Foreign media outlet GameGPU analyzed previously leaked specifications, estimating that the PS6 console's theoretical performance could reach up to 80 TFLOPS, a figure potentially surpassing NVIDIA's RTX 5080, although these specifications have not yet been officially confirmed.
The leaked parameters used in this analysis indicate that this AMD custom graphics chip may utilize a customized RDNA 5 architecture and TSMC's 3nm process, featuring 54 compute units with a maximum frequency of 3.6GHz. Its memory interface is a 160-bit GDDR7 with speeds up to 32Gbps, and its thermal design power (TDP) is limited to 160W.
Based on the above parameters, the PS6 GPU's raw rasterization performance is estimated to be around 34 to 40 TFLOPS, roughly comparable to the Radeon RX 9070 and RX 9070 XT. For comparison, the RX 9070 XT adopts the RDNA 4 architecture, features 64 compute units and a 256-bit memory interface, has a power consumption of approximately 304W, and operates at frequencies around 2.9 to 3.0GHz.
The analysis suggests that although the PS6 graphics chip has fewer compute units, its higher frequency and architectural optimizations may narrow the performance gap. In sustained high-load rasterization scenarios without AI algorithms, the RX 9070 XT is expected to remain about 5% to 10% faster, with its wider memory interface being one of its advantages.
However, if Dual Issue and PSSR 3 are taken into account, the calculated results change. The analysis states that RDNA 5's dedicated architecture units' support for Dual Issue instruction parallel execution can increase FP32 theoretical computing power from about 40 TFLOPS to 68–80 TFLOPS; in games with full path tracing, its performance might exceed that of the RX 9070 XT by 70% to 100%.
In terms of theoretical peak performance, the GeForce RTX 5080 stands at 56.3 TFLOPS, and the GeForce RTX 5090 at 104 TFLOPS. This places the estimated results for the PS6 GPU between the two. The analysis also predicts that, with the upgraded PSSR 3, its effective graphics output speed could be 35% to 50% higher than the nominal figures, but these conclusions are all based on leaked specifications and theoretical calculations.
Original report from foreign media:
Based on analysis of previously leaked specification information, AMD's custom GPU for the PS6 is expected to adopt a customized version of the RDNA5 architecture, built using TSMC's 3nm process. It will feature 54 Compute Units (CUs) with a maximum frequency of up to 3.6GHz, and carry a 160-bit GDDR7 memory bus width with memory speeds reaching up to 32Gbps. The GPU's TDP is reportedly 160W, with theoretical rasterization performance around 34–40 TFLOPS, placing its overall performance roughly on par with AMD's Radeon RX 9070 and RX 9070 XT.
If we directly compare the PS6 GPU with the Radeon RX 9070 XT, there are clear differences in their hardware specifications. The RX 9070 XT uses the RDNA4 architecture, has 64 CUs, a 256-bit memory bus width, and a power consumption of about 304W; although the PS6 GPU has only 54 CUs, its maximum frequency reaches 3.6GHz, significantly higher than the RX 9070 XT's frequency of approximately 2.9–3.0GHz.
According to this analysis, thanks to architectural optimizations in RDNA 5 for high-frequency operation and performance, each CU in the PS6 may deliver traditional graphics performance about 15%–20% higher than the RX 9070 XT. However, in pure rasterization rendering without AI technologies, the RX 9070 XT may still be about 5%–10% faster than the PS6 GPU during sustained operation due to its wider memory bus width.
The real differentiator could be Dual Issue and PSSR 3. According to leaked information, in games with full path tracing enabled, the PS6 GPU's actual performance might even exceed that of the RX 9070 XT by 70%–100%. RDNA 5's dedicated architecture modules support Dual Issue parallel execution, which can further boost FP32 computing power, raising the GPU's theoretical peak from a nominal 40 TFLOPS to approximately 68–80 TFLOPS.
If we simply compare theoretical computing power, this means the PS6 GPU's peak performance will fall between the GeForce RTX 5080's 56.3 TFLOPS and the RTX 5090's 104 TFLOPS. Combined with the rumored PSSR 3 upscaling technology, the effective output performance of final game visuals is said to improve by another 35%–50% over the theoretical performance, thereby further unlocking GPU potential within the 160W power limit.
