Framework, the new desktop is pure power: Ryzen AI Max+ 495 and up to 192 GB of RAM, but the price?

Framework

Framework, a company known for its modular notebooks for professionals, has previewed a new configuration of the Framework Desktop during the AMD Advancing AI 2026 event. The star of the show is the Ryzen AI Max+ PRO 495, a processor that elevates the platform to a new level along with 192 GB of unified LPDDR5X memory, a capacity expressly designed for the local execution of large language models.

The new AMD chip retains 16 Zen 5 cores and 40 RDNA 3.5 Compute Units, but boosts the CPU frequency to 5.2 GHz. The graphics are powered by the Radeon 8065S, while the integrated NPU reaches 55 TOPS of dedicated AI computing power. The LPDDR5X memory operates at 8533 MT/s with a bandwidth of 273 GB/s. Compared to the previous solution with the Ryzen AI Max+ 395 and 128 GB of RAM, the new memory configuration offers a 50% increase in capacity and a 6.6% improvement in memory bandwidth.

Today at AMD's Advancing AI event, we're previewing the first (as far as we know) 192GB system with AMD Ryzen™ AI Max+ PRO 495. This lets you run models like DeepSeek-V4-Flash in Q8 on a single box, with room to spare for context length. — Framework (@FrameworkPuter) July 22, 2026

According to Framework, the desktop is optimized for local LLMs and can run DeepSeek-V4-Flash in Q8 GGUF format on a single computer. This model contains a total of 284 billion parameters and 13 billion active parameters. A build of this kind occupies about 162 GB of memory. Thus, the 192 GB available not only allows loading the model but also maintains free space for a larger context window, a fundamental aspect for advanced AI applications.

Framework has also updated the PCIe x4 slot, which now adopts an open-ended design. This allows for the installation of longer PCIe cards while continuing to use four PCIe lanes.

To showcase the platform's capabilities, the company shared images of a cluster made up of two 192 GB desktops connected via two 50 GbE network cards with RDMA over Ethernet (RoCE). Thanks to tensor parallelism, the two systems can share 384 GB of memory with very low latencies.

Framework has also confirmed the arrival of a pre-assembled version with pre-installed Linux distribution. Prices and availability have not yet been announced, but the company has hinted that the 192 GB configuration will have a significantly higher cost than the current 128 GB model, as usual, due to rising prices of LPDDR5X memory.