Snapdragon

The Snapdragon
Using Stable Diffusion 1.5, a famous machine learning model initially published in 2022, the Californian giant compared the processor with an Intel Core Ultra 7 155H, highlighting the power gap between the two chips: Snapdragon X Elite far surpasses the competition in image generation, with surprising performance.

Snapdragon X Elite: some numbers

Snapdragon X Elite is the new ARM processor Qualcomm based on a 4 nanometer production process that can count on 12 Cores and 12 Threads, with a peak frequency of up to 4.3 GHz and a TDP of 45 W, while on the graphics side it shows off the new Qualcomm Snapdragon X Elite Adreno GPU.
Just like Intel's rival, too Snapdragon X Elite can count on an integrated NPUcalled Hexagon, dedicated to the local processing of tasks assigned to artificial intelligence.

Stable Diffusion 1.5 should make extensive use of the NPUs integrated into both processors, but in this specific case the Qualcomm chip outclassed the Intel rival: in the first prompt used in the tests, which was about “a berry bush“, the Snapdragon SoC completed the request in 7.25 secondsgenerating images three times faster than Intel Core Ultra 7 155H which took 22.26 seconds.

In the test relating to “a majestic lion tanning in the golden afternoon sun“, Snapdragon X Elite has taken off, generating the image 10 times faster than the competitor: The Qualcomm chip took about 1 second, while the Intel Core Ultra 7 155H completed the request in a full 12 seconds.

Not just AI

In the latest benchmarks that emerged online, Snapdragon the processor reached 2574 points in single-core And 12562 points in multi-core.
From the tests it can be deduced that Snapdragon of Intel, however remaining above that of AMD, which however has 8 cores instead of 12.
In the previous official tests performed on multi-core performance, Qualcomm had declared important performances: put to the test with Geekbench 6, the Californian company's chip would have demonstrated up to 21% more powerful than Apple M3not to mention that it can operate on two power limits of 23W and 80W respectively depending on the scenario.

Only time will tell if these numbers will truly be confirmed. However, the success of the SoC could depend a lot on the optimization of the reference software. All that remains is to wait for the first road tests to make a definitive judgement.