Framework Desktop with Ryzen AI Max+ Pro 495 and 192GB Memory Opens Pre-Orders at $6,799

Framework Desktop Opens Pre-Orders with Massive Memory Configuration

Framework has announced that pre-orders for its most capable Desktop configuration will open on Wednesday, September 30, in the morning Pacific Time2

2

3

3

. The small-form-factor Desktop PC is built around AMD's Ryzen AI Max+ Pro 495 processor and features 192 GB unified memory, positioning it as a specialized tool for AI workloads and memory-intensive development. The DIY Edition starts at $6,799, with initial systems scheduled to ship during November 20261

1

.

The pricing reflects rising LPDDR5X costs, prompting Framework itself to suggest buyers consider lower-capacity models at 32GB, 64GB, or 128GB based on their budget2

2

4

4

. The current Ryzen AI Max+ 395 system with 128GB starts around $3,4493

3

, making the 192GB configuration nearly double the price. Rival systems with similar memory capacity include GMKtec's EVO-X5 Pro at $6,800, while Apple's Mac Studio with M5 Ultra starts at $5,500 and can exceed $6,800 in higher configurations2

2

.

AMD Ryzen AI Max+ Pro 495 Delivers Enhanced Performance

The Ryzen AI Max+ Pro 495, also known as Gorgon Halo, was first previewed at AMD's Advancing AI event in July2

2

. The processor combines 16 Zen 5 CPU cores with multithreading and Radeon 8065S integrated graphics containing 40 RDNA 3.5 compute units running at 3.0 GHz4

4

. Framework reports that both CPU and GPU clock speeds are slightly higher than those of its original Ryzen AI Max+ 395 configuration, with the boost clock reaching 5.2 GHz2

2

4

4

. The system delivers 131 TOPS of combined AI performance2

2

.

The core layout matches last year's Strix Halo-based chip, but the frequency improvements and upgraded graphics make it more capable for AI-accelerated processor tasks2

2

4

4

. Framework specifically designed this configuration to handle local AI models that demand substantial memory allocation without requiring separate discrete graphics cards.

192 GB Unified Memory Enables Large Language Model Deployment

The most significant upgrade centers on memory capacity and bandwidth. The system provides 192 GB of soldered LPDDR5X-8533 connected through a 256-bit interface, delivering 273 GB/s of memory bandwidth1

1

4

4

. This represents a 6.6% increase in memory speeds and bandwidth, plus a 50% boost to capacity compared to the Ryzen AI Max+ 395 systems, which offered 256 GB/s bandwidth and up to 128 GB capacity4

4

.

Because the processor uses a unified-memory architecture, up to 160 GB can be allocated to the Radeon GPU2

2

. This allows unusually large local AI models to run without the separate VRAM limit imposed by conventional graphics cards. Framework cites quantized versions of DeepSeek V4-Flash at Q8 and MiMo V2.6 Flash RL as examples of models that can run locally on a single box, with room left for longer context length1

1

2

2

4

4

.

The unified memory approach does not make LPDDR5X equivalent to HBM, however, and compute throughput can remain the limiting factor after a model fits into memory1

1

. Actual performance will depend on the inference engine, quantization format, context size, and workload configuration. The memory is soldered onto the board and cannot be upgraded later, making the initial capacity choice critical for buyers planning long-term use2

2

.

Connectivity and Expansion Options for AI Workloads

The Mini-ITX mainboard includes two M.2 SSD positions and an open-ended PCIe 4.0 x4 slot1

1

3

3

. Full-length x8 or x16 cards can now be inserted physically without a riser, although they remain electrically limited to four PCIe lanes1

1

. Framework specifically mentions 50GbE adapters such as Intel's E810 for connecting multiple systems, and demonstrated a concept cluster using two 50 GbE NICs that enable RDMA over Ethernet (RoCE)1

1

4

4

. This configuration makes it possible to effectively pool 384 GB of memory across two desktops with tensor parallelism, offering ultra-low latency for distributed AI workloads4

4

.

A firmware update increases point-to-point USB4 networking throughput from approximately 9Gbps to around 30Gbps1

1

. This creates a lower-cost alternative for experimenting with clustered inference, though performance remains well below specialized high-speed GPU interconnects. Standard connectivity includes two USB4 Type-C ports, two USB-A ports, two front Expansion Card positions, two DisplayPort 2.1 outputs, HDMI 2.1, 5Gb Ethernet, Wi-Fi 7, and a 3.5mm audio connection1

1

.

Cooling, Operating System, and Limited Availability

Both the DIY and prebuilt systems include a Noctua NF-A12x25 HS-PWM 120mm fan pre-installed1

1

2

2

. The prebuilt version adds a 2TB SSD and Fedora Linux1

1

. Framework explicitly states that Windows configurations are not offered for the Ryzen AI Max+ Pro 495 model, although Windows 11 is technically supported and this does not necessarily mean Windows is impossible to install1

1

3

3

.

Only one initial pre-order batch is planned because of constrained memory inventory1

1

. Framework says later pricing and availability will depend on memory costs, an important qualification given the broader increases affecting high-capacity DRAM. The company has retained the current motherboard and case design to preserve compatibility with existing parts, following an approach it has also used for Framework Laptop upgrades3

3

.

At $6,799, the 192 GB configuration targets private local inference and memory-heavy development where capacity matters more than conventional value propositions1

1

. The system positions itself against NVIDIA's DGX Spark, which retains 128 GB LP5X memory and costs close to $5,0004

4

. For buyers who need maximum memory for running large language models locally without cloud dependencies, the Framework Desktop with Ryzen AI Max+ Pro 495 offers an AI punch in a tiny box, though the price premium reflects both component costs and the specialized nature of the configuration.