By using this site, you agree to our Privacy Policy and our Terms of Use. Close

From neogaf: BadBreathoftheWild (his explaination of this)

"At the speed a CPU runs at, having to go get some data from RAM is like walking to the next building, where having it in your cache is like having the data you need in your room.

When you only have 8ms per frame (120 FPS) this kind of latency matters.

Bandwidth is like how big your pockets are, and latency is like how fast you run to go and get that data to stuff into those pockets.

It doesn’t matter if you have massive pockets if the data turns up too late to get the frame out before 8ms has elapsed.

A system that needs to take less frequent trips to RAM because its caches are faster and don’t get entirely invalidated (Coherency Engine) should see less frame drops and have a higher average frame rate as frame times get smaller and smaller.

High frame rate is critical for something like VR, and Sony may have specifically included Coherency Engines, cache scrubbers and be running the GPU as fast as they are doing all to aid latency so PS5 can deliver better VR than PS4 could.

From the wording of what Matt was saying, and from RGT’s sources (no idea how reliable he is), it could be that PS5 CPU will spend less time waiting for a RAM fetch and be doing useful work more often as it either has a unified L3 cache between all cores like Zen3 architecture, or it manages the problems of cache coherency better.

The GPU will already be doing this, and Cerny did say that “one” of the jobs of the Coherency Engines was working with the GPU cache scrubbers, implying it also has a role to play on the CPU side, too.

It could be that XSX has the potential to push more pixels per frame, but PS5 has the potential to push more frames per second.

The relationship between both aren’t linear exactly because of cache/RAM latencies.

PlayStations IO patent also has just as much emphasis on latency as it does on bandwidth, which is why so much of the pipeline is offloaded to fixed function accelerators. Low latency is down as a key requirement, just as much as high bandwidth is."