Recent analysis of leaked specifications suggests that the upcoming PlayStation 6 console could feature a custom AMD GPU with a theoretical performance peak of up to 80 TFLOPS. These estimates are based on a custom RDNA 5 architecture manufactured using TSMC's 3nm process, featuring 54 compute units and a maximum clock speed of 3.6GHz.
The hardware configuration reportedly includes a 160-bit GDDR7 memory interface with speeds reaching 32Gbps, all operating within a 160W thermal design power limit. While raw rasterization performance is estimated between 34 and 40 TFLOPS, comparable to the Radeon RX 9070 series, the console's performance profile shifts significantly when utilizing advanced features.
The integration of Dual Issue instruction support and the PSSR 3 upscaling technology is expected to boost FP32 theoretical performance from 40 TFLOPS to a range of 68–80 TFLOPS. In scenarios involving full path tracing, this could result in performance gains of 70% to 100% over the RX 9070 XT, despite the lower number of compute units.
When compared to NVIDIA's current lineup, the PS6 GPU's theoretical peak sits between the GeForce RTX 5080 at 56.3 TFLOPS and the RTX 5090 at 104 TFLOPS. Analysts further suggest that PSSR 3 could enhance effective graphical output by an additional 35% to 50%, though these figures remain based on unconfirmed leaks.
