PlayStation 4 Lead Architect Mark Cerny is one of the men that took the decision to build the previously almost unbelievable eight gigabytes of GDRR5 RAM in a nice little circle around the console’s CPU, and to arrange them in an unified architecture that shuffles the memory dynamically between GPU and CPU. Today he explained the reasons behind that choice in an interview on Engadget.
You know, performance has a lot to do with memory speed and GDDR5 has very high bandwidth. In our case it’s 176 gigabytes per second, but I mean, it depends on the width of your BUS and so on and on… The number of gigatransfers per seconds you can have with your hardware…
The reason why unified is important is that you have access that primarily the CPU will use, acces that the GPU will use, and the ratio of that changes fluidly throughout the game.
So, as you’re driving down the street in one of these simulated worlds, the particular data that you’re loading in, you don’t really know whether it’s on CPU or GPU.
With a hardware like PlayStation 3 for example, with 256 megs on the CPU side and 256 megs on the GPU side, that gets very awkward, because you’ll run out of one or the other pretty much continuously, and you have to come up with crazy schemes to try to shuffle it all back back and forth.
If it’s unified you don’t need to think in those terms.
One thing is for sure. Cerny is the only one that manages to make terms like “gigatransfers” sound like poetry. Will the choice prove a winning one? So far it definitely looks like a good approach, but we’ll have to see it tested in a couple years, when we’ll be deeper in the coming generation.