Navi 48 and 44 Specs Revealed: AMD’s Mid-Range Revolution
During its Financial Analyst Day event in June 2022, AMD updated its GPU architecture roadmap for the RDNA/CDNA series, showcasing the RDNA 4 architecture GPUs as the Navi 4x series chips. These will be manufactured using an even more advanced process and are scheduled for release in 2024. Rumors suggest that the RDNA 4 architecture will feature just two chip models, namely Navi 48 and Navi 44.
Recent disclosures from internet users have revealed preliminary specifications for the Navi 48 and Navi 44.
The Navi 48 is set to be equipped with 32 Workgroup Processors (WGP), totaling 64 Compute Units (CU), and will feature a 256-bit memory interface with a bandwidth of 693GB/s, achieving an effective bandwidth of 2770GB/s, and a GPU die size of 240mm². By comparison, the current Radeon RX 7800 XT, which is equipped with a Navi 32, measures 346mm², has 60 CUs, a similar 256-bit memory interface, a memory bandwidth of 624.1GB/s, and an effective bandwidth of 2708GB/s.
Besides the more compact size, the original specifications of both are relatively similar. However, the cache size and frequency of the Navi 48, which significantly influences GPU performance, remain unclear. It is claimed that the Navi 48 will deliver 50 TFLOPs of computational performance at a power draw of 215W, whereas the Radeon RX 7800 XT offers 37 TFLOPs at 263W, and Nvidia’s Geforce RTX 4070 Ti achieves 40 TFLOPs at 285W.
The Navi 44, targeting the mid to low-end market segment, features only 16 WGPs, amounting to 32 CUs, with a 128-bit memory interface, a memory bandwidth of 288GB/s, an effective bandwidth of 515GB/s, and a GPU die size of 130mm². For context, the current Radeon RX 7600, equipped with Navi 33, measures 204mm² and also features 32 CUs, a 128-bit memory interface, and a bandwidth of 288GB/s. This indicates that the most significant advantage of the Navi 44 is its smaller chip size and potentially lower power consumption.
It is understood that the Navi 4x series chips will adopt a 4nm manufacturing process to further enhance energy efficiency.
