AMD unveils RDNA 3 GPU
AMD yesterday introduced new graphics cards based on the next-generation, high-performance, energy-efficient RDNA 3 architecture
RDNA3 is first and foremost a multi-chip architecture that allows AMD to boast the “first GPU chiplet for gaming”. An RDNA 3 GPU is thus characterized by a Graphic Compute Die (300 mm²) engraved in the 5nm process by TSMC and associated with several Memory Cache Die (MCD) engraved in the 6nm process (and whose area is 37 mm² each).
AMD is moving forward with a total of 58 billion transistors for the RDNA 3 GPU, which is significantly less than the 76.3 billion transistors on Nvidia’s Ada Lovelace GPU.
Inevitably, a multi-chip architecture requires ultra-fast chip interconnects. This is the case here with 5.3 TB/s of throughput between the various chiplet components, a new record in this area. This is almost 3 times the bandwidth of RDNA 2.
Chiplet performance will reach dizzying performances. AMD in fact advertises a graphics computing power of 61 tflops, that is, almost 3 times the performance of the RDNA 2 (23 tflops). All this is certainly very promising from a technological point of view, but at what cost in the end?
Newer Articles
- What is Ergonomics at work?
- PSVR 2: PlayStation wants to captivate Gamers with its Games
- Samsung Galaxy S23 to exclusively use Qualcomm’s Snapdragon chips
