It's been a very long time since my computer engineering course, and we didn't cover this topic specifically, but I highly doubt it's a full dump and reload. What likely happens is each ram address has a ttl flag or some way for the CPU to know when to rewrite the data, and it does it as needed.

Plus, the bus between the CPU and ram is ridiculously fast. Your pc could dump and reload all of its ram in the time it takes you to blink. And, with multiple cores, the task can be allocated to a single core, or divided up among all of them.

[–] al177@lemmy.sdf.org 3 points 1 year ago

Modern RAM just needs to be told to refresh. The device itself will go through the refreshing process. But the whole array needs to be refreshed, there's no LRU scheme to tell what bank or row was last accessed.

Starting with DDR3 it's not so easy. Density is so high that reading or writing one row affects cells in adjacent rows. Partial target row refresh (PTRR) counters this, where any access of a row is followed by a refresh of adjacent rows. Flaws in this process in early DDR3 controllers was at the heart of rowhammer exploits, where repeated accesses to a memory location could work out what's stored in physically adjacent memory, even if it's not privileged. IIRC DDR4 pulled the PTRR process into the RAM's built in refresh circuitry so it's transparent to the memory controller.

[–] PeterPoopshit@lemmy.world 2 points 1 year ago

At least on older x86 motherboards, there used to be a dram refresh interrupt. It would trigger every 15 or so milliseconds and tell the dram controller to do a bus hold request and then refresh the ram. This bus hold request means the cpu can't access the ram when this happens (it can still run stuff in the cache) but at least you aren't wasting as much cpu time on dram refresh this way.