Feed: Engadget is a web magazine with obsessive daily coverage of everything new in gadgets and consumer electronics
Feed Link: https://www.engadget.com/
---

Title: Framework Desktop (2025) Review: Powerful, but perhaps not for
everyone

Date: Thu, 07 Aug 2025 15:00:11 +0000
Link: https://www.engadget.com/computing/framework-...

The most obvious question is ΓÇ£Why?ΓÇ¥

Framework builds modular, repairable laptops that anyone can take apart and
put back together again. ItΓÇÖs a big deal in an era where laptops are
regularly sold as a single unit that, should one part break, goes in the
trash. Since every part of a Framework machine can be swapped out, you can
keep one going for as long as your patience, and the supply of spare parts,
allows. Desktops, however, are already modular and repairable ΓÇö company
founder Nirav Patel said ΓÇ£desktop PC ethos was one of the core inspirations
for the Framework laptop to begin with.ΓÇ¥ So, if desktops are already
modular and repairable, why do we need one from Framework?

When the Desktop was announced, Patel said the genesis of the product came
from seeing a preview of AMDΓÇÖs Ryzen AI Max. ItΓÇÖs an APU (Accelerated
Processing Unit) ΓÇö AMDΓÇÖs term for a chip combining a CPU, GPU and NPU in
a single package, much like Apple Silicon ΓÇö with plenty of hyped-up claims
about its performance. Those claims were so compelling that Patel added the
Desktop to the companyΓÇÖs roadmap just to harness that potential power. The
big selling point for this chip is the sheer volume of RAM you can employ (up
to 128GB) and the massive memory bandwidth (up to 265GB/s) it can take
advantage of. AMD described it as a ΓÇ£workstation-levelΓÇ¥ chip thatΓÇÖll
work in a regular ΓÇÿol PC, with the base model priced at $1,099.

But thereΓÇÖs a devilΓÇÖs bargain in opting for such a powerful chip, since
to get it means Framework has had to give up a lot of its founding
principles. As someone probably once wrote, for what shall it profit a
computer manufacturer if it shall gain searing power but lose its own soul?

Hardware

The major issue with the Ryzen AI Max is its inflexibility since itΓÇÖs made
as a single package. Much like Apple Silicon products, youΓÇÖll need to pick
your chip spec in the knowledge that you donΓÇÖt get to change things later.
Consequently, youΓÇÖll be ordering the Framework Desktop in one of three
unchangeable flavors:

Ryzen AI Max 385 with 32GB RAM, Radeon 8050S GPU

Ryzen AI Max+ 395 with 64GB RAM, Radeon 8060S GPU (the model IΓÇÖm testing).

Ryzen AI Max+ 395 with 128GB RAM, Radeon 8060S GPU

With every other Framework machine, the mainboard has the CPU and fan
soldered in place, but thatΓÇÖs it. Every other component can more or less be
removed and reinstalled on the replacement mainboard. Here, if that APU goes
or if your needs do evolve, then youΓÇÖre losing pretty much everything
(including the heatsink) since itΓÇÖs all soldered to the mainboard. We'll
get into pricing considerations later, but replacement mainboards from
Framework run between $799 and $1,699. Otherwise, the only things you can
recover here are the Wi-Fi module, SSD (thereΓÇÖs space for two), power
supply and case.

Speaking of which, the Framework Desktop is a Mini-ITX desktop in a 4.5L
case, complete with optional carrying handle. ItΓÇÖs less understated than
the cases youΓÇÖd have seen at a LAN party circa 2006, but thatΓÇÖs not the
point. While the box itself is a stark black, you can add a big chunk of
personality to it with the front panel, which has space for 21 plastic tiles.
These tiles come in a variety of colors (including black, green, orange and
lavender) for you to mosaic to your heartΓÇÖs content. You can also pick up
single tiles with specific images printed on, including the Framework, AMD
and Linux logos, plus this fetching pride heart. Naturally, if youΓÇÖre
crafty, you can also make your own.

Below the front panel and irritatingly small power button, youΓÇÖll find two
of FrameworkΓÇÖs trademark expansion card slots. These are USB-C ducts into
which you can slot any of the companyΓÇÖs expansion cards, letting you pick
and choose what I/O you have up front. But the flexibility, so necessary on a
laptop, is less of an issue here since this is a Mini-ITX mainboard. Lean
over to the back and youΓÇÖll find two USB-C, two DisplayPorts, two USB-A
sockets as well as dedicated connections for HDMI, Ethernet and 3.5mm audio.

Installation

Framework is only selling its desktop in a ΓÇ£DIY Edition,ΓÇ¥ but thatΓÇÖs
less of a big deal than you might initially expect. Whereas the companyΓÇÖs
DIY laptops require you to put every component in the chassis, on the Desktop
thereΓÇÖs very little to do at all. Everything bar the SSD is already in
place, and all you need to do to add that is remove the heatsink and slot
your drive into the M.2 slot. After that, you just need to attach the beefy
120mm fan to the equally beefy heatsink, pop the cowl on top and screw in the
four screws. FrameworkΓÇÖs wonderful iFixit-style guides claimed getting the
hardware together would take between 30 and 45 minutes. I hit stop on the
watch after 15 minutes and 15 seconds, and canΓÇÖt imagine many folks will
take much more time than that to put everything together. From there, you
just need to install your operating system of choice and youΓÇÖre ready to
go.

In use Daniel Cooper for Engadget

With the Desktop, Framework is targeting two groups: gamers, and developers
eager to use AMDΓÇÖs Ryzen AI Max. The pitch to the former group could easily
be boiled down to ΓÇÿthis is a fast PC you donΓÇÖt need to do much to
build.ΓÇÖ ThatΓÇÖs an easy enough metric to judge it by, since we can just
run some games on it and see how well it performs. IΓÇÖll admit that I am not
an AI developer, and so canΓÇÖt speak as authoritatively on the latter or how
effective it would be at running large models if you ΓÇö as the company
expects ΓÇö buy several mainboards to run in a cluster.

I basically ran every title in my admittedly limited game library with the
settings dialed up to max, and it didnΓÇÖt break a sweat. AMD claims the
Radeon 8060S GPU inside my machine goes toe-to-toe with an RTX 4070 laptop
GPU. This is a ten pound hammer for the one ounce nail that is Fortnite, but
even demanding titles like Hardspace: Shipbreaker breezed through. My gut
tells me, however, that people wouldnΓÇÖt be eyeing this up as a primary
gaming machine. ThatΓÇÖs not where this unitΓÇÖs power lies, really, but in
the more work-y tasks that better suit the APU.

As I said, IΓÇÖm not an AI developer but I did mess around with LM Studio,
which I tested with a chatbot running GoogleΓÇÖs Gemma 3 27B model.
Performance was a little slower than you may see on a web-based AI client,
but not enough for it to be an issue. LM Studio, too, suggested that running
this was only taking around a third of the DesktopΓÇÖs CPU power, so
thereΓÇÖs probably plenty more headroom there to run bigger and more
demanding models.

I am, however, on surer footing with big workstation tasks, like video
editing and exporting, and I was impressed with the results here. For this, I
took a 39GB HD video file with a runtime of 2 hours, made a few minor trims,
and then compressed and exported it as an MP4 file. Crunching the file down
to 6GB took just one hour and 12 minutes, a staggering speed boost for a job
that could take half a day to export on lesser hardware. ItΓÇÖs worth
remembering, too, that IΓÇÖm testing the middle-tier version of the Desktop
with 64GB RAM.

When announcing the Desktop, Patel said the machine would run quietly even at
peak power. He described it as ΓÇ£silent while sitting on your desktop under
normal loads, and even under gaming, itΓÇÖs impressively quiet.ΓÇ¥ If
youΓÇÖre familiar with FrameworkΓÇÖs track record and products so far, that
quote will have instantly provoked scoffing. The company does many things
well, but it wildly overpromises on how quiet and cool its machines run.

Framework trumpeted how much better its cooling was on its recently-released
Ryzen AI 300 mainboard for the Laptop 13, which was noisy and lap roasting.
Here, youΓÇÖve got a beefy APU expected to run for sustained periods of time
at 120W and up to 140W in boost. I half expected to be able to use this thing
as a space heater but, mercifully, the company does seem to have made good on
its promises. The 120mm fan barely got noisy at all, and I can only recall it
becoming noticeable when running heavier AI models in LM Studio and when I
started exporting the video file.

Pricing

The starting price for the base model Desktop with the Max 385 and 32GB of
RAM is $1,099. For that, youΓÇÖll get the case, power supply and mainboard,
which includes its own Wi-Fi module. What the company is listing as optional
extras, however, includes the SSD, CPU fan, OS and even the power lead. So,
if you were looking to buy the base model as an essentially off the shelf
purchase, including decorative tiles and two front-facing expansion cards,
the price rises to $1,386. If you want to opt for the mid-tier option (the
395 with 64GB RAM) add $500 to the base model price. If you want to go for
the high end 395 with 128GB RAM, then youΓÇÖll be adding $900 to the base
price, bringing the total for an off-the-shelf model to around $2,286.

Because of the distinct nature of the Ryzen AI Max, an apples-to-apples
comparison isnΓÇÖt going to be perfect. But, if you were looking to spend
around two grand on a high-performance PC, you could snag something like
LenovoΓÇÖs Legion Tower 5 with AMD. $2,200 buys you a Ryzen 7 with 16GB DDR5
RAM, a 1TB SSD and NVIDIAΓÇÖs GeForce RTX 4070 Ti with 12GB RAM.
Alternatively, that figure could get you an ASUS ROG G700 with a Ryzen 7,
Radeon RX 9070 XT Prime (with 16GB VRAM), 16GB RAM and a 2TB SSD. The real
differentiator is how much you would value the faster memory bandwidth and
speed the AI Max offers over its discrete rivals.

Framework will also sell users the DesktopΓÇÖs mainboard on its own, with the
base model priced at $799. Certainly, if youΓÇÖve got boxes full of spare
parts and the necessary Mini-ITX case, you could save a chunk of change that
way. This will also be the most cost effective way for power users to run
clusters of boards for bigger projects. Pre-orders are open, but the company
wonΓÇÖt start shipping boards on their own until itΓÇÖs fulfilled all of its
system orders.

Wrap-up

In my time with the Framework Desktop, IΓÇÖve flip-flopped a few times on if
this product is a hit or a miss. Framework was open about the fact this was
something akin to a side project, outside its regular remit to build modular
laptops, based on a particularly exciting chip. So while I think the all-in-
one approach is a backward step compared to regular PCs, I get the rationale
for doing so here.

Where I think Framework whiffed was to pitch this as a machine to make ΓÇ£PC
gaming more accessibleΓÇ¥ by reducing ΓÇ£the mental and physical loadΓÇ¥
associated with building your own. Nightmares about thermal paste aside, I
donΓÇÖt think thatΓÇÖs a real issue for would-be gamers as they could easily
pick up a pre-built system for similar cash. And I suspect most gamers would
much rather use a PC with a standalone graphics card rather than slum it with
an integrated GPU.

Because even mentioning gaming, really, does the machine a massive
disservice, pulling the focus from its real strength. Which is the ease at
which this machine handled productivity tasks, like running AI models and
crunching video. The effortlessness at which it handled that brought to mind
products like the Mac Studio, a creative powerhouse in its own tiny package.
ItΓÇÖs this that Framework should have led with, especially since itΓÇÖll do
all of those tasks and play games on the side.

IΓÇÖm not sure IΓÇÖd recommend this product to people who are just looking to
buy a powerful PC or a gaming PC. ItΓÇÖs a tool for a specific group of users
capable of taking advantage of the AI MaxΓÇÖs benefits that youΓÇÖd otherwise
need a workstation for. So while its review score is justifiably high when
the Framework Desktop is judged on its own merits, that doesnΓÇÖt mean you
need to own one.

This article originally appeared on Engadget at
https://www.engadget.com/computing/framework-...
perhaps-not-for-everyone-150011909.html?src=rss

---
VRSS v2.1.180528