seems these findings apply to Intel D975XBX bad axe as well, loosened tRFC/tRD from 35/6 to 40/7 and increased my max 32M benchable fsb from 433fsb to 440fsb!
Maybe all motherboard manufacturers need to do is open their bioses to the other memory timing options that memset has available so we can further fine tune how high the fsb can go ?
Decided to see how far i can tighten the additional timings knowing the effects tRD has on higher fsb.
Software used: Systool dual 32M Pi tests as 99.99% of time if you can pass dual Systool 32M pi you can pass Super Pi 32M.
Using the memset values for the following:
tRFC | tRD | Write to Precharge Command | Write to Read Command | Read-Write Turnaround clocks | Write-Read Turnaround clocks | tRTP
Managed to pass with following settings:
- 440fsb = 40 | 7 | 12 | 10 | 8 | 5 | 4
- 438fsb = 36 | 7 | 12 | 9 | 8 | 5 | 4
- 438fsb = 35 | 7 | 12 | 10 | 8 | 5 | 4
- 438fsb = 34 | 7 | 12 | 11 | 8 | 5 | 4
- 438fsb = 33 | 7 | 12 | 11 | 8 | 5 | 4
- 438fsb = 32 | 7 | 12 | 11 | 9 | 5 | 4
- 438fsb = 30 | 7 | 12 | 11 | 9 | 5 | 4
- 438fsb = 28 | 7 | 12 | 11 | 9 | 5 | 4
@438fsb = 36 | 7 | 12 | 9 | 8 | 5 | 4
Well so far 3954mhz is max 32M.
Here's the break down of results for max 32M
@3898mhz - 9x433fsb max
with tRD at board default 6
@3934mhz - 9x437fsb
@tRD of 7 with tRFC 28 and looser additional timings
@3934mhz - 9x437fsb
@tRD of 7 with tRFC 32 and tighter additional timings
@3954mhz - 9x439fsb
@tRD of 7 with tRFC 36 and additional timings needed for 439fsb
Couldn't get 32M @440fsb for now probably more tweaking needed or more vcore/mch volts.
But here's 8M
All 32M times run from same session of windows, so probably not optimal for fastest times yet
So by playing with tRD and tRFC and additional timings, I managed to increase my max fsb from 432-433fsb to 439-440fsb :D