According to current speculations, AMD RDNA3’s flagship GPU Navi 31 could attain FP32 performance of 92 TFLOPS, up from the previously reported 75 TFLOPS. That’s four times the performance of the top-tier Navi 21 GPU (RX 6900XT) chip in the current RDNA2.
If AMD keeps the 60 WorkGroup Processor (WGP) and 15360 stream processor arrangement, a core speed of 2995MHz is necessary to accomplish 92 TFLOPS of FP32 single-precision computing performance, but the exact alterations are still unknown. However, AMD may develop a new 32bit instruction matrix to increase floating-point performance, akin to NVIDIA’s Ada architecture.
However, Greymon55 noted that the 92 TFLOPS figure was for the rumored 15360 Stream Processor configuration. He’s recently released an update, claiming that there are now 12288 Stream Processors in the system. This means the GPU with a 3 GHz frequency would have 73 TFLOPS less than the 92 he claimed.
AMD may also preserve Compute Units for its RDNA3 (GFX11) architecture, contrary to prior speculations. The AMD Radeon RX 7000 series should be available by the end of the year. AMD’s Navi 31 GPU should be the first to be introduced, with the RX 7900 XT, a new consumer-grade flagship graphics card, as a probable model name.
Why I am excited for AMD to bring out RDNA3?
Over the last few years, AMD has been largely absent from the flagship/high-end GPU market. With the rDNA architecture and the Radeon RX 5000 and RX 6000 series, AMD’s exile from the high-end GPU market came to an end.
In terms of raw rasterization performance, the RX 6900XT comfortably outperforms NVIDIA’s fastest consumer GPU, the RTX 3090. Despite the lack of ray tracing, these RDNA 2 GPUs are still a superb pick.
The last time AMD offered a rival to NVIDIA was in 2011 when the HD 7970 was pitted against the GTX 580 and GTX 680. We should see a significant improvement in price-per-performance from both businesses if AMD makes another competitive comeback with RDNA 3 to amp up the battle.