SK Hynix is dropping GDDR7 memory that hits 48 gigabits per second per pin at the 2026 ISSCC conference, which absolutely demolishes the earlier guesses that topped out around 32 to 37 gigabits. The 24-gigabit-density chips pump out 192 gigabytes per second of bandwidth per chip, leaving the current 28-gigabit solutions in the dust for GPU rendering and AI training workloads.
The company is also rolling out LPDDR6 modules clocked at 14.4 gigabits per second, way faster than the 9.6 gigabit speeds from LPDDR5. Mobile devices and edge computing boxes get better memory access for AI stuff without destroying battery life.
If these specs actually hold up when the products ship, graphics cards and laptops could see serious performance bumps across rendering tasks and generative AI applications that keep hitting memory bandwidth walls.
The company is also rolling out LPDDR6 modules clocked at 14.4 gigabits per second, way faster than the 9.6 gigabit speeds from LPDDR5. Mobile devices and edge computing boxes get better memory access for AI stuff without destroying battery life.
If these specs actually hold up when the products ship, graphics cards and laptops could see serious performance bumps across rendering tasks and generative AI applications that keep hitting memory bandwidth walls.
