GDDR7 memory density and capacity evolution

The graphics memory market is transitioning toward higher-density GDDR7 modules to meet increasing VRAM demands from 4K gaming, ray tracing, and generative AI workloads. Micron has reportedly begun phasing out 2GB GDDR7 chips, marking several 16Gb/2GB models as ‘End of Life’ in favor of 3GB (24Gb) modules. This shift allows for increased video RAM capacity without increasing chip counts or complicating PCB designs.

Industry leaks suggest that 4GB and 6GB GDDR7 chips may arrive within the next two years, which could potentially double the capacity of graphics cards. For example, a 64-bit bus could theoretically support up to 24GB of capacity using 6GB chips.

In tandem with these hardware shifts, software developments indicate upcoming architecture support. Recent updates to the open-source AMDGPU Linux driver include explicit support for GDDR7, which is widely viewed as preparation for AMD’s rumored RDNA 5 architecture.

New technical details regarding the GDDR7 standard reveal the use of PAM3 (Pulse-Amplitude Modulation) signaling, supported by JEDEC and manufacturers including Samsung, Micron, AMD, and NVIDIA. This signaling method manages three bits per two cycles, providing approximately 25% better energy efficiency than previous standards and allowing frequencies up to 36 Gbps. Such speeds could enable bandwidth to exceed one terabyte per second.

Regarding AMD’s RDNA 5 desktop architecture, recent leaks suggest clock speeds may exceed 4 GHz, a significant increase from earlier speculations of the 3 GHz range. The combination of high-speed GDDR7 and these increased clock speeds is expected to represent a major generational leap for the Radeon family.