Recent leaks have shed light on AMD’s upcoming graphics cards based on the RDNA 5 architecture. Instead of using the traditional GDDR memory normally found in dedicated GPUs, AMD is expected to adopt LPDDR memory, which is commonly used in laptops.
Why AMD Is Moving to LPDDR
This change could lower production costs and improve availability, since LPDDR is produced in larger volumes and faces fewer supply issues.
The main challenge of using LPDDR is its lower bandwidth compared to GDDR. However, AMD is reportedly working on a solution that combines large L2 cache blocks and Infinity Cache, helping offset this limitation and keeping performance strong in demanding games and graphics applications.
Two main chips have been mentioned in the leaks. The first, known as Alpha Triton 4, would include 24 Compute Units, 10 MB of L2 cache, and a 128-bit memory bus supporting LPDDR5X. It could launch with configurations of 12 GB or even 24 GB of memory.
The second chip, Alpha Triton 3, is a more ambitious design with 48 Compute Units, 20 MB of L2 cache, and a 384-bit LPDDR6 memory controller. This version could offer between 16 GB and 32 GB of memory, aimed at the mid-high performance segment.
Performance Expectations
In terms of improvements, RDNA 5 is expected to bring around a 10% boost in rasterization performance per compute unit compared to RDNA 4. The biggest leap, however, would come in Ray Tracing, where major gains are expected.
According to leaks, the Alpha Triton 4 would perform between an RTX 3060 and RTX 4060, while the Alpha Triton 3 could reach levels similar to an RX 9070 in rasterization, with even stronger ray tracing performance.
This strategy suggests that AMD aims to redefine the entry-level and mid-range GPU market by offering more affordable products without sacrificing memory or key features like Ray Tracing and AI-driven technologies such as FSR scaling. At the same time, AMD will likely continue developing high-end models with GDDR6 or GDDR7 memory for users who demand maximum power.
