AMD’s next-gen RDNA 3 flagship will be twice as fast as its predecessor, the Radeon RX 6900 XT, much like its archrival-to-be, the RTX 4090. We’ve been hearing a lot about the next generation of graphics cards from Team Red and Green, most notably the performance of the enthusiast offerings. (www.dr4health.com) The other day we reported that the RTX 4090/4090 Ti is looking over 2x as fast as the RTX 3090 in both ray-traced games as well as synthetic benchmarks.
The fully enabled AD102 die (when overclocked?) reportedly nets more than 160 FPS in Control at 4K with all the settings including ray-tracing set to max and DLSS enabled. The current fastest GPU, the RTX 3090 Ti manages just around 80 FPS in the same scenario. If popular tipster Kimi is to be believed, then this isn’t even the final performance of these SKUs. With the boost clocks, TDP, and L2 cache optimized, the frame rates should increase by another 5-10%, at the very least.
AMD’s Radeon RX 7900 XT is going to be no pushover either. Greymon55 believes that the chiplet-based GPU will be twice as fast as the 6900 XT but the retail variant might even up being a little faster or slower as the TGP is still being optimized. Furthermore, the GCD (Graphics Die) on the RX 7900 XT will have an area of 350mm^2 which makes it slightly smaller than the AD103, the die powering the RTX 4080. Or so Kimi claims.
This discrepancy in die sizes despite leveraging the same process nodes arises due to two reasons. Firstly, NVIDIA is using a fully monolithic design with all the GPU core, VRAM chips, and memory controllers on the same die. AMD, on the other hand, has offloaded the memory controllers and L3 cache to the MCDs (Memory Complex Dies), thereby improving yields and complexity. Secondly, NVIDIA’s fixed-function ray-tracing and AI hardware, namely the RTCores and the Tensor Cores take up quite a bit of die space, most notably so when you’ve got over 700 of them on the fully enabled (AD102) die.