Gigabyte P965-P35 X38 overclocking and BIOS tweaking Guide

Sorry I havent found a source for the pushpins. I replaced mine with nylon 4-40 size (#4 screw 40 threads per inch) from the lowes specialty hardware drawers but you have to be able to get to the bottom of the board to install them. No idea of what that would be in metric. 3/4 to 1 inch in lenght should do find and if long since they are nylon you can snip extra lenght off after washers and nuts are applied.

Thermal pads just do not transfer heat as well from the chip to the heatsink. thats why you see so many references to Artic Silver 5 thermal grease compound. If there is no other choice a pad is better than nothing, and they are much improved over the last couple of years, but for best heat transfer a good grease properly applied is favored.


350x9 2.5 memory multiplier with 4-4-4-12 timings I can get at least one hour out of orthos. Running 350 x9 on 2.66 mem multiplier I fail after 5 mins even with very loose timings like 5-5-5-18. Thats all chipset related right?
Click to expand...

350 x 2.5 = 875MHZ good OC for memory at tight timings. Much more and typically timings will require loosening up.

350 x 2.66 = 931MHZ thats pushing it even with relaxed timings. Put the memory on auto at the 2.66 divider and see what the board sets the timings at. I would say that failure is ram, but I am often wrong. At high speeds like that (480x2 = 960), my board starts setting speeds like 5 6 6 27 !!!

So hard to say with that 2.66 mulitplier, most likely ram as you have proven the chipset can do higher ( I think, I might be getting posts mixed up, I should be in bed LOL )
 
Bill, let me first say that this was an amazing guide and it helped me tremendously. I am very grateful of your hard work that you have put into this.

Let me preview my hardware:
Gigabyte 965P-DS3 Rev. 3.3
Intel Core 2 Duo E6420
Corsair XMS2 (2x 1GB) PC2-6400

Currently I have my system overclocked to 2.96GHz @:

370 FSB x 8
and my memory at 5-5-5-15 (default timings are 4-4-4-12) 2.5 multiplier

I ripped off the MCH heatsink and put on a thermaltake heatsink + fan combo on it, and it runs quite cool now. I also use a big typhoon thermaltake CPU heatsink.

I really want to get it to go to at least 3.0 GHz or 3.1 GHz. But when I set my FSB at 400 x8, and the memory multiplier at 2.0 with 4-4-4-12 timings, Orthos encounters errors 15 seconds in. My memory capped at 939 MHz @ 2.1V memtesting it, so I can't set my FSB higher than 375 or so without reaching my memory's cap with my memory multiplier anything higher than 2.0.

Any suggestions what to try to get over 3 GHz of processing speed?
 
Have you tried leaving the timings at 5-5-5-15 and not their default values with 400 x 8? Also, what are your voltages? CPU temps?
 
Danny Bui said:
Have you tried leaving the timings at 5-5-5-15 and not their default values with 400 x 8? Also, what are your voltages? CPU temps?
Click to expand...

No, I haven't tried those timings yet. My voltages are 1.375 on CPU, 2.1V on memory. CPU temps range from 32-35 idle, to 48-51 load.
 
I agree with Danny and you will see better performance at a 2:1 memory divider anyway.

You have done a great job with the thermals as I can see from your good temps even with your present OC.

Give it another notch of Vcore (Intel max is 1.55 V so you have plenty of room, I dont like higher Vcores but the 8 x 400 is such a nice sweet spot and you have worked the thermals so well that if I had your board I would be willing to go to 1.425 before I chickened out, hopefully just another notch or two and you will get there, its booting and doing windows at 8x400 I am hoping that extra tad of Vcore will stabilize it.


The 2:1 divider is the "synchronous mode" in P965 boards where memory speed is a "match" for other system speeds ( I am simplifying because: A) its complicated. B) the result is well know and not disputed. C) I do not want to even try to type in what little I do know about the details ). Basically a 2:1 divider eliminates wait states (CPU doing nothing) and buffering (delays in memory access just so the timing match up), think of it as everyone in the boat rowing at the same speed. Think of non 2:1 as some people in the boat not rowing at the same speed as others. Even if you only get another 200MHz out of the processor with a 2:1 divider change, usually that extra CPU performance will far outweigh any gain of running 200MHz slower CPU with the memory at some higher divider. (you can run benchmarks once you are done and prove it to yourself, however a pure synthetic bench like Sanda memory bandwidth will of course always show faster is better, so be sure to use something "real". There is a sticky somewhere on benchmarking sw.
 
I kept the Vcore at 1.375, and bumped down the multiplier to 7. I was able to stabilize my OC @ 444 FSB x 7, for 3108 MHz and 4-4-4-12 timings. Both cores run at 30-32 Celsius idle, and 50-51 on load. My computer runs circles around intensive games and I'm amazed every time I play them... Should I continue trying to get a higher OC with a 8 multiplier or is 7 just as well?
 
does anyone have specs for a e6600 ds3 rev 3.3 with buffalo pc2 800? learning this oc and some help is appreciated
 
BillParrish said:
Sorry I havent found a source for the pushpins. I replaced mine with nylon 4-40 size (#4 screw 40 threads per inch) from the lowes specialty hardware drawers but you have to be able to get to the bottom of the board to install them. No idea of what that would be in metric. 3/4 to 1 inch in lenght should do find and if long since they are nylon you can snip extra lenght off after washers and nuts are applied.

Thermal pads just do not transfer heat as well from the chip to the heatsink. thats why you see so many references to Artic Silver 5 thermal grease compound. If there is no other choice a pad is better than nothing, and they are much improved over the last couple of years, but for best heat transfer a good grease properly applied is favored.




350 x 2.5 = 875MHZ good OC for memory at tight timings. Much more and typically timings will require loosening up.

350 x 2.66 = 931MHZ thats pushing it even with relaxed timings. Put the memory on auto at the 2.66 divider and see what the board sets the timings at. I would say that failure is ram, but I am often wrong. At high speeds like that (480x2 = 960), my board starts setting speeds like 5 6 6 27 !!!

So hard to say with that 2.66 mulitplier, most likely ram as you have proven the chipset can do higher ( I think, I might be getting posts mixed up, I should be in bed LOL )
Click to expand...

Hi Bill,
I just came back from the hardware store I couldnt find any nylon #4-40 plastic machine screws I could only find #6 screws will that be too big to fit on the heatsink holes? I noticed you are using nylon, will anything happen if i use metal machine screws? those are the only ones i can find with a 4-40 size.
 
So far, so good. I just ran a baseline after running orthos for about 15 minutes.

E6420 at 36C/39C under load
I'm getting 23C/28C normal

I'm a bit concerned about the discrepancy between the two cores, and wonder if I should take off the tuniq and reapply my AS5.

Other questions. I'm using a tuniq tower, which comes with its own fan controller. I'm keeping the fan on halfway between high and low. One question, though, is with fan settings in the BIOS. Using a Tuniq and an active northbridge cooler, should I enable both the System Fan fail warning and the Power Fan Fail warning?

The other question is with respect to the Bios Recovery 2 utility. There's an Xpress Recovery 2 utility, but it's not clear that this is a bios recovery utility--or at least the word "bios" isn't prominently displayed. This newbie needs key terms displayed prominently.

--ceolstan
 
So I have gone and made some minor tweaks to my rig, started with the suggested 9 x 266 and both CPU-Z and Core Temp still show me at 9x200. I go into my bios, and my changes are still in place, boot back into Vista and it appears as if it's completely ignoring what I have done.

I did enable the No-Execute Memory Protect to no avail, it didn't seem to like SAT changes recommended for Vista so I put those back.

Is there anything else I should try? is Vista the roadblock here or am I missing something simple?

All my info is in my sig, thanks for the help!
 
Well C2D do downclock themselves when nothing major is happening. Try running Orthos with CPU-Z and Coretemp open at the same time. It should show the true speed then.
 
How long do I need to run Orthos for? I'm at almost 1 hour and the readings haven't changed...
 
Really? Damn. Thats weird, it should've showed up. Well, then I'm out of ideas. Sorry. Maybe Bill will help ya out.
 
My setup:

Vista x64, just to complicate things

E6600 (stock voltage reports at 1.35v)
2GB G-Skill 4.4.4.12 @ 800Mhz @ 1.9v
DS3 v3.3 BIOS F11

First, this BIOS reports the Tc at 5-10 degrees HIGHER than the Td - pretty weird. F10 reported it a couple degrees lower.

System is rock solid at stock speeds and I never have an issue with it. Of course, I didn't pick this motherboard for stock speeds!

The system boots no problem @ 333Mhz FSB and x9 multi, but the Orthos "stress some RAM" test fails (see odd note below) around 40 minutes. From experience, Everquest 2 also hangs after an hour or so at this session - obviously the system is unstable. After doing a few tests, I've determined that it's the CPU giving me problems - if I drop the multi, the system is stable. I crank the voltage up to 1.4v and doesn't make any difference except causing my temps to go up under load. I back down to 300Mhz FSB and still no dice, even at 1.4v - same instability. I want the magic 3Ghz number and I figured a E6600 should have been able to make it. Did I just get unlucky and buy the least clockable E6600 out there?

For the record, here is a screenshot of Speedfan at stock speeds under load:

my.php


NOTE: Oddly, the only Orthos setting that EVER gives me problems is the "Large, in-place FFTs - stress some RAM" - the "stress CPU" and "blend" both can run for hours on end even with settings that I know are unstable.
 
After playing some more, I went back to F10 so I can put my alarm back to 70c. I noticed another anomaly - I boosted CPU Voltage to 1.45v in BIOS, but both CPU-Z and Speedfan report my CPU Voltage at 1.41v at boot up, eventually dropping to 1.38v. Is this normal? I turned off all the options I could find in the BIOS which indicated any type of throttling to me (I believe three different options).
 
musika said:
Hi Bill,
I just came back from the hardware store I couldnt find any nylon #4-40 plastic machine screws I could only find #6 screws will that be too big to fit on the heatsink holes? I noticed you are using nylon, will anything happen if i use metal machine screws? those are the only ones i can find with a 4-40 size.
Click to expand...

yes #6 wil be too big, metal is fine, just find some plastic or fibre washers to put under the screw head between the screw head and the bottom of the board. I acually used metal myself but always recommend the plastic for extra safety on other peoples boards.
 
Nailz said:
How long do I need to run Orthos for? I'm at almost 1 hour and the readings haven't changed...
Click to expand...

Hour is fine, its a stablily test, and not a particuarlly good one imo. Run yuor games hard for an evening, bottom line, if the computer will do what you normally do without ANY lockups or freezes, good enough (in my book). Arguments as to what "defines" stable have raged for years. IMO if it works for you, it works good enough.

If it does have problems, back off some.
 
Null1fy said:
I kept the Vcore at 1.375, and bumped down the multiplier to 7. I was able to stabilize my OC @ 444 FSB x 7, for 3108 MHz and 4-4-4-12 timings. Both cores run at 30-32 Celsius idle, and 50-51 on load. My computer runs circles around intensive games and I'm amazed every time I play them... Should I continue trying to get a higher OC with a 8 multiplier or is 7 just as well?
Click to expand...


7 is just at good in your case, better actually, as you are getting more out of your memory because if you go to 8 you might have to back off the FSB and that will reduce your memory bandwidth/speed.
 
Thanks Bill, is there anything you would recommend to show that what I have done has actually sped up my computer?

My changes are only seen in my BIOS, they don't show up in Core Temp or CPUZ, and my 3DMark06 Benchmarks have stayed just about identical. I guess I assumed that going from 9x200 to 9x266 was going to show me something a bit more to this point.

I have had no lockups playing games or doing anything else, but I have no idea if my changes have actually taken effect.

Thanks again for your time.

BillParrish said:
Hour is fine, its a stablily test, and not a particuarlly good one imo. Run yuor games hard for an evening, bottom line, if the computer will do what you normally do without ANY lockups or freezes, good enough (in my book). Arguments as to what "defines" stable have raged for years. IMO if it works for you, it works good enough.

If it does have problems, back off some.
Click to expand...
 
PostItNote said:
After playing some more, I went back to F10 so I can put my alarm back to 70c. I noticed another anomaly - I boosted CPU Voltage to 1.45v in BIOS, but both CPU-Z and Speedfan report my CPU Voltage at 1.41v at boot up, eventually dropping to 1.38v. Is this normal? I turned off all the options I could find in the BIOS which indicated any type of throttling to me (I believe three different options).
Click to expand...

Just updating my own note in case it helps someone else, upping the voltage to 1.45v did the trick - 9x333 is now stable, although my core0 & core1 temps both hit 69c during an 8hr stress test. Higher than I would like, but still a safe temperature from my understanding.
 
PostItNote said:
My setup:

Vista x64, just to complicate things

E6600 (stock voltage reports at 1.35v)
2GB G-Skill 4.4.4.12 @ 800Mhz @ 1.9v
DS3 v3.3 BIOS F11

First, this BIOS reports the Tc at 5-10 degrees HIGHER than the Td - pretty weird. F10 reported it a couple degrees lower.

System is rock solid at stock speeds and I never have an issue with it. Of course, I didn't pick this motherboard for stock speeds!

The system boots no problem @ 333Mhz FSB and x9 multi, but the Orthos "stress some RAM" test fails (see odd note below) around 40 minutes. From experience, Everquest 2 also hangs after an hour or so at this session - obviously the system is unstable. After doing a few tests, I've determined that it's the CPU giving me problems - if I drop the multi, the system is stable. I crank the voltage up to 1.4v and doesn't make any difference except causing my temps to go up under load. I back down to 300Mhz FSB and still no dice, even at 1.4v - same instability. I want the magic 3Ghz number and I figured a E6600 should have been able to make it. Did I just get unlucky and buy the least clockable E6600 out there?

For the record, here is a screenshot of Speedfan at stock speeds under load:

my.php


NOTE: Oddly, the only Orthos setting that EVER gives me problems is the "Large, in-place FFTs - stress some RAM" - the "stress CPU" and "blend" both can run for hours on end even with settings that I know are unstable.
Click to expand...

I got exactly the same thing happening. Depending on load my Tc is 5-10C higher than Td which seems a bit odd. Since my Scythe infinity feels rather cool I tend to believe the Td values. Bill, any thoughts on this?
 
PostItNote said:
After playing some more, I went back to F10 so I can put my alarm back to 70c. I noticed another anomaly - I boosted CPU Voltage to 1.45v in BIOS, but both CPU-Z and Speedfan report my CPU Voltage at 1.41v at boot up, eventually dropping to 1.38v. Is this normal? I turned off all the options I could find in the BIOS which indicated any type of throttling to me (I believe three different options).
Click to expand...

Another thing that looks very similar to my board. The drop on the vcore under load is massive. If I set my vcore to 1.475 in the Bios I get 1.424 in idle in CPU-Z which then drops down to 1.39 under load. There is a pencil vmod for the rev1.0 boards that stabilizes the vcore but it wont work on the rev 3.3 boards.
 
svenski said:
Another thing that looks very similar to my board. The drop on the vcore under load is massive. If I set my vcore to 1.475 in the Bios I get 1.424 in idle in CPU-Z which then drops down to 1.39 under load. There is a pencil vmod for the rev1.0 boards that stabilizes the vcore but it wont work on the rev 3.3 boards.
Click to expand...

I disagree, that is a 3.45% delta at idle and 5.76% delta at load from a voltage regulator circuit that is tying to maintain a voltage over a load of 0 to 30+ amps and that load can change instantly (almost).

Here is a good write up of why Vdroop exists, and in moderation is a good thing.

The referenced Intel document shows that to be in conformance with Intel approvals Vdroop must be designed into the VRD circuit.
I highly suggest reading the Intel VRD spec as well as the link for more insight as to why things are as they are.


http://www.thetechrepository.com/showthread.php?t=126
 
Bill, is there anything you would recommend to show that what I have done has actually sped up my computer?

My OC changes are only seen in my BIOS, they don't show up in Core Temp or CPUZ, and my 3DMark06 Benchmarks have stayed just about identical. I guess I assumed that going from 9x200 to 9x266 was going to show me something a bit more to this point.

I have had no lockups playing games or doing anything else, but I have no idea if my changes have actually taken effect.

Thanks again for your time.
 
Nailz said:
Bill, is there anything you would recommend to show that what I have done has actually sped up my computer?

My OC changes are only seen in my BIOS, they don't show up in Core Temp or CPUZ, and my 3DMark06 Benchmarks have stayed just about identical. I guess I assumed that going from 9x200 to 9x266 was going to show me something a bit more to this point.

I have had no lockups playing games or doing anything else, but I have no idea if my changes have actually taken effect.

Thanks again for your time.
Click to expand...

I'm not Bill, but I'm going to try and answer you question anyway :)

Coretemp should show you overclock results(2400 9x266.67) and if it not.... it's not OCed and in 3dmark06 you should focus on your CPU score result.
 
Can someone please help me? I'm at the Standard High Performance Setup With No Overclock - MB Intelligent Tweaker(M.I.T.) section. I have the DS3 with the latest bios version (F11) and I'm stuck at the "Refresh to ACT Delay" setting. I don't have an option to choose auto because it requires me to enter a value. What should I enter?

:confused: :confused: :confused: :confused: :confused: :confused: :confused: :confused: :confused:

Thanks in advance.
 
GwarGor said:
I'm not Bill, but I'm going to try and answer you question anyway :)

Coretemp should show you overclock results(2400 9x266.67) and if it not.... it's not OCed and in 3dmark06 you should focus on your CPU score result.
Click to expand...

Hey, i'll take an answer from anyone. :)

SO basically what I am hearing is that my BIOS is lying to me and not doing any type of overclock whatsoever? In the past when I have OC'd, if it fails, it automatically changes the settings back to some low level safe default, why wouldn't the gigabyte board do the same?

Or for that matter, what the heck am I doing wrong that it's not OC'ing? Is Vista getting in my way or is it something else?

Thanks!
 
Nailz said:
Hey, i'll take an answer from anyone. :)

SO basically what I am hearing is that my BIOS is lying to me and not doing any type of overclock whatsoever? In the past when I have OC'd, if it fails, it automatically changes the settings back to some low level safe default, why wouldn't the gigabyte board do the same?

Or for that matter, what the heck am I doing wrong that it's not OC'ing? Is Vista getting in my way or is it something else?

Thanks!
Click to expand...

I use Vista 32 bit and have my e6400 OCed to 3gig. I'm not sure why you're OCing isn't successful, but I know that when you try to OC too much you're bios will reset to defaults. I use the F10 BIOS BTW...
 
Hmm, so we're pretty similar in setup then, my E4300 is different, but should still OC, maybe I will try going back to F10 and see how things work out.
 
Nailz said:
Bill, is there anything you would recommend to show that what I have done has actually sped up my computer?

My OC changes are only seen in my BIOS, they don't show up in Core Temp or CPUZ, and my 3DMark06 Benchmarks have stayed just about identical. I guess I assumed that going from 9x200 to 9x266 was going to show me something a bit more to this point.

I have had no lockups playing games or doing anything else, but I have no idea if my changes have actually taken effect.

Thanks again for your time.
Click to expand...

This small program (super pi) should show you the effects of overclocking your processor and does not rely on the video system which is why 3DMark is not showing you much as it is mainly a graphical benchmark (or so I a heard , I havent looked at the new versions, try an older verision 2001 or so. ) Fraps is a nice utility which shows framerates while you play your games. Dont have a link but its easy to find and free. Ray Adam's ATI tray tools is also good if you have an ATI card, shows frame rates and lets you get rid of the bloated control center program ATI sticks us with.

http://en.wikipedia.org/wiki/Super_PI

download from one of the references or google "super pi" get a newer mod. I use mod 1.5 XS of super pi.

btw 1/2 second in run times using 1M places of accuracy is huge.
 
petern said:
Can someone please help me? I'm at the Standard High Performance Setup With No Overclock - MB Intelligent Tweaker(M.I.T.) section. I have the DS3 with the latest bios version (F11) and I'm stuck at the "Refresh to ACT Delay" setting. I don't have an option to choose auto because it requires me to enter a value. What should I enter?

:confused: :confused: :confused: :confused: :confused: :confused: :confused: :confused: :confused:

Thanks in advance.
Click to expand...

I think you found a mistake I made, enter 0 (which is auto) I think I went overboard with the cut and paste. I have to reboot to see and if so I will fix the guide, if I am wrong about being wrong I will be back with a recommendation in a few minutes. My apologies. I screwed up one way or another.

Yep I messed up, should be 0 which is auto (the board will calculate timings. )
 
BillParrish said:
I think you found a mistake I made, enter 0 (which is auto) I think I went overboard with the cut and paste. I have to reboot to see and if so I will fix the guide, if I am wrong about being wrong I will be back with a recommendation in a few minutes. My apologies. I screwed up one way or another.

Yep I messed up, should be 0 which is auto (the board will calculate timings. )
Click to expand...

Thanks for clearing things up. Now another question:

I have the E4300 and when trying to overclock it, I set the voltage to 1.35V. When using CPU-Z it shows that the CPU is at 1.296, meaning that the vdrop is .054, right. Does this mean that I can increase the CPU an additional .054 to make it 1.35V without heating or stressing it out?
 
petern said:
Thanks for clearing things up. Now another question:

I have the E4300 and when trying to overclock it, I set the voltage to 1.35V. When using CPU-Z it shows that the CPU is at 1.296, meaning that the vdrop is .054, right. Does this mean that I can increase the CPU an additional .054 to make it 1.35V without heating or stressing it out?
Click to expand...


yes sorta kinda, I also show about a .05V droop but the question is what exactly (where) is cpuID getting the voltage from. Frankly I havent investigated that and dont know. The good news is a setting of anything 1.4V or below and you are nowhere near the absolute max of 1.55V per Intel data sheet and are at values I belevie will not shorten the life etc. etc. . Just keep an eye your temps.
 
Welp, for whatever reason, going to F10 helped me out. I'm now sitting at 9x300 and running Orthos for about an hour now and things seem to be running fine.

I have to say it is a trip to see CPUZ jump up to my OC settings once I stress it out a bit.

At any rate, things seem to be running faster now, thanks for the assistance!
 
I put together my new system a couple of weekends ago. The good news is that it posted and the XP Pro install was smooth. System specs are in my sig. My cpu is an E6420.

I followed the guide on the first page of the thread to tweak my settings so that I was running my RAM at 2.2v (upped it from the default of 1.8) and adjusted my timings to 4-4-3-8 (again, manufacturer's specs). I saved this as my baseline.

I have a mild overclock at 333MHz FSB. After 2 minutes on Orthos, I'm at 37C/40C according to CoreTemps. I didn't up the FSB any further, as I wanted to check out the limits of my RAM.

I started to explore my RAM limitations as per the instructions here: CPU clock ratio to 6, cpu host frequency to 293, system memory multiplier to 3, the DRAM timing to Auto, and the DDR voltage to 1.09 (.01 over the default setting). With patience in hand, I started to raise the cpu host frequency in 3 MHz increments, booting into Memtest for 3 tests.

I stopped at 350 because I was running out of spare time. However, I have some questions now about the process and what I'm looking for.

General OC noobie questions

1) According to CPU-Z at my baseline BIOS, PC2-6400 has a max bandwidth of 400 MHz, with timings of 5-5-5-15 for 400MHz. What does this mean for me in overclocking? That I shouldn't clock past 400MHz? Or that increases need to coincide with relaxing the timing?

2) I notice that my voltage fluctuates between 1.312 and 1.328. Should I be concerned?

3) CoreTemps says 24/28 at baseline idle while TAT says 25/30. Which is right, and is the 4-5 degree difference between cores acceptable or should I reseat my heatsink?

4) At baseline, 2 minutes of Orthos gives me 35/37C. Are these numbers acceptable for an aftermarket heatsink? I should mention that if I raise the FSB to 333 and run orthos, I idle at 24/28 while 2 mins under Orthos gives me 37/40.

Memory Testing Questions

5) I assume that the purpose of the memory testing, using the lower RAM voltages while increasing the system memory multiplier and the cpu host frequency is to test the memory for its stability. Under these circumstances, it's more important for the computer to boot into Memtest until errors show up than it is for the computer to be able to boot smoothly into Windows. For example, I ran out of testing time after increasing the frequency to 350. For fun, I tried to boot into Windows and ran into a boot cycle til getting the message that Windows couldn't load, and options on how to try to load Windows (last known good configuration, normally, etc.). I'd wanted to run the system under Orthos for a couple of minutes, or at least to look at the information under cpuid. Instead, I rebooted into the bios and loaded my baseline again. Should I still try to boot into Windows at periodic intervals during memory testing? Or just wait til errors show up?

6) Since I interrupted my memory testing, can I try to reload the testing BIOS or should I again work my way up incrementally? In other words, is it better always to oc a component incrementally or is it acceptable to load the last known stable configuration?

7) I plan to work my way through the memory testing with the different voltages: a minimum for the RAM as well as at the maximum voltage for the RAM as well as the stages in between. This is the test data to learn my limitations for my memory. Once I have this data, then the process of the actual overclock is to figure out the sweet spot with respect to the different variables recorded? Is there a standard rubric of when to mess with fsb, when to try to underclock the clock ratio, and when to increase/decrease the DDR voltage?

Thanks a lot for the guide. If nothing else, I'm learning a lot more about my BIOS. The ability to save portraits absolutely rocks!

--ceolstan
 
All good questions and thank you for taking the time to go thru the process, it is tedious but at the end you will have some hard data on your memory that will ensure you dont end up running around in circles if you attempt a "bleeding edge" overclock and in general will know exactly what your memory is capable of. It is of course not really necessary to do the whole bit if you can find out what you need to know to hit some particular OCing target.

1) According to CPU-Z at my baseline BIOS, PC2-6400 has a max bandwidth of 400 MHz, with timings of 5-5-5-15 for 400MHz. What does this mean for me in overclocking? That I shouldn't clock past 400MHz? Or that increases need to coincide with relaxing the timing?
Click to expand...

cpu-z is reporting the contents of your SPD chip on the ram which is programmed at the factory to plain jane (JEDEC) timings to ensure (we hope) that the ram will work (POST) with any motherboard. These are just generic starting values and with OCing memory you are expected to manually adjust the timings etc. in the BIOS per the manuf published specs. Now those published specs should be good at (assuming a memory divider of 2:1) 400MHz FSB (800MHz memory speed) or else I would return the memory. IN my experience even with the tighter manuf tolerances you can typically get to 430-440MHz FSB (860-880 MHz memory speed) before you start seeing errors in memtest.
What this means is that with an cpu mulitplier of 8 x 400MHz you should be able to OC to 3.2GHz with your memory running at manuf tight timings and with a little headroom for about 40MHz more FSB before you have to consider loosening your memory timings due to your ram causing errors. If your CPU will go higher you then loosen the timings to 5 5 5 15 or whatever which in the case of my memory is good for FSB in excess of 480MHz (960MHz memory speed). Being stuck with a cpu with a low (7) mulitplier like I am I have to run high FSB to get the CPU overclock, 8 x 400 or a little higher is nice in that your memory timings are tight and you are getting full bandwidth, and maybe a bit more, out of your memory and 3.2GHz is a very respectable OC. (Would probally have to up the Vcore a bit to get there. )

2) I notice that my voltage fluctuates between 1.312 and 1.328. Should I be concerned?
Click to expand...
Short answer, no not a bit.

3) CoreTemps says 24/28 at baseline idle while TAT says 25/30. Which is right, and is the 4-5 degree difference between cores acceptable or should I reseat my heatsink?
Click to expand...

Close enough not to worry, but as TAT is old and does not know about your processor I would just use it for stressing (which it is the best at, if you want to heat her up, fire up TAT's stress test mode) and use Coretemp for the actual temp readings. Its hard to say the 5 degree difference is more than I am used to seeing in other posts/reports but it could be some software running on one core and not the other. I would just be absolutely sure the heatsink was installed flat and uniformly tight (tighten bolts etc in an "X" pattern) and keep an eye on it. It does warrant further investigation and perhaps the IHS (heatspreader) of the CPU could use a lapp job but I would first get my OC set up (if I was postive about the HS) then go and try to figure out if its an issue or not. Little stumped on this one, I see a 2C difference all the time, 5C seems high but not out of reason.

4) At baseline, 2 minutes of Orthos gives me 35/37C. Are these numbers acceptable for an aftermarket heatsink? I should mention that if I raise the FSB to 333 and run orthos, I idle at 24/28 while 2 mins under Orthos gives me 37/40.
Click to expand...

The cpu contains an internal temp trip that you cannot disable. You cannot destroy this processor by burning it up, it wont let you. Those numbers are extreamly good. I dont even think twice until my machine starts showing numbers in the mid 60's and I set my temp alarm in the bios to 70C most of the time, still leaving me a huge margin of safety as the typical thermal trip temp is 85C. The temps you report are good for that mild OC and stock or close to stock Vcore. Hmm guess I sould mention that if you are in a room where the AC is dumping out ice cubes that temp is just OK, if you are in a sweatbox they are exceptional. Room ambient temp is a major factor.
 
Thanks Bill.

As I mentioned, time is something of a premium, so I probably can't complete any single one of the memory passes in a day. However, I keep two Testing profiles in the BIOS: one for the current oc and one for the past (error-free) oc. I'll just load up the last known good test and continue from there.

I'm keeping myself to the discipline of the 3MHz increments, with the hope that if Memtest starts to show errors, I'll merely have to reload the last good test settings rather than clear the CMOS. Hmmm...That reminds me. I need to go connect the Reset button to the clear CMOS jumper--just in case, you understand. ;)

I wanted to take time with this build to understand my system better and to test its limits more carefully. My goal is to achieve a stable overclock of between 3-3.5GHz that I can use while gaming.

--ceolstan
 
I have some questions about my overclocking attempts with my E4300 and Gigabyte DS3 rev 3.3 board.

I'm currently overclocked to 3.0GHz via 9x multiplier and 335 FSB. My memory is running at 891MHz with the 2.66 multiplier.

I'm content with both the speed of my memory and CPU.

However I'm wondering if I should be striving to lower my CPU multiplier, raise my FSB, and use a 2.0 memory multiplier to run my memory synchronously?

The best way on paper I found to achieve this is with a 7x CPU multiplier and a 430MHz FSB. That would give me about the same CPU and memory speeds with the memory running synchronously.

Problem is I can't seem to get above 380FSB. Seems like I read that these boards are capable of much higher FSBs. So I'm trying to figure out if I've done something wrong.

I have a 50mm fan on my MCH (or NB).
I disabled C1E, EIST, Limit CPUID max to 3, and CIA2.
PCIE has been tried from 100 to 110.
PCIE, MCH, and FSB voltages are all +0.1.
Using the F10 BIOS.

I've even tried a 6 CPU multiplier so that the CPU speed is well below what it is capable of and memory with looser timings just try eliminating these components as problems.

Are there an other tricks that I'm missing to reach a higher FSB speed? Or should I even care of I'm running my memory asynchronously? Thanks.
 
However I'm wondering if I should be striving to lower my CPU multiplier, raise my FSB, and use a 2.0 memory multiplier to run my memory synchronously?
Click to expand...

In a perfect world, yes. running the memory 2:1 eliminates buffers and wait states that slow down the transfer of data. However if you cannot acheive the "sweet spot" of 8 x 400 using a divider to get more bandwidth out of your memory will/should overcome the the drawbacks of running non-synch, just not as much as you would think because of the buffering/waits.

You may be stuck but there is one thing that comes to mind. The infamous FSB "wall" or hole. The northbridge is also directly coupled to the FSB and as you raise the FSB you overclock the NB just like you are OCing the CPU. It has been widely reported (verification is a bit sketchy however) that around 360-375 FSB the NB becomes OCed so much it craps out. The good news is that it is beleived that around 400MHZ the NB will "change straps" (a divider kicks in) so that it runs at somewhat looser timings but can support the higher FSB. Reduce you mulit to 6 and loosen memory timings and set to 2:1, give it an extra notch of MCH voltage, and see if you can get it to boot at 405 FSB (we are tying to jump over the wall and get to the other side). Be prepared to have to reset/clear the bios if this does not work.
 
You must log in or register to reply here.
Back
Top