The high processing power of our smartphones is actually not noticed by most of us in daily use. However, chipsets found in flagship phones can compete with today’s computers.
The biggest proof of this sheer power was presented by Qualcomm. The company ran the image generator artificial intelligence Stable Diffusion with the new chipset Snapdragon 8 Gen 2.
First, let’s explain why we’re surprised why Stable Diffusion works on mobile:
Stable Diffusion is a visualization tool that is trained and used on extremely high-performance hardware, often through cloud-based systems.
In order to train this artificial intelligence and create images using the data generated by the training, you need to have the highest level graphics processors. At least if you don’t want to wait for hours to get the results.
If you have a very powerful computer and you have trained artificial intelligence, you can now start to benefit from its benefits without the need for an internet connection.
How did Qualcomm achieve this performance?
Qualcomm has announced that it has made major optimization tweaks on the smartphone to run Stable Diffusion on mobile. The company’s artificial intelligence research team performed operations on the device, the application used, algorithms, software and even hardware that we would not see on a regular phone.
Of course, some of these settings were also made on Stable Diffusion. The team switched from the floating point format FP-32 data type used for Stable Diffusion and deep learning in general to the lower precision INT8 type. It also used the AI Model Efficiency Toolkit (Artificial Intelligence Model Efficiency Toolkit) technology that it has just released.
With all the optimizations and adjustments made, Stable Diffusion has become able to produce 512×512 pixel resolution photos in 15 seconds on a smartphone. This period is also very close to Stable Diffusion systems integrated into cloud systems.
So what will Qualcomm use this success for?
Emphasizing that running Stable Diffusion on the phone is just the beginning, the company stated that there will be many advantages such as security, efficiency and cost with the giant artificial intelligence technologies working on mobile devices. In addition, the optimization studies carried out became an example in the field.
25868
