AMD’s next-gen Radeon RX 7000 GPUs will be the first in the PC industry to leverage a chiplet architecture, something Apple tried to do with its M1 Ultra SoC (but failed horribly). The RDNA 3 family will be based on two process nodes, namely N5 and N6. The top-end Navi 31 and Navi 32 GPUs will leverage a dual-die design with core counts of over 15,000. The midrange Navi 33 die will be fabbed on TSMC’s 6nm process, and still offer RX 6900 XT levels of performance on a budget.
The Radeon RX 7700 XT will be powered by the Navi 33 core, and compete with the likes of the GeForce RTX 4070. It will reportedly come with an 8GB VRAM buffer paired with a 128-bit bus along with 128MB of L3 “Infinity” cache. According to Tom from MLID, it’ll be faster than the Radeon RX 6900 XT at 1080p but fall behind at 4K due to the higher external bandwidth requirement. In ray-traced titles, however, you can expect markedly better performance than all the existing Radeon RX 6000 GPUs.
In terms of power consumption, the original TGP target was 200W but now has been increased to 230W. Interestingly, the RX 7700 XT will be equipped with an 8-lane PCIe Gen 5 bus which is more than sufficient considering the high bandwidth of Gen 4/Gen 5. However, users with older PCIe Gen 3 boards may run into bottlenecks.
The second part of the report claims that AMD’s Radeon RX 7000 GPUs will absolutely obliterate rival NVIDIA’s GeForce RTX 40 series lineup in power efficiency. This is rather far-fetched considering that both families will be fabbed on TSMC’s N5 process node. The GeForce RTX 30xx offerings are roughly as efficient as the RDNA 2 parts despite leveraging a much older node (Samsung 8nm LPP vs TSMC N7). Therefore, I’m a little skeptical about this particular claim.