UMA removes the PCIe bottleneck and replaces it with a memory controller + bandwidth bottleneck. For most high-performance GPUs, that would be a direct downgrade.