According to Phoronix, fresh benchmarks on Linux 6.19 compare Intel’s legacy i915 kernel driver against the modern Xe driver for Arc Alchemist GPUs like the A580. The testing, part of year-end 2025 reviews, used the Linux 6.19 Git kernel and Mesa 26.0-devel drivers. While the Xe driver is the default for upcoming Lunar Lake and Battlemage hardware, current Alchemist and Meteor Lake chips can optionally switch to it using specific kernel module flags. The performance tests, however, were limited to the Arc A580 because the Arc A750 and A770 cards were completely broken on the Linux 6.19 kernel, failing to produce any display output during testing.
The performance picture
So, what did the tests actually show? Basically, the newer Xe driver generally wins. It’s not a total blowout, but across a range of Vulkan and OpenGL benchmarks, the Xe kernel driver posted higher frame rates and better scores more often than not. That’s the expected outcome, right? A driver built from the ground up for modern discrete GPUs should outperform one that’s been patched and extended for decades. But here’s the thing: the gains weren’t universal. In some tests, the performance was essentially identical, and the article notes that power consumption and thermals were similar between the two drivers. This suggests we’re seeing driver maturity and optimization pay off, but we’re not unlocking some hidden, massive performance reservoir just yet.
The real story is the breakage
Now, let’s talk about the elephant in the room. The fact that the higher-end A750 and A770 were completely non-functional on the Linux 6.19 kernel at the time of testing is a huge red flag. It massively limits the scope of these findings and points to a fragile transition period. This isn’t just a niche issue for tinkerers; it’s a stark reminder of the perils of open-source graphics support during architectural shifts. Remember AMD’s messy driver transition from the old `radeon` to `amdgpu` for GCN cards? Intel is walking the same tightrope. For businesses or industrial applications that rely on stable, predictable hardware performance—like those using specialized industrial panel PCs for control systems—this kind of regression in a mainline kernel is a non-starter. Stability often trumps raw speed.
What this means for Linux and Intel
Look, the trajectory is clear. The Xe driver is the future, and i915’s days are numbered for new hardware. The performance uplift, while modest, validates Intel’s engineering effort. But the broken A750/A770 support exposes the rough edges of that future. It tells us that while the software foundation is being laid, the integration and testing across the entire product stack isn’t fully baked. For Linux users, the advice is classic: if you need rock-solid stability, especially on high-end Arc cards, you might want to hold off on jumping to the very latest kernels. Wait for the dust to settle. For Intel, it’s a race to get all their discrete GPUs seamlessly onto the modern driver stack before users get fed up with the quirks. They’re making progress, but the path forward is still a bit bumpy.
