Erthazus proves for the hundredth time that he doesn’t know anything about system architecture or how games work. Not a single thing.
Here’s a few ways in which more memory can improve game performance and image quality:
1) Assets that would have been streamed in a memory-tight system can be memory-resident, which means vastly lower latency before they can be fetched for use in rendering, eliminating pop-up.
Streaming requires 4 things: Memory bus bandwidth to transfer the compressed package from disk to decompression buffer, a decompression buffer, a processor core (or a time-slice of one) to perform the decompression, and the memory bandwidth required to read the compressed file and write the decompressed assets. All these actions impact the performance of the machine as a whole.
2) Having the memory to retain high resolution geometry and shader buffers allows for higher precision geometry-based antialiasing instead of relying on post-render edge smoothing algorithms.
3) Not having to scrub the geometry buffers also means easier and more accurate stereoscopic 3D rendering. The ‘cheat’ 3D techniques of Trivioz, et al. evolved out of a need for dual-view rendering without access to a full depth buffer.
4) More frames of animation can be stored in memory, meaning not only higher quality animations but less need to interpolate frames, which takes up processor and memory bandwidth.
5) Many other real-time calculations that are currently performed can be tabulated if there is memory to spare. Look-up tables have been a staple in increasing performance since day one. It’s an oldie but goodie: The SIN/COS lookup table was the holy grail of circle rendering back in the day. The principle remains.
On the GPU access to 2Gb rumour: That would only really make sense if it was a discrete GPU. The GPU in the PS4 (correct me if I’m wrong here folks) is built into the CPU die as part of an APU, and so it’s access to RAM would be through the same built-in memory controller as the CPU cores as there is effectively no Northbridge.