Hi all.
I've been experimenting on the bench with a second 4coredual board I have and am seeing some odd results in overclocking experiments, for example, has anyone seen Orthos run for 7 hours and then fail?
Hardware:
4coredual-sata2 modo bios ver 1.6
E2200 CPU
Crucial Ballistix Tracer 1G Memory stick (PC2-6400)
Enermax 350W PS
Video card: oldie ATI 7000/7200
OS: Winxp
HD: 80G IBMGXP model
I was surfing Newegg recently and saw the Crucial Ballistix Tracer 2x1GB memory for $65 with a $40 rebate and free shipping, so I bought some to tinker with. I never would have imagined 2 Gig of memory for $25 in years gone by.
I did NOT do a BSEL mod since forcing the E2200 to boot at 266 FSB is a default bootup CPU freq of 11*266 or 2926 MHz which is a bit hi.
All of this hardware is setup on my workbench, no case is used, bench is basement located, ambient temperature is rather cool so overclocking experiments are fun since temps never get out of hand.
I did implement this Vcore mod:
http://forums.vr-zone.com/showthread.php?t=192571
using a 100K 20-turn precision pot
For those interested in knowing, I am uncertain how the poster for the VR zone Vcore Offset voltage mod came up with his formula to calculate the offset or final Vcore voltage, but this formula from the L6714 chip datasheet is very accurate to calculate the offset votage generated:
Ros=(1.240/Vos)*Rfb
where
Ros = offset resistance inserted in place of the zero ohm resistor (the potentiometer)
Vos = offset voltage desired (to be added to VID programmed Vcore setting)
Rfb = feedback resistor value from the Vcore circuit (which is 4.4 Kohm according to the VR zone link above, and appears to be the same for the 4coredual-sata and the 4coredual-vsta boards)
So, if you want a .1v Vcore increase,
Ros=(1.240/.1)*4400=54560 Ohms
What I have found is this, you MUST be careful when adjusting the variable resistor or potentiometer as the L6714 chip/circuit apparently does NOT respond to power-on Ros adjustments. IOW, one cannot view the Vcore output in the BIOS voltages/temp screen while adjusting the pot/trimmer as the voltage does NOT change until a 5vsb power cycle has occured. You must power down fully (kill standby power even) and then power up the mobo to realize the changed Vcore and Vos. I just about fried the CPU as I was attempting to adjust Vcore by viewing Vcore in the BIOS and cranking on the pot and seeing no changes, then on reboot and jump into BIOS saw Vcore all the way up to 1.8v, oops, a quick kill of the PS was warranted at that moment. Some probing around revealed that the L6714 chip has power applied to it from the 5v standby circuit, thus it appears 5vsb must be turned off to "reprogram" the L6714 chip with a new Ros resistance.
FYI The VR Zone link above provides this formula to calculate the resulting Vout: voltage:
Vout=Vstock/(1 + 3,000/Vos) and I've not bothered to figure out where it came from in the L6714 datasheet.
Anyway, I tried running the system with memory setting of 266 for the Ballistix stick. I could not get a stable overclock above 255 FSB, even with Vcore bumped up by about .1v to 1.42. I then, just for grins, set the memory type/Dram Freq to 333 and all other settings on "auto" for memory, and also enabled the "flexible" mode just to see what the system did, and the BIOS sets the mem/fsb ratio to 1:1. Now comes the odd part, my overclocks are now stable at 266 FSB, and a recent Orthos test at 270 FSB ran 7 hours before a failure. Seems rather odd to have a failure after so many hours of testing. I suppose it could have been a droop of Vcore in response to a load change.
Anyone else seen this type of result? Temps (viewed using the Coretemp program) never got above 55C during this testing, and were mostly in the hi 40s.
Do you suppose the "auto" settings are somehow different for the memory timings with Flexible enabled versus disabled, and 266 or 333 memory speed selected, even tho the FSB-MEM ratio is still 1:1 per CPUz under either setup? CPUz reports memory timings of 4-3-3-9 with 333 Dram Freq setup in BIOS and Flex enabled. The memory sticks are spec'ed at 4-4-4-12 at 400 Mhz and 2.2v per their SPD data.
My other 4coredual-sata rig uses an E2180 CPU (BSEL mod'ed), which runs at 286 FSB for a cool 2860 MHz overclock that is all day stable on prime95/orthos. In my estimation, running at 286 FSB and 10x with the E2180 is probably a better OC than 270 FSB and 11X with the E2200 since memory speeds are a bit better at FSB of 286. So my advice to anyone purchasing a CPU, buy the E2180 and save $10 (or so) and likely you will get a better overclock on this mobo.
OK, I went back and did a little more investigation as these things always have an explanation, and I found out that when I select 266 or 333 for the Dram Freq in the BIOS, and Flexibility Mode disabled (all other mem settings on Auto, AGP voltage HI and Dram voltage on HI), the resulting DRAM freq is 340 (with CPU FSB at 255) as reported by CPUz, a 3:4 ratio, which explains why CPU has problems at 270 MHz with Flex turned "off" as this combination of settings causes the memory to run at about 340, perhaps higher than the chipset can handle, dunno. Guess I need to go back and fiddle with memory settings to see what it takes to get stable operation with the higher Dram freq.
Turning Flexibility "on" in BIOS gives a DRAM freq equal to CPU FSB or 270 MHz regardless of whether Dram Freq is set to 266 or 333 in BIOS. Timings for memory with Flex enabled are 4-3-3-9 at 270 MHz CPU FSB, and mem timings with Flex "off" are 4-4-4-12 and Dram freq of 340 and CPU FSB of 255 (can't boot winxp at 270 CPU FSB and Dram freq of 340, but system DOES boot up at FSB of 255 and Dram freq of 340).
Interesting stuff!
Edit:
Orthos failed again after an 8 hour run so I bumped the Vcore up a bit more, am now at 9 hours 16 mins and running with a vcore that measures 1.432v by DVM.