Notes from the field-- IC7-Max3 plus Prescott, ratios and timings
Straight_Man
Geeky, in my own wayNaples, FL Icrontian
This is mostly about the motherboard, and three Softmenu settings in the BIOS on same.
I discovered, over a month of playing with settings, that part of teh problem with Ocing this board is NOT the CPU itself. BUT, default NB strap, or the speed at whihc data is fed from CPU to NB, IS a problem.
Here's some of what I did, what turned out to be the critical things:
With NB strapped to CPU (default), upper limit for OCing BASE was about 223-224. Take the strap, force it to PSB800 (yes, nominal 200 quad pumped, or Prescott base) and you get this situation: PSB800 forces the NB hyperbus between NB and SB to be locked, and thta lets you run the CPU base asynchronous to the hyperbus linkages to the bridges, and back, from the CPU. I now have a stable CPU at 225-229 base, and the NB fan is not running all out at 5100+ RPM constantly, with hangs every 18-28 hours-- with the NB and hyperbus for bridges and CPU strapped to PSB800. NB is the hangup for OCing, on a stock IC7-Max3 (as far as harrdware, no power circuitry on the board is modeed at all, the OTES fan and tunnel are in place, etc) with an i875 chipset. NB is fanned, SB not, but what locks is off the NB per the chipset specs. See below about RAM for another reason I say this, other than the fact that locking the NB strapping worked to make the box stable at higher CPU bases.
RAM is another issue, you might have to take it off of 1:1 to get RAM to best speeds for your sticks versus an OC'ed CPU base (default is exactly the same as CPU base X 2, ram cannot hack DDR500 timings, that is an "OC too far" for this RAM). With the LLPT PC3200 nominal Corsair, I can run RAM at 1:1 up to a CPU base of 225, which is DDR450 at almost low latency BIOS SET SPD based settings of 2.5,3,2,6 (note nominal for that RAM is 2,3,2,6 and softmenu is adjusting it with RAM set to SPD or otherwise, to the figure given). Now, for RATIOS for async RAM, the IC7-Max3 uses CPU:RAM (per BIOS, I know this MIGHT seem odd to some, who think that it should be RAM:CPU, but the calcs and performance and RAM testing and hang tests I did show it is as the BIOS states it in fact and it truth for this board). So, for above 225 base, I end up with a CPU at 5, RAM at 4 for my RAM, PC3200 Corsair LLPT sticks, QTY 4 of this exact same kind). At or below 225 CPU base, I can run 1:1 or faster ratios, with RAM faster than CPU, so long as RAM is not timed faster than DDR450 as pumped rate.
Third issue, is I have a limited graphics card, not even publishing which one, but for my needs I do not need a faster one so that will not matter for me. So, for my video and PCI bussing's sake, I have PCI and AGP locked, AGP at 66, PCI at 33 (in MHz). Trick with Prescott setups is knowing what ratios you can free-run to limit of devices in box.
A combo issue, influenced indirectly by above:
Fourth issue (not as major, but there) is storage, pump PCI or the hyperbus too fast, you can get pumped storage timings sent to the IDE controller, possibly also SATA, and HDs CAN get fed data too fast. Then you start getting corruption of data, possibly escalating as to speed and how it builds up and aggregates as amount of junk files versus the good original files written when box was not OC'd and not modified since. timings for the other busses can propagate into strange happenings elsewhere in box.... Every time I unlocked AGP and PCI, I got junk data written to HD, had archives that were corrupt at download when checked with checksum routines, locking the AGP\PCI fixed for my IDE HDs on this particualr hardware set. Lock them both, not only does video run right, but storage and NTFS utils and fixers behave much more stably ALSO.
Just figured, given the number of folks trying to OC Prescott (not verus AMD, versus those who hang out here and do not OC or at least TRY TO, most of their personal gear and who happen to be trying Intel versus those with Intel based boxes and do hang here but do not OC at all), that this might help some of us. Similar things are happening with my Northwood box, but not so tight as with Prescott.
I'll talk about the heating issues as separate from setup and timing issues in more detail in another forum area when I have time, OTES temp does climb when the CPU temp is over a relatively certain point, with any of three GOOD HS's in place, and several fans tried for HSs. RADIANT heat effects that yield a higher case temp close to OTES are affecting OTES temps, the biggest OTES tunnel intakes are close to CPU HS and PSU.
I discovered, over a month of playing with settings, that part of teh problem with Ocing this board is NOT the CPU itself. BUT, default NB strap, or the speed at whihc data is fed from CPU to NB, IS a problem.
Here's some of what I did, what turned out to be the critical things:
With NB strapped to CPU (default), upper limit for OCing BASE was about 223-224. Take the strap, force it to PSB800 (yes, nominal 200 quad pumped, or Prescott base) and you get this situation: PSB800 forces the NB hyperbus between NB and SB to be locked, and thta lets you run the CPU base asynchronous to the hyperbus linkages to the bridges, and back, from the CPU. I now have a stable CPU at 225-229 base, and the NB fan is not running all out at 5100+ RPM constantly, with hangs every 18-28 hours-- with the NB and hyperbus for bridges and CPU strapped to PSB800. NB is the hangup for OCing, on a stock IC7-Max3 (as far as harrdware, no power circuitry on the board is modeed at all, the OTES fan and tunnel are in place, etc) with an i875 chipset. NB is fanned, SB not, but what locks is off the NB per the chipset specs. See below about RAM for another reason I say this, other than the fact that locking the NB strapping worked to make the box stable at higher CPU bases.
RAM is another issue, you might have to take it off of 1:1 to get RAM to best speeds for your sticks versus an OC'ed CPU base (default is exactly the same as CPU base X 2, ram cannot hack DDR500 timings, that is an "OC too far" for this RAM). With the LLPT PC3200 nominal Corsair, I can run RAM at 1:1 up to a CPU base of 225, which is DDR450 at almost low latency BIOS SET SPD based settings of 2.5,3,2,6 (note nominal for that RAM is 2,3,2,6 and softmenu is adjusting it with RAM set to SPD or otherwise, to the figure given). Now, for RATIOS for async RAM, the IC7-Max3 uses CPU:RAM (per BIOS, I know this MIGHT seem odd to some, who think that it should be RAM:CPU, but the calcs and performance and RAM testing and hang tests I did show it is as the BIOS states it in fact and it truth for this board). So, for above 225 base, I end up with a CPU at 5, RAM at 4 for my RAM, PC3200 Corsair LLPT sticks, QTY 4 of this exact same kind). At or below 225 CPU base, I can run 1:1 or faster ratios, with RAM faster than CPU, so long as RAM is not timed faster than DDR450 as pumped rate.
Third issue, is I have a limited graphics card, not even publishing which one, but for my needs I do not need a faster one so that will not matter for me. So, for my video and PCI bussing's sake, I have PCI and AGP locked, AGP at 66, PCI at 33 (in MHz). Trick with Prescott setups is knowing what ratios you can free-run to limit of devices in box.
A combo issue, influenced indirectly by above:
Fourth issue (not as major, but there) is storage, pump PCI or the hyperbus too fast, you can get pumped storage timings sent to the IDE controller, possibly also SATA, and HDs CAN get fed data too fast. Then you start getting corruption of data, possibly escalating as to speed and how it builds up and aggregates as amount of junk files versus the good original files written when box was not OC'd and not modified since. timings for the other busses can propagate into strange happenings elsewhere in box.... Every time I unlocked AGP and PCI, I got junk data written to HD, had archives that were corrupt at download when checked with checksum routines, locking the AGP\PCI fixed for my IDE HDs on this particualr hardware set. Lock them both, not only does video run right, but storage and NTFS utils and fixers behave much more stably ALSO.
Just figured, given the number of folks trying to OC Prescott (not verus AMD, versus those who hang out here and do not OC or at least TRY TO, most of their personal gear and who happen to be trying Intel versus those with Intel based boxes and do hang here but do not OC at all), that this might help some of us. Similar things are happening with my Northwood box, but not so tight as with Prescott.
I'll talk about the heating issues as separate from setup and timing issues in more detail in another forum area when I have time, OTES temp does climb when the CPU temp is over a relatively certain point, with any of three GOOD HS's in place, and several fans tried for HSs. RADIANT heat effects that yield a higher case temp close to OTES are affecting OTES temps, the biggest OTES tunnel intakes are close to CPU HS and PSU.
0
Comments
1. There is no hyperbus.
2. There is no CPU base.
3. You can't feed data "Too fast" to an HDD.
4. There is no "PSB"
5. There is no NB strapping
...
20. Everyone knows that PCI/AGP should be locked at 33/66.
Even though I want to, I simply <b>cannot</b> comprehend what you just tried to tell us.