Framework has officially set a Wednesday morning Pacific Time pre-order start for its upgraded compact desktop featuring 192GB of unified memory and AMD’s Ryzen AI Max+ Pro 495 processor. The miniature system targets enthusiasts running massive local artificial intelligence models directly from their desks.
AMD Ryzen AI Max+ Pro 495 Hardware Specifications
Powering the newly dated desktop is AMD’s upgraded chip, which brings 16 Zen 5 cores running at speeds up to 5.2 GHz according to Digitaltrends reporting. The processor pairs with a beefed-up Radeon 8065S graphics architecture featuring 40 RDNA 3.5 cores operating at 3.0 GHz, as detailed by Wccftech.
The standout improvement in this iteration is the massive 192GB capacity leap in unified memory, operating at 273 GB/s across 8533 MT/s LPDDR5X channels. This configuration delivers a 50% capacity boost and a 6.6% bandwidth speed-up compared to earlier 128GB systems built around the Ryzen AI MAX+ 395. Retaining the standard Mini-ITX form factor and FlexATX power supply, the machine keeps the exact physical dimensions and case architecture of its predecessor for complete backwards compatibility.
Local AI Performance and DeepSeek V4-Flash Capabilities
All that hardware headroom serves a very specific purpose for developers and hobbyists working with heavy language models locally. Framework states that the desktop is optimized to run DeepSeek V4-Flash at Q8 single-handedly, leaving extra memory space available to supply the model with expanded context lengths. This gives tinkerers a powerful sandbox for local model experimentation without relying on cloud infrastructure.
Motherboard Expansion and Cooling Enhancements
Beyond the processor and memory upgrades, Framework has refined several physical aspects of the desktop motherboard and chassis. The PCIe x4 slot now features an open-ended design, allowing users to fit larger expansion cards directly into the compact machine. To highlight this flexibility, Framework demonstrated a concept cluster utilizing dual 50 GbE network interface cards running RDMA over Ethernet to pool 384GB of memory across two desktops using tensor parallelism. Out-of-the-box cooling comes handled by a pre-installed Noctua fan, with options including the Noctua NF-A12x25 HS-PWM, Cooler Master Mobius 120, or 120P fans depending on the configuration.
Pricing Tiers and Pre-Built Software Options
Buyers looking to secure a unit during the upcoming launch window face steep costs driven by rising LPDDR5X memory prices. While exact retail figures for the 192GB variant were not locked down prior to pre-orders, the company warns that the top-tier configuration will carry a substantial price increase over the 128GB models. Existing base hardware powered by the 395 chip typically starts around $3,500, while competing high-end pre-orders top $5,000 US.
For software deployment, Framework offers pre-built system configurations with either Windows 11 or Fedora Linux ready to go out of the box.
