Nvidia and AMD’s next-generation flagship GPU leaked
Recently, Twitter user @Greymon55 revealed more information about Nvidia’s GeForce RTX 4090 based on the Ada Lovelace architecture, and AMD’s Radeon RX 7900 XT based on the RDNA 3 architecture. These two graphics cards are the next-generation flagship products, and both will support the new performance benchmarks of Nvidia and AMD.
The AD102 GPU with Ada Lovelace architecture is manufactured using TSMC’s 5nm process and will be equipped with 144 sets of SMs, with 18432 CUDA cores, with GDDR6X video memory, a bit width of 384 bits, a capacity of 24GB, and a rate of 21 Gbps. Its core frequency is between 2.3 GHz and 2.5 GHz, and the performance of FP32 is between 85 TFLOPs and 92 TFLOPs, which is more than doubled on the basis of GeForce RTX 3090 (36 TFLOP).

The Navi 31 of the RDNA 3 architecture will consist of two GCDs and one MCD, using MCM multi-chip packaging, with a total of 240 CUs, that is 15,360 stream processors, and an Infinity Cache of 512MB. The video memory bit width is still maintained at 256 bits, still GDDR6, the rate is 18 Gbps, and the bandwidth is 576 GB/s. Each GCD in the Navi 31 core will have 30 WGPs, and each WGP will have 256 stream processors, which means that each computing module can provide 7680 stream processors, and a Navi 31 core will have a scale of 15,360 stream processors. The core frequency of Navi 31 is between 2.4 GHz~2.5 GHz, and the performance of FP32 is around 75 TFLOPs. As Infinity Cache will also switch to a 3D stack design, GPUs based on the RDNA 3 architecture are likely to adopt this technology. It is rumored that Infinity Cache will be added to the MCD chiplet, similar to the principle of using 3D V-Cache on the Zen 3 architecture. Navi 31’s GCD uses TSMC’s 5nm process, while MCD uses a 6nm process.
It is foreseeable that in about a year, the new graphics performance battle between Nvidia and AMD will be on the verge.