Historically, AMD has chosen to prioritize energy efficiency over peak performance by selecting standard GDDR6 over the faster GDDR6X variant. This approach is complemented by the use of a substantial L3 cache, known as the Infinity Cache, in its recent GPU architectures. This design choice has been effective enough for AMD to compete in bandwidth performance, even as competitors like Nvidia have adopted larger L2 caches in their latest graphics series.
There are also speculations that AMD may further enhance its Infinity Cache in the RDNA 4 GPUs instead of transitioning to GDDR7, potentially adding up to 192MB of Infinity Cache across a 384-bit interface. This strategy could allow AMD to maintain competitive performance levels against upcoming GPU models from Nvidia and Intel, both expected to release new technologies this fall. AMD's consistent use of GDDR6 is likely influenced by a balance of cost, performance needs, and the physical constraints of PCB design. While the upcoming RDNA 4 series might stick with GDDR6, high-end models could potentially adopt GDDR7 to match or exceed competitors' offerings in terms of bandwidth.
As with any pre-release information, these details are subject to change as more concrete specifications become available closer to the official unveiling of AMD's next-generation GPUs.