Framework Desktop review: AMD's Strix Halo in a petite, powerful, pricey PC

1 day ago 19

The Framework desktop is small and powerful, even for 1080p gaming. But for the price, you can get a full-on gaming rig with discrete graphics.

Pros

  • +

    Strong 1080p performance on integrated graphics

  • +

    Fun and customizable

  • +

    Small footprint fits easily on a desk

Cons

  • -

    Expensive

  • -

    Required to buy tiles to decorate the chassis

  • -

    PSU cable not included

  • -

    Not as upgradeable as traditional desktops

Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.

When Framework built its first laptop, it was with the ethos of a desktop: repairable, customizable, and upgradable when new parts came out. Now, Framework has a desktop of its own.

But the Framework Desktop ($1,599 as tested before a myriad of components and add-ons), isn't a traditional PC by any means. Using an AMD Ryzen AI Max+ 395 “Strix Halo”

APU

with soldered RAM, the customization options on this desktop are limited to storage, Wi-Fi, your case fan, and some decorative elements. The soldered RAM is effectively required for the bandwidth this system needs, and the mix of powerful graphics and plentiful RAM is why Framework is pushing this as an AI system.

The Desktop has a ton of charm, from its 4.5-liter size to its surprising power with the right configuration. But it's also very expensive compared to the

best gaming PCs

in its price range, and if you don't use all of the RAM, you might be better off with a cheaper system.

Framework Desktop
(Image credit: Tom's Hardware)

Stay On the Cutting Edge: Get the Tom's Hardware Newsletter

Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.

Framework Desktop
(Image credit: Tom's Hardware)
Framework Desktop
(Image credit: Tom's Hardware)
Framework Desktop
(Image credit: Tom's Hardware)

Swipe to scroll horizontally

Processor

AMD Ryzen AI Max+ 395

Motherboard

Framework FRANMFCPO4

Memory

64GB LPDDR5-8000

Graphics

AMD Radeon 8060S (integrated)

Storage

1TB WD Black SN7100 SSD

Networking

AMD RZ717 Wi-Fi 7, 5Gbit Ethernet

Front Ports

Two expansion card slots of your choosing

Rear Ports (Motherboard)

HDMI 2.1, 2x DisplayPort 2.1, 2x USB 4 Type-C, RJ45 Ethernet, 2x USB 3.2 Gen 2 Type-A, 3.5 mm headphone jack

Power Supply

FlexATX 400W PSU

Cooling

Heatsink, Cooler Master Mobius 120 CPU fan

Operating System

Windows 11 Home

Dimensions

8.9 x 8.09 x 3.81 inches (226.1 x 205.5 x 96.8 mm)

Price as Configured

$1,599 + $99 SSD + $22 fan kit + $5 power cable, before expansion cards and case tiles

Framework Desktop
(Image credit: Tom's Hardware)
Framework Desktop
(Image credit: Tom's Hardware)
Andrew E. Freedman

Andrew E. Freedman is a senior editor at Tom's Hardware focusing on laptops, desktops and gaming. He also keeps up with the latest news. A lover of all things gaming and tech, his previous work has shown up in Tom's Guide, Laptop Mag, Kotaku, PCMag and Complex, among others. Follow him on Threads @FreedmanAE and BlueSky @andrewfreedman.net. You can send him tips on Signal: andrewfreedman.01

  • ottonis

    While it may seem that soldered RAM is antithetical to Framework's product philosophy, one need to have in mind that replaceable RAM moduels would entirely defeat the purpose of a Strix Halo CPU/APU, which NEEDS super-fast and super large memory bandwidth in order to feed both iGPU (by allocating the system RAM as VRAM) and the CPU.
    This system is a dream come true for sound engineers and composers juggling with Gigabytes worth of sample libraries and massive orchestral pieces. Templates for the notation sofware Steinnberg Dorico with a massive orchestral library such as EastWest Hollywood Orchestra already take up 70-80 GB of RAM.

    The other use case is AI training/tuning where large amounts of memory (e.g.128 GB), most of which can be allocated to the GPU, is critical for large LLMs with many billions of parameters. Even if the GPU itself may not be the fastest, computation will drastically benefit from these large amounts of ultra fast memory that can be used as VRAM. Imagine having a modern RTX card with like 96 GB of VRAM. That alone would probably cost 10.000- 15.000 USD, if not more.
    So people who want to tinker with larger LLMs at home, orchestral composers etc will be drooling over this system.

    Reply

  • Antiphonal

    ottonis said:

    While it may seem that soldered RAM is antithetical to Framework's product philosophy, one need to have i mind, that replaceable RAM moduels would entirely defeat the purpose of a Strix Halo CPU(APU, which NEEDS super-fast and super large memory bandwiidth in order to feed both iGPU (by allocating the system RAM as VRAM) and the CPU.
    This system is a dream come true for sound engineers and composers juggling with Gigabytes worth of sample libraries and massive orchestral pieces. Templates for the notation sofware Steionberg Dorico with a massive orchestral library such as EastWest Hollywood ORchestra alreaty take up of 70-80 GB of RAM.

    The other use case is AI training/tuning where large amount of memory (e,g, 128 GB), most of which can be allocated to the GPU is critical for large LLM with dozens of billions of parameters. Even if the GPU itself may not be the fastest, computation will drastically benefit from these large amounts of memory that can be used as VRAM.
    So people who want to tinker with larger LLMs at home, orchestral composers etc will be drooling over this system.

    Totally agree. It seems strange to me to see a review focusing on FPS in gaming instead of the workloads this machines was meant to deliver.

    I'd like to see tokens/sec with one of the larger models (or even the new GPT-OSS 120B).

    Reply

Read Entire Article