Originally Posted by GuardianOdin
What would the benefit of a fusion chip? I like the idea, but would like a better understanding of it.
There are several reasons why a GPU on die with the CPU makes sense: memory access, GPU computing, power savings, and economics.
Memory Access - Integrated GPUs (iGPU)share system memory since it doesn't have its own dedicated VRAM. The way it works now is that the GPU is on the NB. In order for the iGPU to use the memory that is connected off the CPU, the iGPU sends the request out via the Hypertransport bus to the CPU's memory controller (MCT) and then to the memory before making its way back to the iGPU. Because the physical distance is so far apart between the iGPU and the RAM, the latency is greatly increased. Putting the two side by side, the latencies will be greatly decreased. Latencies should further be decreased since the two will be communicating natively with each other without the use of the Hypertransport bus.
GPU computing - Because GPUs are so powerful these days, combining the two to maximize the advantages of both will yield a result that is greater that simply the sum of the two parts. The first generation will simply be a separate GPU right next to a separate CPU. In time, more and more of the GPU will be intertwined with the CPU so that the two will no longer be distinguishable from each other - true "fusion."
Power savings - Before the 780G chipset, all iGPUs are traditionally fabbed on manufacturing processes that are at least one generation behind CPUs so the iGPUs consumed more power. Also, in AMD systems, since the iGPU needs to use the memory that is attached to the CPU, memory access by the iGPU needed to "wake up" the CPU to its full power state, even if the CPU is idling. This is solved with the new generation of AMD CPUs where the memory controller has its own power plane but the Hypertransport bus still consumes power, especially under high memory access.
Economics - By placing the two on a single packaging, the NB will be rendered completely obsolete. As a result of this, mobo manufacturers will no longer need to worry about the VRMs, PWMs, tracing, heatsinks, etc. associated with having a NB. AMD is also guaranteed to make more money since they will be pushing the platform of the CPU and GPU and the mobo makers will probably end up getting more of a price break.