But the intention of mature multi-die GPUs as it would likely be once it reaches consumers (Based on the wealth of research put out not just by academics but the likes of NVidia, Intel & AMD) is for the device to operate as a single processing unit on a single workload, with other components that don't need to be duplicated in a separate uncore die similar to Epyc2, acting as one whole individual processing unit one one workload, no mGPU technique is comparable to this, every existing technique has two or more independent workloads working in cohort to feed two or more independent processing units.
A multi-die GPU built on similar techniques as Epyc/Zen2 would appear over the PCIe bus as a single device.
A multi-die GPU built on similar techniques as Epyc/Zen2 would appear over the PCIe bus as a single device.
Last edited: