The Snapdragon
Qualcomm’s SoC looks pretty tasty with AI acceleration as a new YouTube video (hats off Wccftech) showing off the Snapdragon X Elite is all you can do. In it, the Snapdragon
The clip shows the two CPUs running Stable Diffusion 1.5 (in the GIMP image editor) to test the capabilities of their respective NPUs (or Neural Processing Units) for accelerating AI-driven image generation.
When challenged to create an image of a “bundle of berries,” the Snapdragon . The Snapdragon is indeed more than 3x faster here.
The next test was to generate a photo of a ‘majestic lion basking in the golden afternoon sun’ and the Snapdragon
Qualcomm’s SoC produced an image approximately every second, while the Core Ultra 7 only managed one image in 12 seconds. You can check out the comparison in the video below.
Analysis: Snap happy
This seems like a surprising win for the Snapdragon X Elite. However, with internal testing we should always be somewhat skeptical, as individual (and possibly handpicked) tasks are one thing, while a set of independent benchmarks performed by a third party is quite another.
Still, the margin of victory here is so large that any way you look at it, it’s a real eye-opener to the power of the NPU in the Snapdragon silicon compared to the Intel Core Ultra 7 laptop processor.
Keep in mind, though, that this is specifically focused on AI acceleration (and the NPU), and there are many other performance metrics outside of that narrow (but admittedly rapidly broadening) field.
The Snapdragon (Like using Windows on ARM, and not being able to run popular software – or at least not being able to run traditional x86 apps natively at the full extent of their potential performance).
However, the benefits include the Snapdragon X Elite having very good battery life (long lasting for ‘days’, we’re told). And if the leaks, and indeed Qualcomm’s own pre-release benchmark claims, are correct, this is a CPU that could be a real competitor to Apple’s (also ARM-based) M3 silicon for raw performance.
Laptops powered by a Snapdragon