily from strong single-core CPU speed, so a newer mainstream processor will often feel considerably faster than an older workstation CPU with many more cores.
This type of machine may still perform adequately in certain professional CAD and engineering workflows, but that does not automatically make it a good SketchUp workstation. SketchUp works primarily with polygonal geometry, while its viewport performance and visualisation capabilities increasingly depend on a modern GPU with current driver support.
The same applies even more strongly once rendering and AI workflows are introduced. Most modern render engines increasingly favour NVIDIA’s RTX platform, with sufficient VRAM needed to hold textures, geometry, lighting data and render passes. An 8GB Quadro RTX 4000 may still be usable, but it no longer compares favourably with newer 12GB, 16GB, 24GB or 32GB RTX cards.
Given that you are based in Amsterdam, I would not expect this system to achieve much more than approximately €900 on the private resale market. I would therefore be very careful about purchasing it at anything close to its original workstation price. As several people have already suggested in this thread, it would be better to revise the specification around hardware that is optimised for the next five years of working in SketchUp rather than buying into an ageing workstation platform.
If local AI is also part of the requirement, storage needs to be considered from the beginning. I would not start with less than a 1TB SSD, and realistically I would recommend at least 2TB. A collection of local models, checkpoints, LoRAs and supporting files for ComfyUI and LM Studio can quickly consume hundreds of gigabytes while you test different models and workflows.
You also need to consider the balance between system RAM and GPU memory. This becomes especially important in ComfyUI when working with multiple reference images, start and end frames, animation workflows, style transfers or combinations of several models. The system needs to hold a considerable amount of visual context in memory while processing the result.
For this type of work, VRAM often matters more than raw gaming performance. An RTX 5080 with 16GB may sound extremely capable, but the 16GB memory limit could become its main weakness over time. A GPU with 24GB to 32GB of VRAM would provide a much more useful lifespan for serious local AI work.
For someone genuinely trying to future-proof a workstation for local AI over the next three to five years, I would look towards 128GB of system RAM and a GPU with approximately 32GB of VRAM. That is, however, a significant investment. For many independent consultants and small businesses, paying approximately $20 per month for a hosted AI service remains more economical than purchasing and maintaining the equivalent hardware locally.
A Mac mini can still be an interesting alternative for LM Studio because Apple’s unified memory allows relatively large language models to run locally. That can be valuable when privacy, intellectual property and business-specific workflows are important. It allows sensitive information to remain on the local machine rather than being submitted to a general cloud-based AI service.
