To summarize, in case 1, the network card gets stale data due to a cached write. In case 2, the processor gets stale data due to a cached read.

My understanding is that in this slide, when Processor writes to Cache, there will be a mechanism (network card) that flush the updated value in the Cache to Memory through Interconnect, and then update that value to all caches in all Processors if the same address exist in those caches.

Now, what is the point of cache in Processor since whenever we update the value in cache, it will take cycles to update memory and then update all other caches?

@wanze I'm not sure if that interpretation is accurate. The network card is a computer system's interface to a computer network. Its job is to write the data from the message buffer in memory to the network (whether it be wired or wireless; it isn't involved in flashing data from a processor's cache to memory!

@wanze To answer your second question, a BusWB operation, which is what I assume you mean when you say "update memory," is only triggered in one of the two following situations:

1. A $-line is in the M state in one $-controller and another $-controller issues either a BusRd or BusRdX.
2. A $-line is being evicted either because the $-controller has maxed out its capacity or there has been a $-line conflict.

So, in regards to your question, the cache is still very useful because we don't have to update memory on every update!