Been playing around with tREF on my system in sig. With the memory @ 250 Mhz @ 2-2-2, 1T, 200 divider, I got the following graph: EDIT - Yellow line shows 2GB OCZ @ 213 Mhz. As you can see, a clear relationship between tREF and errors, and a slight reduction in bandwidth @ low tREFs. My questions are: Is the ideal tREF specific to the RAM? Is it specific to the memclock? Is it specific to the divider? From previous discussions, I'm quite a fan of the following: ideal tREF = refresh rate / (1/clockspeed) The problem is that there is so much mis-information about tREF about, and that table showing tREFs grouped into refresh rates (3.9, 7.8, 15.6 us) and clockspeeds (100 - 200 Mhz) only confuses things. Don't even get me started on Dracula's tREF tables . Does anyone have any simple theory on tREF? At least Dracula's four best and four worst tREFs bear some resemblance to my findings: Four best = * 1560 * 3072 * 3120 * 4708 Four worst = * 0016 * 0032 * 0064 * 0128 My four best (1GB BH5) = * 3684 * 3072 * 4708 * 4196 EDIT - My four best (2GB OCZ) = * 2336 * 2048 * 2560 * 1536 I can only assume higher tREF = better, but knowing that the memory must be refreshed to keep the data 'alive', a higher tREF could also result in data loss...
Looking astaris' post here: http://www.dfi-street.com/forum/showpost.php?p=139338&postcount=27 It seems that my best tREFs all fall in his 1.95us category. Namely: 1,95 us 1536= 100mhz -> 195 Mem Clock cycles 2048= 133mhz -> 259 2560= 166mhz -> 324 3072= 200mhz -> 390 1.95 us 3684= 120mhz -> 234 4196= 140mhz -> 273 4708= 150mhz -> 292
Wow, nice work mate, I'll make sense of it in the morning, but looks like you've been working hard on this
I was bored after having to RMA my new 2GB RAM, so I thought a hungover day of tinkering with BIOS settings and running memtest while listening to music and snacking would be just what the doctor ordered
