AMD RDNA3 architecture family of large core codenamed Navi 31, is said to be divided into two 5nm GCD modules, one 6nm MCD module, respectively, compute nodes, input and output nodes, integrated up to 15360 stream processors, 512MB infinite cache, with 256-bit GDDR6.
This is undoubtedly the successfully streamed version, which unsurprisingly will bear the name RX 7900 series.
Further down the line are the Navi 32, also with dual-core processors, 1024 stream processors, 384MB infinite cache, and 256-bit GDDR6 memory.
Further down the line is the Navi 33, a single-core with 5120 stream processors, 256 MB infinite cache, and 128-bit GDDR6 memory.
And for the lower end of the market, the Navi 23 and Navi 24 cores are to be reused and the manufacturing process updated from 7nm to 6nm, with the same specifications and a lower price.
AMD is already working on the production of RDNA 3, the graphics architecture that will power the next generation of Radeon graphics cards.
The AMD Navi 31, which is based on the RDNA 3 architecture, has been taped out
The hint came directly from Greymon55, a well-known leaker, who reported that the Navi 31 GPU has been “taped out.”
The term “tape out” means that the design of the GPU has been completed, an important step towards the start of mass production of the chip. Once the first wafers are produced, AMD will begin testing the chip, a process that will take some time to properly optimize, especially since this is the first time a mass-market MCM design has been used.
The design itself features two side-by-side molds, though it’s not known if the two molds are identical or if there’s a difference in manufacturing. Each has three shader engines with 120 compute engines. So the entire chip will have 15360 FP32 shader units, although this could be less on the commercial Radeon models.
For next-generation Radeon graphics, Navi 31 would use TSMC’s N5 and N6 process technology, although this is not confirmed at this time.
Navi 31 silicon and all other GPUs are expected to bring the RX 7000 series to life by the end of next year. At that time, Nvidia will launch the corresponding RTX 40 series. We’ll see how much better and more competitive the next Radeon can be against GeForce now that it uses an MCM architecture.