A leak in certain ROCm code, found by Twitter user Kepler L2, revealed the unified shader (stream processor) counts of AMD’s future second- and third-largest GPUs based on the RDNA3 graphics architecture. “Navi 32,” which has a compute unit count of 60, and “Navi 33,” which has 32 compute units, are both mentioned in the “performance.hpp” file. We are aware that an RDNA3 compute unit still equates to 64 stream processors from the “Navi 31” specs (although with significant IPC uplifts over the RDNA2 stream processor due to dual-instruction issue-rate).

With 60 compute units, the “Navi 32” silicon would have 3,840 stream processors, which is a 50% improvement over the 2,560 stream processors of the “Navi 22” silicon, which powered graphics cards like the Radeon RX 6750 XT. The “Navi 33” has 2,048 stream processors in total, which is identical to the “Navi 23” powering the RX 6650 XT in terms of numbers. Aside from the dual-issue stream processors, the new RDNA3 compute unit differs significantly from RDNA2 in that it also features two AI accelerators for matrix-multiplication and second-generation Ray Accelerators.

At this time, no additional details on the “Navi 32” and “Navi 33” are known. With a 5 nm GCD that contains these 60 RDNA3 compute units and either three or four 6 nm MCDs, depending on whether AMD decides to give it a 192-bit or 256-bit wide GDDR6 memory interface, the “Navi 32” could have a similar design to the “Navi 31” if AMD is sticking with the chiplet design. If the chiplet architecture even extends to smaller GPUs like the “Navi 33,” it remains to be seen. Also unknown is the release date for the remaining Radeon RX 7000-series graphics card models from AMD. The RX 7900 series will make its debut in December 2022.