A laptop with Snapdragon

According to the video description, the Snapdragon Qualcomm's NPU (Neural Processing Unit) tests included AI image creation in Stable Diffusion and GIMP. As expected, testing was done locally and on-device using the onboard processors, certainly not in the cloud.

The video first shows two computers side by side running a project using Stable Diffusion. The Snapdragon

Qualcomm proved fair in its comparison with Stable Diffusion. As you can see from the screenshot below, the company's engineers have made sure that version 1.5 of the image generator “runs in the most efficient configuration” of the Intel platform, “sharing” processing between the Meteor Lake CPU, NPU and GPU. As appropriate. However, it's worth noting that the Qualcomm device seems to have benefited from a specially optimized version of Stable Diffusion. More specifically, both Stable Diffusion and Fast Stable Diffusion were run on a Snapdragon X Elite PC v1.5. The Intel Core Ultra 7 system also runs Stable Diffusion v1.5 and is configured accordingly GitHub – GIMP AI plugins and OpenVINO backend And User guide for Windows users GitHub.

Qualcomm has also set up both devices to create AI images using GIMP. This time, the prompt is “A majestic lion basking in the golden afternoon sun” with both devices starting up at the same time but finally we see the AI ​​plugin for the device with the Snapdragon Intel's OpenVINO to complete the same task. Qualcomm claims that the 45 TOPS performance processor outperforms its x86 competitor as it can produce 10 times more images at the same time.

However, it is not surprising to see that Qualcomm's platform outperforms Intel's platform in terms of integrated NPU. Qualcomm is talking about 45 TOPS in terms of NPU performance for the Snapdragon

The Meteor Lake family celebrated the commercial launch of Intel's first processors with an integrated NPU, with the company already aiming to significantly increase performance with the next generation – Arrow Lake (desktop) and Lunar Lake (laptop). More specifically, by making architecture-level optimizations for NPUs and GPUs, Intel promises up to three times the performance for AI workloads. At the recent CES, Intel announced that the aforementioned processors expected to be released later this year were in the final stages of development.

It's worth noting that the company is holding a similar demo this week at MWC 2024 in Barcelona, ​​with equally impressive results.