Framework 13 Gets ARM—But Should You Actually

Look, I'm obsessed with the idea of ARM laptops. The promise of Apple Silicon performance and battery life in a machine you can actually repair and upgrade? That's the dream. So when MetaComputing dropped an ARM mainboard for the Framework 13—literally the most modular laptop on earth—I was ready to get hyped.

Then tech journalist Jeff Geerling actually tested it, and uh... we need to talk about expectations versus reality.

The Swap That Shouldn't Work This Well

First, the genuinely cool part: Geerling hot-swapped his Framework 13 from AMD to ARM in ten minutes. Ten. Minutes. That's remarkable when you think about it—this laptop has now run x86, RISC-V, and ARM processors. It's basically the Rosetta Stone of computing architectures.

The MetaComputing board uses a 6P616 chip (or something like that—even Geerling couldn't remember the exact model name, which... tells you something about how memorable this chip is). Twelve ARM cores, four USB-C ports that all support power input and HDMI output, running Ubuntu 25.04 with Linux kernel 6.6.

Booting up? Smooth. General use? Absolutely fine. As Geerling noted: "Everything on here works like a typical computer would. It has 12 cores, so it's not going to be a slouch doing almost anything that it needs to do."

Browsing, YouTube, productivity apps—it all just works. Which is genuinely impressive for beta firmware on a brand new ARM board. The Framework ecosystem is delivering on its modular promise.

But that's where the good news plateaus.

When The Benchmarks Get Honest

Geekbench scores came in... fine. Right between other 6P1 chip systems like the Minisforum MSR1 and Radxa Orion O6. Nothing surprising, nothing alarming.

Then Geerling ran HPL—a floating-point double precision task that hammers memory bandwidth. And something weird happened. The board underperformed compared to other systems using the same chip, with worse efficiency to boot. "I don't know exactly what's happening here," Geerling admitted. "Something's funny there."

Maybe it's a BIOS setting. Maybe it's the memory configuration. But the fact that a reviewer as thorough as Geerling can't pinpoint the issue suggests this hardware is still finding its footing.

The GPU situation is even messier. The Mali G720 Immortalis GPU delivered scores comparable to a several-generations-old Apple A14 chip. Which would be fine for productivity work, except—

Gaming: Please Don't

Geerling tried running Steam games through FEX emulation (translating x86 Windows games to ARM Linux). Horizon Chase Turbo, a lightweight older racing game, ran but barely. "It's not quite a slideshow here, but I wouldn't call this playable," he said.

Portal 2 stuttered so badly it was "a challenge to even enter the test chamber." Abduction quit after the splash screen. Doom Eternal crashed the entire laptop when it ran out of the 16GB of shared GPU/CPU memory.

Now, to be fair: this board isn't designed for gaming. But the gaming stress test reveals something important about where ARM on Linux still struggles—driver support, emulation overhead, and memory management under load.

Windows on ARM? That's a whole different nightmare.

Windows: Technically Possible, Practically Painful

Getting Windows 11 to install required: upgrading the BIOS, dealing with NTFS driver bugs, disabling four CPU cores (a setting that wouldn't stick), switching from performance mode to "work mode" (which Geerling correctly notes is ironic if you're trying to do actual work), and even then the install failed midway through.

MetaComputing will eventually support Windows properly. But right now, this is a Linux-first board, and anyone thinking about dual-booting should recalibrate their expectations.

The AI PC Thing Nobody Asked About

Why is this called an "AI PC"? Because it has a neural processing unit rated for 30 TOPS int8. Which sounds impressive until you realize you can't just install AI tools and go—you need specific tweaks and configurations to actually use the NPU.

MetaComputing doesn't even have documentation yet. Radxa's wiki has some guides for object detection, but that's about it. The "AI PC" label feels like marketing reaching for relevance rather than describing a meaningful capability.

The Problem That Kills Everything

Here's where the math gets brutal. This ARM board idles at 7-8 watts with the screen off—MetaComputing's best-in-class for 6P1 chips. Turn on the display and you're at 10 watts.

Meanwhile, the MacBook Neo idles at under 1 watt. With the screen on: 3 watts. The AMD Ryzen 5 340 Framework mainboard? 2.7 watts idle, 5 watts with display.

For a desktop, 10-15 watts idle is whatever. For a laptop—a device whose entire value proposition depends on battery life—it's disqualifying.

And then there's price. The ARM board costs around $800 with memory. The MacBook Neo is way cheaper and way faster for bursty workloads. The AMD Framework mainboard costs about the same but is faster across the board, more efficient, and compatible with more Linux software.

As Geerling put it: "If you want a good value ARM laptop, get a Neo."

Who Is This Actually For?

The Framework ARM mainboard occupies a weird position. It's better than the RISC-V board that burned 25 watts doing nothing, but that was explicitly a developer-only product. This ARM board could work for general users—if the price drops, if the power draw improves, if BIOS updates unlock better performance.

Right now? It's for people who specifically want ARM on Framework for development or testing purposes, and who understand they're paying a premium for modularity rather than performance or value.

Which is fine! That's a legitimate use case. But it's not the mass-market ARM breakthrough some people were hoping for.

The good news: this is beta 2 firmware. There's room for improvement. The bad news: even with perfect optimization, this board is competing against Apple's frighteningly efficient silicon and AMD's mature x86 ecosystem.

The Framework 13 can run ARM now. Whether you should run ARM on it depends entirely on whether you value experimentation and modularity over, well, everything else a laptop is supposed to do well.

—Zara Chen