Opteron Socket939 Overclocking Discussion Thread

[KingYubbo]

Good Luck T_M! You'll need it!

[groovetek]

lol... hey, instead of chucking it up to 1.77v straight away, how about see how 1.65v goes... lol... uc... if it was me pushing the chip - i wouldnt' even think twice about running high volts, since it's my responsibility, and if it dies then my own bad luck... but if you kill it, i'd be annoyed, and yet, no right to be... if u know what i mean...

anyway... finally i can go to sleep... finished an assignment due tomorrow... started today pretty much ehehe...

here's a sexy graph (Magna engine power output efficiency basically):

[eva2000]

face it groovetek you're not going to be able to sleep well thinking about that opteron

testing 32M at 3007mhz and weirdess experience so far with 2x 512MB Gskill PC4400LE on 180 divider 32M would clonk out 2-3 loops in at 334HTT/300mhz 2.5-4-3-6 at 2.85-2.95v

so decided to try 2x 512MB Patriot PC3200XBL TCCD at 180 divider 32M and would clock out 1-2 loops at 334HTT/300 2.5-4-3-6 at 2.69v

so thought why not try 166 divider with mem at 273mhz 2.5-3-3-6 7-17-2223 with Patriot PC3200XBL

32M
at 2.69v = clonks out 2 loops in
at 2.76v = clonks out 4 loops in
at 2.79v = clonks out 5 loops in
at 2.85v = clonks out 6 loops in

so change from 7-17-2223 to 8-18-2223 loosening Trc and Trfc
at 2.85v = passed in 24m 29s!



Prior to this revelation, i was upping vcore with no effect and it turns out it was vdimm that was lacking! So now i need to lower vcore and see what happens

Strange is at 300-312mhz 2.5-4-3-6 only 2.69v is optimal for this Patriot, but at 2.5-3-3-6 it handles volts fine!

Perc on xtremesystems.org uses same ram and confirmed this weird experience he said 2.7v was optimal for his Patriot PC3200XBL at certain timings 2.5-4-3-x and 2.5-4-4-x but at 2.5-3-3-6 he could force 3.0v into them with no probs and even 2-3-3-6 at 240mhz at 3.0v!

edit: Can confirm so far, lowered vcore by 0.025v in bios and tad da!

[eva2000]

Could be that 180 is a non-standard divider than only some board manufacturers implement i.e. DFI so it could be opteron s939s built for workstation platform doesn't fully support 180 divider or needs bios tweaks which DFI has stated will not provide support for s939 opterons

further proof

[dinos22]

that's a great find

yeah it looks like my and tamille are just not getting our RAM timings right i guess

By eva2000

face it groovetek you're not going to be able to sleep well thinking about that opteron

testing 32M at 3007mhz and weirdess experience so far with 2x 512MB Gskill PC4400LE on 180 divider 32M would clonk out 2-3 loops in at 334HTT/300mhz 2.5-4-3-6 at 2.85-2.95v

so decided to try 2x 512MB Patriot PC3200XBL TCCD at 180 divider 32M and would clock out 1-2 loops at 334HTT/300 2.5-4-3-6 at 2.69v

so thought why not try 166 divider with mem at 273mhz 2.5-3-3-6 7-17-2223 with Patriot PC3200XBL

32M
at 2.69v = clonks out 2 loops in
at 2.76v = clonks out 4 loops in
at 2.79v = clonks out 5 loops in
at 2.85v = clonks out 6 loops in

so change from 7-17-2223 to 8-18-2223 loosening Trc and Trfc
at 2.85v = passed in 24m 29s!


Prior to this revelation, i was upping vcore with no effect and it turns out it was vdimm that was lacking! So now i need to lower vcore and see what happens

Strange is at 300-312mhz 2.5-4-3-6 only 2.69v is optimal for this Patriot, but at 2.5-3-3-6 it handles volts fine!

Perc on xtremesystems.org uses same ram and confirmed this weird experience he said 2.7v was optimal for his Patriot PC3200XBL at certain timings 2.5-4-3-x and 2.5-4-4-x but at 2.5-3-3-6 he could force 3.0v into them with no probs and even 2-3-3-6 at 240mhz at 3.0v!

edit: Can confirm so far, lowered vcore by 0.025v in bios and tad da!

[eva2000]

Yeah seems from ram i tested so far, the Opteron 144 mem controller isn't as good as FX-57 still in not handling as tight timings and lower vdimm but still fairly close and still Opteron 144 mem controller is beter than 3700/4000 sandies and venice cpus

[dinos22]

By eva2000

Yeah seems from ram i tested so far, the Opteron 144 mem controller isn't as good as FX-57 still in not handling as tight timings and lower vdimm but still fairly close and still Opteron 144 mem controller is beter than 3700/4000 sandies and venice cpus

and remember there are no optimised bioses for these CPUs either......i mean people say it's based on San Diego.....that may be the case but it clearly behaves different to other CPUs......

have you tried your previous timings again to see if they still fail in 32M....just in case

also are the new timings 100% 32M stable.....i find that my 2.9GHz 32M runs fail sometimes when i play with timings......i am yet to find a sweetspot for it........i'll give your new mix a shot as well

you've changed your TREF value as well........are you setting a value in bios or leaving it at auto to find optimal TREF values.....they always seem to default to 166 7.8us or 3.xus in windows

[eva2000]

Gskill PC4400LE for liked 3072 TREF, Patriot PC3200XBL likes 3684 on FX-57 same values i'm using on Opteron 144 respectively for them.

I only tried 5x 32M runs with 5 reboots at same speed - if you can call that 100% 32M stable

All values set in bios only, i only use a64tweaker to show the values never to set them in windows