SOLVED i7 950 OC: fails to pass the LinX/IBT max tests at any speed

Hi All,

Here is a new member with a very frustrating problem. I sincerely hope that you may be able to help me . I am an experienced OCer from the past with both Intel and AMD platforms but this is the first time I am trying to OC an i7 system. I have read and re-read the excellent 3-step OC guide of MiahAllen, not once or twice but perhaps 10 times. I have then read every related thread in these forums (and in some others too), in the hope that I'd be able to solve my problem but alas! I have now been spending the past 8 days 24/7 with my PC, living like a zombie, tweaking something, executing tests, failing, re-tweaking and repeating the whole cycle many times over. I have executed the 3-steps at least 6 times from A to Z, choosing slightly different goals each time (all aimed at getting the CPU run at around 4GHz by using various multiplier and BCLK combinations). I wasn't even able to get it pass the max tests for any of those combinations. The system works seemingly stable even at 4.1 GHz (which is the maximum I have pushed it to). I get no BSODs or lock-ups or other problems. I can run all my programs without any problems. My max temp does not exceed 80 degrees Centigrade under 100% LinPac64k load. I use LinX 0.6.4, OCCT and IBT 2.4/2.5 for testing although LinX gradually became my primary test program as it issues an error most quickly out of those three. I can run any test successfully as long as I am not using the max memory. But the max memory tests using 8 threads almost always fail. In those rare occasions when they pass the max mem test, I move on to the next step but when it fails and when I go back to the latest stable situation, the tests usually fail again although they have been successful before using the same settings. So things do not seem to be repeateble. As you can imagine, this is extremely frustrating.

Here is what my system looks like:
CPU: i7 950
MoBo: GigaByte X58A-UD3R Rev.02 Bios: FB
RAM: OCZ Platinum DDR3 OCZ3P1333LV6G: 6x2GB, rated as 7-7-7-20@1333 according to the OCZ specs
Cooler: Noctua NH-D14 air cooler (using 2 fans in push/pull)
OS: Windows 7 Ultimate 64-bit

Right now, I have been trying the step 1 for the millionth time for these goals:
BCLK: 183 MHz
CPU: x22 @4 GHz (HT: On)
RAM: x8 @ 1464 MHz @ 8-8-8-24 (9-9-9-27 would be acceptable too)
PCI-E: 100 MHz

My initial settings in the BIOS for step 1 are:
CPU multi: x15 @2.4 GHz (HT: On)
DRAM: x6 @973 MHz, timings 8-8-8-24
BCLK: 163 (first try to establish a base QPI/Vtt)
CPU/IOH Clock Drive: 800 mV
CPU PLL: 1.8 V
QPI/Vtt: 1.215 initially (gradually increased up to 1.415 w/o any success)
CPU Vcore: Normal (1.072 V: please note that the CPU is underclocked @ 2.4GHz, that is why this is so low)
DVID: +0.05 V
LLC: level 2
DRAM Voltage: 1.66 V
All other DRAM settings: Auto
PCIE: 1.5 V
IOH Core: 1.28 V
QPI PLL: 1.1 V
ICH I/O: 1.5 V
ICH Core: 1.1 V

I have started testing this with Vtt @ 1.215V and increased till 1.415V. All the LinX max tests (i.e. problem size 36585 MiB: 10240) have failed at the first step. The message it issues is: stopped upon error after n mins:secs.

I am really desperate now. I know for a fact that I have all the right components and the system should have no problems whatsoever with being OCed at 4 GHz. Actually, it seems to work w/o any problems under normal circumstances. So why I cannot make it get to work also using the max tests? All your ideas or suggestions will be greatly appreciated. Thanks.

PS: I have just executed the final test cycle using QPI/Vtt 1.435V for BCLK 163 and the LinX max tests are now green. But I find this QPI/Vtt for such a low value of BCLK to be way too high since Miah Allen also indicates that the max Vtt for my CPU should be 1.45 V. What do you reckon?

Have you actually tired increasing the Vcore at all? I would say you may need more than stock for 4GHz. See if that helps make things stable, then you might not need such a high Vtt...

Badbonji said:

Have you actually tired increasing the Vcore at all? I would say you may need more than stock for 4GHz. See if that helps make things stable, then you might not need such a high Vtt...

Click to expand...

Hi Badbonji, thanks for chiming in. Yes, during many tests I have also played with various Vcore values. In a few series, I have increased the DVID up till +0.2V. In another, I have set the LLC to standard (i.e. off) and set the Vcore manually to various values between 1.2 and 1.46V. Even when the V core was above 1.35V the QPI/Vtt needed was still more than 1.35V. I really do not understand why I seem to be needing these high voltages at all.

I just built a i7 950 rig myself your vcore is 1.072 and its been kept there? Seems low,that might be your problem there. I was having a heck of a time getting above 4ghz stable and what finally realllly helped me was keeping close tabs on my vdroop. im working on the rampage 3 formula so the options you have may not be named the same. Basically i need 1.36 voltages to hit 4.1ghz and this puts my linpack temps at 90ish max. Thats going to be has high as i want to keep it because even if itll never get that hot real world usage, if it gets much hotter i may just get bsod from heat regardless of potential "stability" while im running stability tests. So to get that 1.36ish vcore to get where it needs to be and stay I have enabled 100% load line calibration and inputed 1.35 vcore into the bios (100% "spikes" up .01 on my system). I tried 50% load line calibration and it required me to put 1.40 into the bios to watch the vcore droop to 1.35 before freezing to a bsod)

This is all at 1.275 qpi/dram and that may be adjusted as iv lowered my bclock recently trying to shoot for decent 24/7 usage temps. Im currently at 175x23 and those i left those same voltages for now my max temps are in the high 80's, my only beef on that bclock is my memory is supremely underclocked.

All that being said your chip may require slightly different values to get where you want(or extremely different values as you start to bang your head against the expontentially growing heat/ required voltage wall, so know that if i lower my clock below 4ghz it starts requiring ALOT less vcore, Im definately feeling like iv been banging my head against an expontentially growing vcore/heat wall once i pass 4ghz. 4.2 has been my max at 1.38 (1.39 after 100%calibration) but thats just too hot for me 24/7.

Anywhoot hope something/anything i said made any sense and hopefully helped you. Try 1.25 qpi and 1.2 vcore, drop the multi to closer to stock speed and then monitor/take notes of your vcore and qpi droop and compensate, very minimal adjustments should be needing until you start knocking on the upper end of 3ghz/lower end of 4ghz depending on the chip.

TL;DR

up your vcore, watch your vdroop, yup

added in * It isnt stable with higher vcore at 4ghz or a lower setting? like closer to stock speed

added in again* my mobo has a qpi LLC jumper which i have enabled, you may need to input a high qpi value just be sure to watch it under full load to see where it really sits.

jmccann said:

I just built a i7 950 rig myself your vcore is 1.072 and its been kept there? Seems low,that might be your problem there. I was having a heck of a time getting above 4ghz stable and what finally realllly helped me was keeping close tabs on my vdroop. im working on the rampage 3 formula so the options you have may not be named the same. Basically i need 1.36 voltages to hit 4.1ghz and this puts my linpack temps at 90ish max. Thats going to be has high as i want to keep it because even if itll never get that hot real world usage, if it gets much hotter i may just get bsod from heat regardless of potential "stability" while im running stability tests. So to get that 1.36ish vcore to get where it needs to be and stay I have enabled 100% load line calibration and inputed 1.35 vcore into the bios (100% "spikes" up .01 on my system). I tried 50% load line calibration and it required me to put 1.40 into the bios to watch the vcore droop to 1.35 before freezing to a bsod)

This is all at 1.275 qpi/dram and that may be adjusted as iv lowered my bclock recently trying to shoot for decent 24/7 usage temps. Im currently at 175x23 and those i left those same voltages for now my max temps are in the high 80's, my only beef on that bclock is my memory is supremely underclocked.

All that being said your chip may require slightly different values to get where you want(or extremely different values as you start to bang your head against the expontentially growing heat/ required voltage wall, so know that if i lower my clock below 4ghz it starts requiring ALOT less vcore, Im definately feeling like iv been banging my head against an expontentially growing vcore/heat wall once i pass 4ghz. 4.2 has been my max at 1.38 (1.39 after 100%calibration) but thats just too hot for me 24/7.

Anywhoot hope something/anything i said made any sense and hopefully helped you. Try 1.25 qpi and 1.2 vcore, drop the multi to closer to stock speed and then monitor/take notes of your vcore and qpi droop and compensate, very minimal adjustments should be needing until you start knocking on the upper end of 3ghz/lower end of 4ghz depending on the chip.

TL;DR

up your vcore, watch your vdroop, yup

added in * It isnt stable with higher vcore at 4ghz or a lower setting? like closer to stock speed

Click to expand...

Hi JMcCann,

Thanks for the kind response. To prevent a misunderstanding from spreading any further, the Vcore of 1.072V I have mentioned is when the CPU is running at 15x163; i.e. 2.44 GHz, which is way below its stock frequency. When I push the CPU to 4 GHz, which I have done using various combinations, the Vcore needed varied between 1.35V to 1.46V.

This is a new PC I have built and I will be running a lot of very expensive programs on it such as Photoshop CS5, Lightroom, DxO, etc. All photography related apps which need to be activated on-line and any change in the PC config can void the activation. So I want to make 100% sure of the final OC config before committing to the activation process. The PC runs seemingly stable but I really want to make sure by passing the max LinPack64 tests using 8 threads.

You are right about the virtues of keeping the BCLK as low as possible, which to my opinion causes less heat/stress on the system components outside of the CPU.

Thanks again.

hmm, with those voltages id check into the batch number and VID and see if maybe your stuck running higher volts and hotter to get your 4 ghz than others. I know I felt like I should be running faster and cooler when i got my stuff in but alas I has no golden chip. batch #3020A953.

edited in* dang i need to pay more attention. You get linx errors at the higher bclocks regardless of multi? Or only when you pushing 4ghz?

jmccann said:

hmm, with those voltages id check into the batch number and VID and see if maybe your stuck running higher volts and hotter to get your 4 ghz than others. I know I felt like I should be running faster and cooler when i got my stuff in but alas I has no golden chip. batch #3020A953.

edited in* dang i need to pay more attention. You get linx errors at the higher bclocks regardless of multi? Or only when you pushing 4ghz?

Click to expand...

I get LinX errors at higher BCLKs (i.e. higher than 175) regardless of the multi indeed. My batch # is 3949B121 but the only google hit is from an Italian OC site .

First off Cem... to OCForums....I'm glad you like my guide

Cem said:

Right now, I have been trying the step 1 for the millionth time for these goals:
BCLK: 183 MHz
CPU: x22 @4 GHz (HT: On)
RAM: x8 @ 1464 MHz @ 8-8-8-24 (9-9-9-27 would be acceptable too)
PCI-E: 100 MHz...

....................

...I have started testing this with Vtt @ 1.215V and increased till 1.415V. All the LinX max tests (i.e. problem size 36585 MiB: 10240) have failed at the first step. The message it issues is: stopped upon error after n mins:secs.

Click to expand...

Based on these statements, let me first say, that you should not be increasing the memory multiplier in step 1. Step 1 is not so much about finding the max bclock as it is about just getting a feel for the bclock overclocking process....and alos looking for "holes". This makes step 2 a bit easier, and you really start to push the IMC.

Which brings me to your second point.....the IMC and your 12GB of memory. Is that 3x4GB or 6x2GB (just curious)? Either way, the IMC has to work pretty hard and explains why you're needing a bit more QPI/VTT/IMC voltage than average. Bloomfield IMCs have a large range of capability....and some do great with large memory pools, others do not. I'm working with a new chip right now that seems to have trouble with 12GB. Did you already test the memory with memtest86+ ?

The other thing I noticed is that your primary goal is with the x22 multi, and I'd suggest sticking to x21 when you get back to step 3.....Bloomfield doesn't seem to like even multipliers much

Good luck and keep us posted

Cem said:

....
My initial settings in the BIOS for step 1 are:
CPU multi: x15 @2.4 GHz (HT: On)
DRAM: x6 @973 MHz, timings 8-8-8-24
...

Click to expand...

MIAHALLEN said:

First off Cem... to OCForums....I'm glad you like my guide

Based on these statements, let me first say, that you should not be increasing the memory multiplier in step 1. Step 1 is not so much about finding the max bclock as it is about just getting a feel for the bclock overclocking process....and alos looking for "holes". This makes step 2 a bit easier, and you really start to push the IMC.

Which brings me to your second point.....the IMC and your 12GB of memory. Is that 3x4GB or 6x2GB (just curious)? Either way, the IMC has to work pretty hard and explains why you're needing a bit more QPI/VTT/IMC voltage than average. Bloomfield IMCs have a large range of capability....and some do great with large memory pools, others do not. I'm working with a new chip right now that seems to have trouble with 12GB. Did you already test the memory with memtest86+ ?

The other thing I noticed is that your primary goal is with the x22 multi, and I'd suggest sticking to x21 when you get back to step 3.....Bloomfield doesn't seem to like even multipliers much

Good luck and keep us posted

Click to expand...

Hi Miah,

Thanks for the warm welcome. Your 3-step guide, it is a great document which must have helped countless OCers out there

As you can see in my self-quote above, I did set the memory multi to x6 in the first step as you've recommended. So it should not be the main culprit. Nevertheless, I have already done 2x full memtest86+ runs with this memory and it has passed, phew .

Based on my various intermediate results (of tests passed and failed), I have come to realize a pattern here. My rig does seem to fail tests with higher than 175 BCLK rates more often than not. So based on that, I have changed my goal set to use BCLK x175 and x23 for the CPU multi. In the meantime, I have started steps 1 and 2 on this goal set. I have just completed step 2 successfully with the following parameters:

CPU multi: x15 @2.625 GHz (HT: On)
DRAM: x8 @1400 MHz, timings 8-8-8-24
BCLK: 175
CPU/IOH Clock Drive: 800 mV
CPU PLL: 1.8 V
QPI/Vtt: 1.315 V
CPU Vcore: Normal
DVID: +0.05 V
LLC: level 2
DRAM Voltage: 1.66 V
All other DRAM settings: Auto
PCIE: 1.5 V
IOH Core: 1.28 V
QPI PLL: 1.2 V
ICH I/O: 1.5 V
ICH Core: 1.2 V

Now I am busy with the step 3 but I am running in to too high Vcore values. Already the Vcore is at 3.8125 V and not yet stable.... Any advice will be highly appreciated.

PS: forgot to answer that I am using 6 x 2GB memory sticks.

bclock holes in the 180-190 range are common....have you tried 195-205 range? Do a quick check with the bclock at 200 and mem at x6 to see

FYI - bclock holes can also chge with different CPU ratios as well.

Help pls: I'm losing my sanity

I have tried every possible combination, but still no solution. I did execute the steps 1 and 2 with BCLK values of 166, 170, 173, 175, 177, 180, 183, 185, 187, 190, 195, 200, 205. Almost all of these BCLK values have passed the max test. Then, when I got to the step 3, I've used various CPU multis such as 19, 20, 21, 22, 23 and even 24. Almost all of them have failed. A couple of them did pass the max tests initally. But when I have restarted the PC or when I did use exactly the same settings after changing to something else, the tests have failed again whereas at first they have been successful. So far, I have not been able to get past 3 GHz in a stable manner! This is so frustrating .

Is it even possible that a CPU cannot be overclocked at all? If not, then why I can't make this work? It is not because I did not try, I have tried all the possible combinations in the past 10 days. What else can one do? Please help me regain my sanity.

Problem solved: DRAM was the problem

Hi folks,

Good news: in the end I have discovered what was going wrong. I did re-run the memtest86+ and it came up with errors all over the memory range. On hindsight I should have done this much earlier but having previously done the memtest w/o any problems, I hadn't bothered repeating it. This is a valuable lesson: check, double check and then triple check, lol.

So I brought back the OCZ Platinum DDR3-2x6GB-1333 7-7-7-20 memory to the store and I bought 6 sticks of Kingston HyperX DDR3 2GB 1600-8-8-8-24 memory. I first have done the memtest86+ and all was fine. After that, the tests which have been failing before have all come up green.

I redid the 3 steps and easily achieved an awesome OC of 4.09 GHz. Fully stable and the maximum temp is around 83C (idle 40C).
Here are my final settings:
BCLK: 195
CPU multi: x21 @4.09 GHz (HT: On)
DRAM: x8 @1600 MHz, timings 8-8-8-24
QPI/Vtt: 1.315 V
CPU Vcore: Normal
DVID: +0.1625 V (idle Vcore=1.328V, 100% load Vcore=1.344V)
LLC: level 2
DRAM Voltage: 1.66 V
IOH Core: 1.2 V
PCIE: 100 MHz
All other settings on default/auto.

So I am extremely relieved and can go back to having a life again. Thanks to all of you who have tried to help, much appreciated.

Cheers.

Awesome! Thanks for reporting back Cem.....I'm glad you got it working

now you can more