Unless you know how a specific card/driver combination is handling memory management, you can't gather any useful information from any measure you make of utilized memory.
You don't know if it's cache.
You don't know if it's memory that has yet to be freed, but can be on demand.
You only know the number that one specific counter is giving you.
Imagine the confusing state of memory usage that you would have, if, on a windows 7 system you could not see how much memory is functioning as cache. The values would be nearly useless.
Until AMD and NV start publishing their memory management techniques, we really have very little clue what using a specific amount of memory on one of their cards means. We can relatively compare on the same card and the same driver revision that activity X creates a memory utilized amount of Y, and that activity A creates a memory utilized value of B, but we cannot extrapolate what that actually means functionally for other situations nearly as much as people seem to be trying to do.
You don't know if it's cache.
You don't know if it's memory that has yet to be freed, but can be on demand.
You only know the number that one specific counter is giving you.
Imagine the confusing state of memory usage that you would have, if, on a windows 7 system you could not see how much memory is functioning as cache. The values would be nearly useless.
Until AMD and NV start publishing their memory management techniques, we really have very little clue what using a specific amount of memory on one of their cards means. We can relatively compare on the same card and the same driver revision that activity X creates a memory utilized amount of Y, and that activity A creates a memory utilized value of B, but we cannot extrapolate what that actually means functionally for other situations nearly as much as people seem to be trying to do.
