Framework Previews Desktop With AMD Ryzen AI Max+ PRO 495 and 192 GB Memory

The push for local AI execution requires massive memory pools, and Framework is attempting to solve this by scaling the unified memory ceiling.

Ryzen AI Max+ PRO 495 Technical Specifications

The core of the new system is the Gorgon Halo-series chip. According to VideoCardz, the processor retains 16 Zen 5 cores and 40 RDNA 3.5 Compute Units, but pushes the CPU boost clock to 5.2 GHz. On the graphics side, it utilizes Radeon 8065S graphics and includes an NPU rated for up to 55 TOPS.

The most critical shift is the jump from 128 GB to 192 GB of unified memory. As Wccftech reports, this memory operates at 273 GB/s (8533 MT/s LPDDR5X), representing a 6.6% increase in bandwidth and a 50% boost in capacity compared to the previous 395 systems.

Feature	Ryzen AI Max+ PRO 395 (Previous)	Ryzen AI Max+ PRO 495 (New)
Max Memory Capacity	128 GB	192 GB
Memory Bandwidth	256 GB/s	273 GB/s
CPU Boost Clock	Lower than 5.2 GHz	5.2 GHz
Cores / GPU Units	16 Zen 5 / 40 RDNA 3.5	16 Zen 5 / 40 RDNA 3.5

DeepSeek-V4-Flash and Local LLM Performance

Framework is positioning this hardware specifically for AI Large Language Models (LLMs). The 192 GB capacity allows the system to run DeepSeek-V4-Flash at Q8 precision on a single box. This specific model has 284 billion total parameters and 13 billion active parameters; a Q8 GGUF build of it occupies roughly 162 GB of space.

By providing 192 GB, Framework ensures that the model fits while leaving enough overhead for a longer context window. This puts the machine in a different performance bracket than existing 128 GB platforms, such as NVIDIA’s DGX Spark, which Wccftech notes costs close to $5,000 US.

PCIe Modifications and the 384 GB Memory Cluster

Beyond the silicon, the physical architecture of the Framework Desktop has changed. The PCIe x4 slot is now an open-ended design, which allows for the installation of longer PCIe cards directly. While these cards still operate through four PCIe lanes, the open design removes previous physical restrictions.

To demonstrate this, Framework showcased a proof-of-concept cluster connecting two 192 GB systems using two 50 GbE network adapters with RDMA over Ethernet (RoCE). This setup uses tensor parallelism to effectively pool 384 GB of memory across the two computers, enabling the execution of even larger language models.

Pricing Pressure and LPDDR5X Costs

High-capacity memory is becoming prohibitively expensive. Framework has warned that the 192 GB configuration will cost substantially more than the 128 GB model due to rising LPDDR5X prices. While official pricing for the 495 model has not been announced, Wccftech expects a retail price exceeding $5,000 US later this year.

This pricing volatility is already appearing in other Framework products. Sweclockers reports that Framework has raised prices on Laptop configurations with 32 GB and 64 GB of primary memory by nearly 100 percent, with the 64 GB option now costing 1 600 dollar in the USA.

For context, the base price of the older MAX+ 395 system from Framework is approximately $3,500 US. Framework currently recommends that users choose between 32 GB, 64 GB, or 128 GB models based on their specific budget if they cannot justify the jump to the 192 GB tier.

Linux Integration and Market Competition

To better serve the developer and AI community, Framework is preparing a prebuilt version of the Desktop that comes with a Linux distribution pre-loaded. This move aligns with the system’s target audience of researchers and engineers who typically prefer Linux for LLM deployment.

Framework is not alone in this hardware race. VideoCardz notes that GMKtec has also confirmed an EVO-X3 configuration utilizing the same Ryzen AI Max+ PRO 495 processor and 192 GB memory capacity, expected to arrive later this year.