I've been watching local AI hardware get more interesting all year, but one thing keeps happening: the number everyone talks about is usually the processor.
With Framework's newest Desktop, that is not the most important number.
It is 192GB.
Framework has opened pre-orders for a new 4.5-liter Desktop built around AMD's Ryzen AI Max+ PRO 495, with 192GB of soldered LPDDR5X unified memory.
The starting price is $6,799.
And despite that eye-watering number, the first production batch sold out within hours. Framework then secured additional memory for another limited batch.
So what are buyers actually paying for?
Framework's new 192GB Desktop pairs AMD's Ryzen AI Max+ PRO 495 with a huge pool of unified memory for local AI workloads.
What Is the New Framework Desktop?
The machine is the latest evolution of Framework's compact Desktop platform.
Instead of using a conventional desktop CPU plus a separate graphics card, it centers on AMD's Ryzen AI Max+ PRO 495, an APU that combines Zen 5 CPU cores with Radeon 8065S integrated graphics and a dedicated NPU.
Framework rates the chip at 16 CPU cores and 32 threads, with boost clocks up to 5.2GHz.
The GPU has 40 compute units, while AMD lists up to 55 TOPS of NPU performance and up to 131 TOPS of total AI compute.
But those numbers are not why this machine is unusual.
192GB of Unified Memory Is the Real Story
The new Desktop uses 192GB of LPDDR5X-8533 memory across a 256-bit bus.
That memory is unified, meaning the CPU and integrated GPU share the same pool instead of having completely separate system RAM and discrete VRAM.
For local AI, however, unified memory can be incredibly useful because model weights do not have to fit entirely inside a small dedicated graphics-memory pool.
Framework says the extra 64GB over its previous 128GB configuration allows larger models to remain loaded in memory on one machine instead of relying on SSD offloading or multiple systems.
The Upgrade Is Bigger Than the CPU Upgrade
It is easy to look at the Ryzen AI Max+ PRO 495 and assume the processor itself is the reason this computer costs so much.
That is only part of the story.
Framework's own comparison shows that the newer architecture keeps the same basic Zen 5 CPU and RDNA 3.5 GPU design used by Strix Halo.
Memory bandwidth rises from roughly 256GB/s on the previous 128GB configuration to 273GB/s here, an increase of about 6.6%.
That is meaningful, but it is not a generational performance explosion.
The giant jump is memory capacity.
Why Local AI Users Care So Much About Memory
Running a large AI model locally is often less about raw CPU speed than whether you can keep the model in memory without constantly moving data back and forth.
The larger the model, the more memory pressure you face.
Framework's own local-AI documentation now highlights models such as DeepSeek-V4.1-Flash Q2 and MiMo V2.6 Flash RL MXFP4 as examples that can be run on the 192GB configuration.
The practical benefit is straightforward.
You can experiment with much larger open-weight models without immediately moving to a multi-GPU workstation or a cloud server.
Andrew Ng's Old AI Prediction Fits This Hardware Perfectly
— Andrew Ng
Ng made that comparison because he expected AI to become infrastructure that spreads across industries.
Framework is betting on another part of that future: more people owning the compute themselves rather than renting every serious AI workload from a cloud provider.
That matters for privacy-sensitive workloads, experimentation, development and situations where recurring inference costs can become expensive.
The $6,799 Price Needs Context
This is where the story becomes genuinely interesting.
Framework's existing 128GB Desktop with Ryzen AI Max+ 395 currently starts around $3,449.
The new 192GB model starts at $6,799.
That means you are getting 50% more memory, a newer processor configuration and somewhat higher memory bandwidth, but the entry price is almost twice as high.
What That $6,799 Actually Buys
- AMD Ryzen AI Max+ PRO 495: 16 Zen 5 cores and 32 threads.
- 192GB unified memory: LPDDR5X-8533 shared by CPU and GPU.
- Radeon 8065S: 40 compute units based on RDNA 3.5.
- 4.5-liter chassis: an unusually small footprint for this memory capacity.
- Linux-focused configuration: Framework's new system is offered with Fedora in the pre-built version and does not offer Windows as a pre-installed option.
The reason for that huge price jump is not simply the processor.
Framework has repeatedly warned that high-capacity memory has become substantially more expensive, and the company expects memory costs to continue rising.
The 192GB Memory Is Soldered — and That's Easy to Miss
Here is the uncomfortable trade-off.
Framework's Desktop is built around a modular philosophy, but the LPDDR5X memory itself is soldered.
That means the most expensive and strategically important component in this configuration is not something you can replace with a larger memory kit later.
Think About Your Model Requirements Before Buying
Do not buy 192GB simply because the number looks impressive. Choose the capacity based on the models, context lengths, concurrent workloads and inference engines you actually intend to run.
This is especially important because Framework itself now offers 32GB, 64GB, 128GB and 192GB configurations for different local-AI workloads.
Framework Also Made an Interesting PCIe Change
The 192GB configuration includes an open-ended PCIe x4 slot.
That unusual design lets users physically install larger x8 or x16 cards directly without a riser, opening the door to networking and accelerator hardware that would normally be awkward in a small system.
Framework specifically discusses using an Intel E810 networking adapter for high-speed RoCE deployments and clustering multiple Framework systems.
This is one of the more interesting details because it shows Framework is not treating the machine only as a desktop PC.
It is also positioning the platform as a building block for compact AI clusters.
USB4 Is More Important Than It Looks
Framework also says a firmware improvement raises USB4 point-to-point networking throughput from around 9Gbps to approximately 30Gbps.
That makes USB4 more useful for linking multiple systems together.
In other words, the machine can start as one local AI workstation and eventually become part of a larger cluster.
Framework Desktop 192GB: Pros and Cons
Potential Advantages
- 192GB unified memory in a 4.5-liter chassis
- Excellent fit for large local AI models
- 16-core Zen 5 CPU
- 40-CU Radeon 8065S integrated GPU
- 273GB/s memory bandwidth
- Strong expansion and clustering options
- Mini-ITX mainboard format
Important Trade-Offs
- $6,799 starting price is extremely high
- 192GB LPDDR5X memory is soldered
- Shared memory is not equivalent to 192GB dedicated VRAM
- Linux is the focus of this configuration
- Storage adds to the DIY system cost
- Availability depends heavily on memory supply
Amazon Alternatives for Local AI
MINISFORUM MS-S1 MAX 128GB
The MS-S1 MAX is another compact AMD Ryzen AI Max+ 395 system with up to 128GB unified memory. It targets large local-AI workloads while offering features such as high-speed networking and expansion.
Check MS-S1 MAX on Amazon →GMKtec EVO-X2 128GB
The EVO-X2 is another Ryzen AI Max+ 395 mini PC aimed at local AI users. Its 128GB unified-memory configurations provide a lower-memory alternative to the new 192GB Framework system.
Check EVO-X2 on Amazon →Samsung 990 PRO 2TB NVMe SSD
The DIY Framework configuration does not include storage, so a fast PCIe 4.0 NVMe SSD is one of the obvious additions for anyone planning to keep large local AI models, datasets and development environments on the machine.
Check Samsung 990 PRO on Amazon →Why the First Batch Sold Out
The sellout is easier to understand when you look at the product as a piece of AI infrastructure rather than a normal desktop PC.
People buying this machine are likely not comparing it only with gaming desktops.
They may be comparing it with workstation-class systems, multiple GPUs, cloud inference bills or the engineering complexity of running a local cluster.
That makes the $6,799 price look very different depending on the workload.
For ordinary office computing, it makes little sense.
For someone specifically trying to keep very large models resident in memory locally, the machine occupies a much more unusual category.
What Happens Next?
Framework's first batch sold out quickly, but the company has since secured enough memory for another limited batch.
The current systems are expected to begin shipping in November 2026.
Framework has also warned that future pricing will depend on the memory market and supplier costs.
That makes availability an important part of the story rather than a footnote.
The Bigger Local-AI Trend
- More memory: AI PCs are moving beyond conventional desktop memory capacities.
- Unified architectures: CPU and GPU memory pools are becoming increasingly useful for local inference.
- Smaller hardware: large AI workloads are moving into compact systems rather than only traditional towers.
- Clustering: USB4 and high-speed networking make multiple small systems more practical.
- Ownership: enthusiasts are increasingly interested in running open-weight models locally instead of paying for every inference request.
The Real Question Is Whether You Need 192GB
That is the question hidden underneath all the excitement.
If your local AI work fits comfortably inside a 32GB, 64GB or 128GB system, buying 192GB simply because it is the largest option may be difficult to justify.
But when your workload is constrained by model size, context length, concurrent models or memory capacity, the additional 64GB can be much more important than a small increase in CPU or GPU speed.
That is what makes the Framework Desktop 192GB unusual.
It is not trying to be the fastest desktop for everyone.
It is trying to be a surprisingly small machine that can keep unusually large AI workloads in memory without requiring a conventional multi-GPU server.
And that explains why a computer costing $6,799 attracted so much attention so quickly.
Meet the 192GB FEVM FAEX2 Mini PC
If the $6,799 Framework Desktop is sold out or outside your budget, it isn't the only 192GB option for running large models locally. Read our complete guide to the FEVM FAEX2 to see how this rival high-capacity Mini PC compares in specs, pricing, and everyday local AI performance.
Read the FEVM FAEX2 Guide →Sources checked for this article:
Framework — 192GB Framework Desktop open for pre-order
Framework — What 192GB Changes for Local AI on the Framework Desktop
Framework — Choosing a Framework Desktop for Local AI
Framework — Framework Desktop 192GB Specifications
AMD — Ryzen AI Max+ PRO 495 Specifications
Phoronix — Framework Desktop With Ryzen AI Max+ PRO 495 Starts at $6,799
TechPowerUp — AMD Ryzen AI Max+ PRO 495 Framework Desktop Batch 1 Sells Out
Amazon — MINISFORUM MS-S1 MAX 128GB
YouTube — Framework Desktop 192GB Ryzen AI Max+ PRO 495
Stanford Graduate School of Business — Andrew Ng: Why AI Is the New Electricity
Frequently Asked Questions (FAQ) About the Framework Desktop 192GB
How much does the 192GB Framework Desktop cost?
The new Framework Desktop with AMD Ryzen AI Max+ PRO 495 starts at $6,799 for the DIY Edition. A pre-built configuration with a 2TB NVMe SSD and Fedora pre-installed costs $7,449.
Why does the Framework Desktop have 192GB of memory?
The 192GB unified-memory configuration is designed primarily for large local AI workloads. The additional capacity lets users keep larger models in memory on one machine and can also allow multiple smaller models to remain loaded simultaneously.
Is 192GB unified memory the same as 192GB VRAM?
No. The Framework Desktop uses unified memory shared by the CPU and GPU. It is not equivalent to a discrete graphics card with 192GB of dedicated VRAM.
What processor is inside the 192GB Framework Desktop?
The system uses AMD's Ryzen AI Max+ PRO 495, with 16 Zen 5 CPU cores, 32 threads, boost clocks up to 5.2GHz, Radeon 8065S graphics with 40 compute units, and an NPU rated for up to 55 TOPS.
When will the Framework Desktop 192GB ship?
Framework says shipments for the new 192GB configuration are scheduled to start in November 2026. Availability depends on memory supply, and Framework has said future batch pricing and timing will depend on supplier costs and availability.
No comments:
Post a Comment