CATEGORY_INDEX // 16 ENTRIES

Hexagon NPU

Infographic on Qualcomm's three engines before Maui, all memory-focused: Oryon Flex Cache merges two 12MB L2 pools into a 24MB shared pool (+40% working set hit rate, -35% DRAM traffic); Adreno Matrix Cores paired with an 18MB high-performance memory pool unchanged since prior generation, showing performance ceiling is memory capacity not clocks; Hexagon NPU with 50% larger shared memory activates only 3B of 50B MoE parameters per token (+45% inference efficiency, -40% energy per token). Sidebar notes LPDDR crisis (bandwidth wall, power wall, memory tax) and SRAM-as-DRAM-hedge vs 5GHz marketing. Tagline: Every one of them is about memory. m1k.tech, redefining performance one memory decision at a time. Summit in four days
CHIPS & AI · 09/18/2026

Qualcomm Showed All Three Engines Before Maui

Qualcomm has unveiled its next flagship mobile chip architecture, highlighting three main components: the Oryon CPU reaching 5GHz, Adreno GPU incorporating AI-dedicated units, and…

22 MIN
READ →
Nvidia H20 Export Ban: CUDA Lock-In vs Qualcomm Edge — m1k.tech article thumbnail featuring Nvidia and Qualcomm Snapdragon chips
CHIPS & AI · 07/21/2025

The H20 Reversal Was Never About the Chip

Jensen Huang's presence in Beijing coincided with the U.S. Commerce Department's lifting of a chip ban, allowing Nvidia and AMD to resume shipments to…

9 MIN
READ →
// SUBSCRIBE_FEED

Signal over noise.

One maker's tech digest, weekly. Chips, AI, gaming, EVs — no spam.