Companies in the technology sector tend to plan and develop for the long term, i.e AMD is already working RDNA 4a future architecture that will likely be used in the generation of Radeon RX 8000 graphics. The first information available on RDNA 4, which would be better taken with a grain of salt given the data we are in, suggests that it would provide an improvement of 50 to 60 percent higher power per watt than its predecessor RDNA 3, which is implemented in the RX 7000.

According to YouTube channel RedGamingTech, sources of information circulating around RDNA 4, this will bring some changes, such as the presence of a single graphics compute cube (GCD), which would be split into one or more GCXs. Navi 41, the future flagship model and top of the range (at least initially), could consist of 64 or 128 logical arithmetic units per compute unit (CU), in addition to reportedly having 48 CUs and 3 shader engines (SE) for each GCX.

Memory Cache Dice (MCD) based chips aim to stay while The memory type would be GDDR7 (at least for the top-of-the-line models), PCIe 5 interface, and use TSMC 3 and 4 nanometer manufacturing processes. Cache size and infinite cache are two queues that are likely to increase in capacity, but the filter did not provide data on this. On the other hand, the second version of the Wave Matrix Multiple Accumulation (WMMA) instructions should aim to speed up the performance on the CU for certain instructions by up to two times.



As we have already said, one of the alleged excellent improvements is 50 to 60 percent performance improvement per watt versus RDNA 3 and RDNA 4 it would be able to exceed 3.5 GHz. Navi 41, which would become the main graphics for future architecture use, would include a total of 48 CUs per GCX, and there would be a total of three GCXs per GCD, for a total of 144 CUs. versus the 96 CU RX 7900 XTX. As for her part, the number of shaders in Navi 41 would be 18,432.

In terms of VRAM, AMD would consider two configurations for the Navi 41, one with four MCDs with 32GB of GDDR7 memory and the other with six MCDs and 24 and 48 gigabytes of GDDR7 variants. Navi 42 would like to be a more “modest” graphics with two GCXs which would probably add 96 CUs, 12228 shaders and 4 MCDs in total. Finally, the Navi 43 would have a single GCX for a total of 48 CUs and two MCDs with GDDR7 memory.

It’s still too early to confirm, but with the bittersweet launch of the RX 7900 XT and RX 7900 XTX graphics, many will surely see all this information as more smoke than anything else. Let’s hope AMD and Intel get their way because NVIDIA is starting to feel like they’re winning just by inertia.