Digging into the GPU internals reveals differences that reinforce the pattern established in the Performance group. The base GU605 houses a GPU with 10,496 shading units, 328 TMUs, and 128 ROPs, while the Ultra 9 / RTX 5080 variant operates with 7,680 shading units, 256 TMUs, and 96 ROPs. These are not minor gaps — fewer shading units directly reduce the parallelism available for rendering complex scenes, while lower ROP and TMU counts constrain pixel output and texture throughput respectively. Both GPUs share the Blackwell architecture, a 256-bit memory bus, and identical effective memory speeds, so the bandwidth infrastructure is equivalent; it is the compute resources attached to that bandwidth that differ.
The TDP split is also notable: the base model's GPU operates at 95W versus 80W for the Ultra 9 variant. A higher TDP generally allows the GPU to sustain its performance envelope for longer under load, which matters in extended gaming sessions where thermal throttling can erode frame rates. The 15W difference is meaningful and aligns with the base model's broader performance lead established across spec groups. CPU-side attributes — L3 cache, socket, instruction sets, big.LITTLE architecture, and maximum RAM speed support — are identical across both, confirming once more that the processor is not a differentiating factor.
The base GU605 holds a clear advantage in this category. More shading units, more TMUs, more ROPs, and a higher GPU TDP collectively paint a consistent picture: the base configuration carries a more capable discrete GPU despite the Ultra 9 variant's premium positioning in the product name.