Framework Opens Pre-Orders for $6,799 192GB AI Desktop as AMD's "Gorgon Halo" Chips Go on Sale

Local AI's first real workstation wave is here: desk-side boxes with server-class memory — but constrained DRAM supply means today's prices may be the cheapest they ever get.

October 1, 2026 — Framework opened pre-orders this week for a 192GB version of its compact Framework Desktop, built around AMD's new Ryzen AI Max+ PRO 495 processor — the flagship of the Ryzen AI Max PRO 400 series, known in coverage by its codename "Gorgon Halo" — starting at $6,799 for the DIY Edition. It caps a week in which AMD's partners finally put 192GB local-AI systems on sale, with GMKtec, MINISFORUM, and HP all attaching prices to machines built to run very large AI models entirely offline.

Image: A compact desktop AI workstation with a glowing processor and a holographic neural-network brain rising above it. Credit: Red Eye Eagle News / original AI illustration.

Editor's Highlight: The story is not the chip — it is the memory. With 192GB of unified memory and up to 160GB assignable to the GPU, a desk-side box can keep enormous AI models resident without cloud help — and Framework's single-batch warning means today's prices may not survive into the next run.

Table of Contents

• What Went on Sale This Week
• Inside the Ryzen AI Max+ PRO 495
• Why 192GB of Unified Memory Is the Real Story
• The Price Landscape: What the Machines Cost
• Local vs. Cloud: AMD's Cost Argument
• Bottom Line
• FAQ
• Sources

Key Facts

Detail	Information
Chip	AMD Ryzen AI Max+ PRO 495 — 16 Zen 5 cores / 32 threads, boost up to 5.2 GHz
Graphics / NPU	Radeon 8065S, 40 compute units; NPU rated up to 55 TOPS
Memory	192GB LPDDR5X-8533 unified; up to 160GB allocatable to the GPU
Framework pre-order	Opened Oct 1, 2026; DIY from $6,799; shipping November 2026
Rival systems	GMKtec EVO-X5 Pro $6,599; MINISFORUM MS-S1 MAX-P495 $7,399; HP ZBook Ultra G3a $7,449
AMD's headline claim	World's first x86 client chip to run 300B+ parameter models at 4-bit (per AMD)
The catch	Single initial pre-order batch; later prices tied to memory supplier costs

What Went on Sale This Week

Framework anchored the news. On Wednesday morning Pacific Time, the modular-PC maker opened pre-orders for the 192GB Framework Desktop: $6,799 for the DIY Edition, first systems shipping in November 2026, and a pre-built configuration with a 2TB SSD and Fedora Linux at $7,449.

Only one initial batch is planned, limited to memory already in stock. Framework says later batches' pricing will depend on memory supplier costs — a real qualification given the broader run-up in high-capacity DRAM pricing.

Framework was not alone. MINISFORUM put its MS-S1 MAX-P495 compact workstation on sale in the US at $7,399 with 192GB of memory and a 2TB SSD — reportedly the first Ryzen AI Max+ PRO 495 system actually on sale. GMKtec launched its EVO-X5 Pro at $6,599 for a 192GB and 2TB configuration, and HP's 16-inch ZBook Ultra G3a with the same chip and 192GB of memory was priced at $7,449.

Newsroom Focus: This is the first week the "Gorgon Halo" platform has moved from announced silicon to buyable systems from multiple vendors — a milestone for the idea of a private, desk-side AI server.

Inside the Ryzen AI Max+ PRO 495

The PRO 495 is the top SKU of AMD's three-chip Ryzen AI Max PRO 400 lineup:

• CPU: 16 full-size Zen 5 cores / 32 threads, boost up to 5.2 GHz
• Integrated GPU: Radeon 8065S, 40 RDNA 3.5 compute units, up to 3.0 GHz
• NPU: rated up to 55 TOPS of AI throughput
• Memory: up to 192GB LPDDR5X-8533 on a 256-bit interface, about 273 GB/s of bandwidth

The rest of the family scales down: the PRO 490 offers 12 cores / 24 threads with a ~50 TOPS NPU, and the PRO 485 offers 8 cores / 16 threads, also ~50 TOPS. All three share the 192GB memory ceiling and the up-to-160GB GPU allocation cap.

Why 192GB of Unified Memory Is the Real Story

Most PCs split memory: the CPU uses system RAM while the graphics card uses its own VRAM. A high-end consumer graphics card might offer 16GB or 24GB of VRAM — nowhere near enough to hold a very large AI model. Once a model exceeds VRAM, performance collapses as data shuttles back and forth.

The Ryzen AI Max PRO 400 platform uses unified memory: the CPU and integrated GPU share one large pool. Up to 160GB can be handed to the graphics engine, enough to keep very large quantized models resident in memory. AMD says the 192GB ceiling is a 1.67x increase over the prior generation's 128GB total and 96GB GPU cap.

Framework cites quantized versions of models like DeepSeek-V4.1-Flash as examples of what can run locally. Actual performance will depend on the inference engine, quantization format, and context size.

Tech Analysis: Capacity is the headline, but bandwidth is the quieter story. Bandwidth rises only about 7% while capacity jumps 50% — and LLM token generation is usually bandwidth-bound, so compute can remain the bottleneck even after a model fits in memory. Fit does not automatically mean fast.

The Price Landscape: What the Machines Cost

System	Config	Price
GMKtec EVO-X5 Pro	192GB RAM + 2TB SSD	$6,599
Framework Desktop (DIY)	192GB, no storage	$6,799
MINISFORUM MS-S1 MAX-P495	192GB + 2TB SSD	$7,399
HP ZBook Ultra G3a 16	192GB laptop	$7,449

For context, the older 128GB Framework Desktop now lists at $3,449 — and it originally launched at $1,999 before Framework pushed prices up repeatedly as DRAM costs climbed. The 192GB machine brings 50% more memory for nearly double the money; the memory market, not the silicon, is setting the bill.

Practical Takeaway: Framework itself says the current price is likely the lowest in the near term, because later batches track supplier memory quotes. Waiting for a price drop may be a losing bet.

Local vs. Cloud: AMD's Cost Argument

AMD is pitching these systems as "agentic PCs" for business: machines built to develop and run AI agents locally rather than dispatching every query to a cloud API. The arguments — laid out in AMD's own business-customer blog post — are privacy, data residency, and cost.

The privacy case: an agent working against internal documents or source code can operate on the local system instead of sending every request to an external inference service. The cost case: agentic workloads consume enormous token counts, and cloud inference turns each operation into a recurring expense — a vendor claim AMD backs with its Tokenomics Calculator model. HP, for example, is pairing the PRO 495 with Perplexity Portable Computer in the ZBook Ultra G3a, so users can run local agents alongside professional apps and tap the cloud only when a step needs frontier intelligence.

Bottom Line

The local-AI workstation has crossed from announcement to storefront: Framework's $6,799 pre-order, GMKtec's $6,599 launch, and MINISFORUM's $7,399 sale mean buyers can now actually purchase 192GB "Gorgon Halo" systems. What to watch next: whether later Framework batches hold the price as DRAM costs rise, independent benchmarks of 300-billion-parameter-class models on these boxes, and whether enterprise buyers treat local inference as a genuine alternative to cloud spend.

FAQ

What does "Gorgon Halo" mean?

"Gorgon Halo" is the codename used in press coverage for AMD's Ryzen AI Max PRO 400 series — the follow-up to the earlier "Strix Halo" platform. AMD's official product name is the Ryzen AI Max and Ryzen AI Max PRO 400 Series.

Can these machines really run 300-billion-parameter AI models?

That is AMD's headline claim: the Ryzen AI Max+ PRO 495 is, per the company, the world's first x86 client processor able to run models over 300 billion parameters at 4-bit quantization, thanks to up to 160GB of GPU-addressable memory. Framework points to quantized models like DeepSeek-V4.1-Flash as practical examples. Actual speed will depend on the inference engine, quantization, and context size.

Will the $6,799 Framework price last?

Probably not as a guarantee. Framework opened only one pre-order batch tied to its current memory inventory and says later pricing will track memory supplier costs. Given rising DRAM prices, the company suggests the current price is likely the lowest in the near term.

Who are these systems actually for?

Developers, researchers, creators, and enterprises that want to run large AI models and agents locally for privacy, latency, or cost reasons. At $6,799, the appeal is concentrated in private local inference and memory-heavy development — not mainstream gaming value.

Sources

• Guru3D — Framework Desktop Adds Ryzen AI Max+ PRO 495 and 192GB LPDDR5X for $6,799 (pre-order details, specs, shipping)
• AMD (official blog) — How AMD Ryzen AI Max PRO 400 Series Processors Bring Local Agentic AI to Business Customers (chip specs, 300B-parameter claim, Tokenomics Calculator)
• Hardware Busters — Framework Desktop Goes to 192GB With Gorgon Halo (price history and DRAM context)
• VideoCardz — Framework Desktop 192GB pre-order and MINISFORUM MS-S1 MAX-P495 at $7,399
• Wccftech — GMKtec Prices EVO-X5 Pro AI Workstation at $6,599
• TheOutpost.ai — ACEMAGIC and GMKtec AI mini workstations
• YouTube — Wait, AMD Made a CHEAP AI Computer!? (explainer on the PRO 495, 192GB memory, and local model claims)
• Hero image: original AI illustration generated for this article by Red Eye Eagle News.