Solor,
Higher tREF only gives better results because increasing the turnaround between Self Refresh cycles frees up memory cycles which in turn reduces latency. This is a cheat way to decrease latency, and a good way to corrupt your memory for stable system usage. Self Refresh is a procedure to ensure that data stored in memory is available is kept valid and not lost. Otherwise you just consume more bandwidth on the bus having to request the data that wasn't paged where it was to supposed to be for the requesting device. On top of that it's also to make sure that all memory is refreshed successfully during the cycle. If its too short there is a chance that it wont be completed, if its too long there is a good chance that you memory ic's will lose charge before the next refresh and result in page faults or increase access time to fetch data, the time to fetch data across the bus requires far more cycles than it does to fetch from DRAM. It may seem irrelevant but if you are doing that often it will add up to a lot of wasted system cycles and bus time. Worst case is you some how end up with memory pages holding incomplete data from parts being lost due to expired memory or being rewritten with something completely different.
My point is it's not good to use it full time, its a way to drop latency in benchmarks and thats all its safe to use it for. By setting it so high you cripple your systems memory performance for tasks which run for extended periods of time, and get lots of page faults
I've already asked Felix if he can add to Memset, auto tREF value setting according to DRAM frequency.
Also put in request to Asus bios engineers to add self-tracking tREF in bios or manually set tREF whichever they rather do.
Reply With Quote
Bookmarks