128GB+ desktops for local AI: Mac Studio and mini PCs
Compare high-memory desktop configurations by usable memory, software support and exact UK purchase options.
By Laptop Deals UK · Updated
A desktop gives you different choices
If you do not need a battery or built-in screen, a high-memory desktop can offer a more convenient permanent workspace. Start with the runtime you need: Apple Silicon, AMD integrated graphics and NVIDIA GB10 support different software paths. A 128GB label does not make them interchangeable.
This is a small specification-led shortlist, not a whole-market ranking or a set of hands-on tests. Budget for a display, keyboard and mouse, and choose storage for your model files as well as ordinary work.
Exact configurations to consider
Compact Windows workstation with a large shared-memory pool
HP Z2 Mini G1a — 128GB
128GB unified/shared memory · 2TB SSD · Ryzen AI Max+ PRO 395 · Windows 11 Pro
A desk-based alternative to a high-memory laptop, with 128GB onboard memory and Radeon 8060S integrated graphics. Compare it for a Windows workstation role rather than assuming it replaces a CUDA GPU.
Before you buy: Check the exact configuration, current purchase terms and delivery date with the retailer before ordering.
128GB LPDDR5x shared system memory, not dedicated VRAM. Check HP firmware graphics-allocation limits for this exact machine and leave room for Windows, context and runtime buffers.
Software
Use a runtime with a supported AMD or Vulkan path; confirm the current Windows driver and model operators. CUDA-only software does not run unchanged on Radeon graphics.
Performance
Capacity is not a speed measurement. We have not benchmarked this configuration; compare sustained performance using your model, context and runtime.
Configuration: 8T219EA#ABU. Specifications reviewed 14 October 2026.
macOS users who want a permanent high-memory MLX workspace
Mac Studio M5 Max — 128GB configuration
128GB unified/shared memory · 2TB SSD · Apple M5 Max · macOS
Consider the 18-core CPU and 40-core GPU option with 128GB unified memory for an Apple Silicon workflow. Choose that exact memory and storage combination in the configurator; the starting model is a different machine.
Before you buy: Select the configuration listed here at Apple and confirm that it can be ordered, its final price and delivery date. Memory and internal storage are chosen at purchase; budget for a display and peripherals.
128GB unified memory is shared by the operating system, CPU and GPU. Leave workload headroom and check the runtime memory limit; it is not 128GB wholly available for model weights.
Software
MLX targets Apple Silicon; llama.cpp also supports Metal. Confirm your model format and required operators. NVIDIA CUDA packages need an alternative implementation.
Performance
This recommendation is based on configuration and software fit, without a hands-on speed claim. Compare prompt processing and generation separately for your workload.
Configuration: M5 Max 18-core CPU / 40-core GPU / 128GB unified memory / 2TB SSD. Specifications reviewed 9 October 2026.
Allow for the operating system, model weights, context/KV-cache and runtime buffers. AMD's firmware graphics allocation and Apple's GPU memory limits need checking for the exact configuration; neither machine gives every installed gigabyte to your model.
If your project depends on NVIDIA CUDA and Linux on Arm, compare DGX Spark and GB10 systems. If you need portability, compare 128GB+ laptops. For a conventional desktop with a discrete GPU, check dedicated VRAM separately from system RAM.
