Streaming 4-byte stores at a device over PCIe

A tight loop writes small 4-byte values to a memory region marked write-combining (WC), e.g. a GPU or device-mapped ring buffer over PCIe, rather than normal cacheable write-back memory. The core provides only a handful of write-combining buffer entries, each able to accumulate one cache line's worth of pending writes before it must be flushed as a single bus transaction. Case A: the loop finishes writing every byte of one cache line's worth of data before moving on to the next line. Case B: the loop interleaves small writes across more distinct cache lines than there are WC buffer entries, touching each line only sporadically. How do these two access patterns differ in practice?

18 of 47 in Computer Architecture