• Specs from the manufacturer, nothing guessed
  • Every machine matched to the models it can run
  • Prices in pounds, charged in euros, duties and taxes included

Start with the model

What do you want to run?

Pick a model size and see every machine and card with the memory to hold it, cheapest first. Or browse the whole range: mini PCs, local-AI desktops, Apple desktops, graphics cards and AI accelerators, with honest specs.

NVIDIA GeForce RTX 5090
GeForce RTX 509032 GB · runs 30–34B models at 4-bit

Graphics cards

NVIDIAGeForce RTX 5090

32 GB of graphics memory, NVIDIA Blackwell (GB202).
Runs models up to 30–34B at 4-bit.

  • CUDA cores 21,760
  • Memory interface 512-bit, 1,792 GB/s
  • Graphics card power 575 W
Minisforum MS-S1 Max
Ryzen AI Max+ 395128 GB unified memory

Mini PCs

MinisforumMS-S1 Max

128 GB unified memory, AMD Ryzen AI Max+ 395.
Runs models up to 120B+ at 4-bit.

  • Networking 2 × 10 Gigabit Ethernet
  • USB4 2 × USB4 v2, 80 Gb/s
  • Power supply 320 W internal
NVIDIA DGX Spark
DGX Spark128 GB unified memory

AI accelerators

NVIDIADGX Spark

128 GB unified memory, NVIDIA GB10 Grace Blackwell.
Runs models up to 120B+ at 4-bit.

  • AI performance Up to 1 petaFLOP FP4 (with sparsity)
  • Size 150 × 150 × 50.5 mm
  • Power supply 240 W
GIGABYTE GeForce RTX 5070 Ti WINDFORCE OC SFF 16G
GeForce RTX 5070 Ti16 GB · runs 13–14B models at 4-bit

Graphics cards

GIGABYTEGeForce RTX 5070 Ti WINDFORCE OC SFF 16G

16 GB of graphics memory, NVIDIA GeForce RTX 5070 Ti (Blackwell).
Runs models up to 13–14B at 4-bit.

  • Memory 16 GB GDDR7
  • Card size 304 × 126 × 50 mm
  • Recommended PSU 750 W
Apple Mac Studio (M3 Ultra)
Mac Studio M3 Ultra96 GB unified memory

Apple

AppleMac Studio (M3 Ultra)

96 GB unified memory, Apple M3 Ultra.
Runs models up to 70B at 4-bit.

  • GPU 60-core; 80-core option
  • Neural Engine 32-core
  • Storage 1 TB SSD (up to 16 TB options)

Browse by category

Local-AI desktops

Specified in full

Four machines with large unified memory, side by side. Every line is the manufacturer's figure.

  • Beelink GTR9 Pro

    Beelink

    Beelink GTR9 Pro

    Processor
    AMD Ryzen AI Max+ 395
    Memory
    128 GB LPDDR5X-8000 unified memory (soldered)
    Networking
    2 × 10 Gigabit Ethernet
    Model fit
    Up to 120B+ at 4-bit
    £3,639Shop
  • Minisforum MS-S1 Max

    Minisforum

    Minisforum MS-S1 Max

    Best fit for local AI

    Processor
    AMD Ryzen AI Max+ 395
    Memory
    128 GB LPDDR5X-8000 unified memory (soldered)
    Expansion
    PCIe x16 slot (x4 electrical), low-profile cards
    Model fit
    Up to 120B+ at 4-bit
    £2,059Shop
  • NVIDIA DGX Spark

    NVIDIA

    NVIDIA DGX Spark

    Superchip
    NVIDIA GB10 Grace Blackwell
    Memory
    128 GB LPDDR5x coherent unified memory, 273 GB/s
    Networking
    NVIDIA ConnectX-7 (200 Gb/s) and 10 Gigabit Ethernet
    Model fit
    Up to 120B+ at 4-bit
    £3,929Shop
  • Apple Mac Studio (M3 Ultra)

    Apple

    Apple Mac Studio (M3 Ultra)

    Chip
    Apple M3 Ultra
    Memory
    96 GB unified memory; 256 GB option; 819 GB/s
    Storage
    1 TB SSD (up to 16 TB options)
    Model fit
    Up to 70B at 4-bit
    £3,339Shop

Best fit for local AI: holds the largest models at 4-bit for the lowest price in this row.

PowerColor AMD Radeon AI PRO R9700 32GB

Graphics cards

PowerColor AMD RadeonAI PRO R9700 32GB

PowerColor's Radeon AI PRO R9700 is AMD's RDNA 4 workstation card, with 32 GB of memory and a blower cooler, aimed squarely at local AI. It offers the same memory as an RTX 5090 at a much lower power draw, and its blower design suits several cards in one machine.

  • Memory32 GB GDDR6, 20 Gbps
  • Board power300 W

See the hardware at work

Official footage straight from the manufacturer's own channel. Every spec on our pages comes from the same source.

NVIDIA DGX Spark | A Grace Blackwell AI Supercomputer on your deskOfficial video · NVIDIA

Mini PCs and home servers

Small machines that stay on

Built for

Who it is for

Before you buy

Guides

All guides

Which machine runs which model?

Pick a model size and see the hardware that can hold it. Memory decides what fits; speed comes second.

Open the model guide

Need something we don't list?

Tell us what you want to run, your budget and where you are. We will reply with a configuration and a quote.

Request a custom build

Shop by brand

Why Atomrig