Qualcomm Says Snapdragon 8 Elite Extreme Gen 6 Runs 30B-Parameter AI Models Locally on Phones

Qualcomm says its Snapdragon 8 Elite Extreme Gen 6 can run 30-billion-parameter Mixture-of-Experts AI models directly on phones by loading needed experts from flash storage instead of requiring roughly 32GB of RAM.

• Qualcomm claims its Snapdragon 8 Elite Extreme Gen 6 chipset runs 30-billion-parameter Mixture-of-Experts AI models locally on mobile devices, more than seven times the roughly 4-billion-parameter on-device ceiling cited for current phones.
• The chipset stores MoE experts in flash storage and predicts, loads, runs, and swaps the needed expert for each next token or action rather than holding all 30 billion parameters in RAM.
• Chris Patrick says a comparable dense model would require about 32GB of phone RAM, while current top-end phones typically offer 12GB to 16GB.