My triple-core 720BE unlocked to a quad-core. To verify stability of the 4th core, I primed it for 12 hours at stock, I also ran OCCT Lin pack for 12 hours with no errors. I was following advice, and tutorials on different OC sites, first time here though. I had a thread going in a different forum about my setup, and 1300 views and over 80 post's later, I still cannot get good information. Here is what I'm trying to figure out.
The DDR3 memory controller on the CPU supports 2 sticks of DDR3 memory at slower than 1600MHz, right? I have 4 sticks (2GB each) of DDR3-1600, GSkill recommends that I set the CPU-NB at 1.4 to start, since I'm using 4 sticks of memory that is faster than the supported maximum of the cpu. I've read recommendations to set the NB at 2400MHz, and not go over 1.3v, and I had someone tell me last night to keep the CPU-NB voltage the same as the vcore once the vcore passes 1.4v
It completed 4 passes of memtest86 v4 at 1600, with a CPU-NB voltage of 1.3, but that is also at stock CPU-NB frequencies, 2000MHz.
I can boot this thing up to 3.3GHz on stock voltages (only cpu multi increases). A stress test (Either OCCT Lin pack, and Prime95, I use both, sometimes simultaneously), causes an instant-reboot at 3.3GHz stock vcore, but I can boot up and get into windows.
Once I get back in BIOS, I push the vcore from 1.325, to 1.350, and OCCT and Prime both detect errors. During this phase of voltage increases and stability test's, I noticed that Prime kept erroring on the same thread, the third one. I read that was the whole purpose for Advanced Clock Calibration, was so that you can apply finer tuning on a per-core basis, such as adding positive percentage to a 'weaker' core, and allowing the weaker core to achieve higher clock speeds than without the ACC.
So after several failures on the same thread in Prime95, I decided to go with the theory that maybe this thread was the same processing core that came disabled (for some reason), and perhaps it was weaker and needed some ACC in it's life. So going with a recommendation straight from AMD, I applied a +2% to the corresponding core that kept erring in Prime95. It blew my mind when it worked, that thread no longer errors, and while watching the display during a stress test, that same thread is now noticeably quicker at completing iterations then the other 3 cores. Even when that core is clocked a .5 multi slower than the other cores. Ex: 18 18 17.5 18
I kept pushing vcore until I got to 1.375, at which I am OCCT and prime stable. At this point, I figured I can continue upping the multi. I do so a .5 multi at a time. It seems like from this point, each multi increase needs two sequential vcore bumps. Example. 1.375 is needed for 3.3GHz, well 3.4GHz needs 1.425. My BIOS allows voltage increases as such: 1.325, 1.350, 1.375, 1.400, 1.425, 1.450
It appears that for each multi increase, I need two increases in vcore. So for 3.4GHz, I needed 1.425. And to get it stable at 3.5GHz, I need 1.475, although I cannot say 'stable'. It will pass 10, 20 and even 30 minutes of Prime and OCCT Lin pack at 3.5 1.475, and just when you start getting excited about finding another stable point on your quest upward, BAM, Emeril steps in and throws something in the mix and an instant-reboot occurs. No 'error detected' with a duck quacking at you, and no Rounding was 0.5, expected 0.4. None of that, flat out pixelation of the screen and instant-reboot. No blue-screen, no memory dump, just a pretty blank screen displaying a deep, true nothing.
What the heck should I be doing with the CPU-NB at this point? I feel like that is what's holding me back. I've tried keeping everything else stock, I've tried with the CPU NB at 2400, 2600. I've used CPU-NB voltages from 1.3-1.5 (following recommendations in other threads I found saying to keep the NB voltage the same as the vcore, crazy, but I tried it) When I play with the CPU-NB settings, I don't get instant reboots, instead I just get error's during stability tests.
Seems like progress, to go from instant reboots to blue screens and recoverable error's. I've read the guide in this forum, and what I get from it, my CPU-NB frequency should be somewhere between 2200-2400, for a 3.4-3.5GHz core speed. Coupled with fact that I'm running 4 sticks of 1600, does my CPU-NB voltage need serious attention? Should I cut the Memory back to the next slowest setting (which is the supported value of the CPU) and tighten up the timings a bit (i can go down to 8-8-8-24 at 685MHz), so that the CPU-NB wont have such a hard time handling all 4 sticks? It's likely that I might get the recommendation to remove two of the sticks of memory, but that's sorta out of the question at this point, my Ziggy Dark Knight does a good job of making sure you cant switch out modules with the HSF mounted, and I dint know if I would want to chance a re-seat of the HSF, I have unbelievably sweet temps with this thing. A test run at 3.6GHz @ 1.5 vcore and my core temps never went above 48c. The test did crash eventually. Not errored, but crashed.
I reiterate the part where I make the distinction between an error and an instant-reboot. I feel that information may be important in diagnosing which component needs attention.
I was 1 hour Prime stable at 3.6 GHz but that was only 3 cores and it needed more vcore then what I need on all 4 cores, but here is a snapshot of my settings at that time:
Base: 206
vCore: 1.4875
DRAM: 1.5
CPU/NB: 1.3875
I stopped it after an hour, excited about finding a stability. But I became worried when reading post's referencing a similar setup stating that my vcore was way too high for that speed on only 3 cores, and that something else was wrong. They also told me that if the 4th core unlocked, and that if it was stable at stock, then the core was most probably good and that I had a good processor. I'm kind of believing it, and I can see that it's a beast, just can't seem to find a sweet spot, and now I'm looking for recommendations or guidance.
Thanks in advance
The DDR3 memory controller on the CPU supports 2 sticks of DDR3 memory at slower than 1600MHz, right? I have 4 sticks (2GB each) of DDR3-1600, GSkill recommends that I set the CPU-NB at 1.4 to start, since I'm using 4 sticks of memory that is faster than the supported maximum of the cpu. I've read recommendations to set the NB at 2400MHz, and not go over 1.3v, and I had someone tell me last night to keep the CPU-NB voltage the same as the vcore once the vcore passes 1.4v
It completed 4 passes of memtest86 v4 at 1600, with a CPU-NB voltage of 1.3, but that is also at stock CPU-NB frequencies, 2000MHz.
I can boot this thing up to 3.3GHz on stock voltages (only cpu multi increases). A stress test (Either OCCT Lin pack, and Prime95, I use both, sometimes simultaneously), causes an instant-reboot at 3.3GHz stock vcore, but I can boot up and get into windows.
Once I get back in BIOS, I push the vcore from 1.325, to 1.350, and OCCT and Prime both detect errors. During this phase of voltage increases and stability test's, I noticed that Prime kept erroring on the same thread, the third one. I read that was the whole purpose for Advanced Clock Calibration, was so that you can apply finer tuning on a per-core basis, such as adding positive percentage to a 'weaker' core, and allowing the weaker core to achieve higher clock speeds than without the ACC.
So after several failures on the same thread in Prime95, I decided to go with the theory that maybe this thread was the same processing core that came disabled (for some reason), and perhaps it was weaker and needed some ACC in it's life. So going with a recommendation straight from AMD, I applied a +2% to the corresponding core that kept erring in Prime95. It blew my mind when it worked, that thread no longer errors, and while watching the display during a stress test, that same thread is now noticeably quicker at completing iterations then the other 3 cores. Even when that core is clocked a .5 multi slower than the other cores. Ex: 18 18 17.5 18
I kept pushing vcore until I got to 1.375, at which I am OCCT and prime stable. At this point, I figured I can continue upping the multi. I do so a .5 multi at a time. It seems like from this point, each multi increase needs two sequential vcore bumps. Example. 1.375 is needed for 3.3GHz, well 3.4GHz needs 1.425. My BIOS allows voltage increases as such: 1.325, 1.350, 1.375, 1.400, 1.425, 1.450
It appears that for each multi increase, I need two increases in vcore. So for 3.4GHz, I needed 1.425. And to get it stable at 3.5GHz, I need 1.475, although I cannot say 'stable'. It will pass 10, 20 and even 30 minutes of Prime and OCCT Lin pack at 3.5 1.475, and just when you start getting excited about finding another stable point on your quest upward, BAM, Emeril steps in and throws something in the mix and an instant-reboot occurs. No 'error detected' with a duck quacking at you, and no Rounding was 0.5, expected 0.4. None of that, flat out pixelation of the screen and instant-reboot. No blue-screen, no memory dump, just a pretty blank screen displaying a deep, true nothing.
What the heck should I be doing with the CPU-NB at this point? I feel like that is what's holding me back. I've tried keeping everything else stock, I've tried with the CPU NB at 2400, 2600. I've used CPU-NB voltages from 1.3-1.5 (following recommendations in other threads I found saying to keep the NB voltage the same as the vcore, crazy, but I tried it) When I play with the CPU-NB settings, I don't get instant reboots, instead I just get error's during stability tests.
Seems like progress, to go from instant reboots to blue screens and recoverable error's. I've read the guide in this forum, and what I get from it, my CPU-NB frequency should be somewhere between 2200-2400, for a 3.4-3.5GHz core speed. Coupled with fact that I'm running 4 sticks of 1600, does my CPU-NB voltage need serious attention? Should I cut the Memory back to the next slowest setting (which is the supported value of the CPU) and tighten up the timings a bit (i can go down to 8-8-8-24 at 685MHz), so that the CPU-NB wont have such a hard time handling all 4 sticks? It's likely that I might get the recommendation to remove two of the sticks of memory, but that's sorta out of the question at this point, my Ziggy Dark Knight does a good job of making sure you cant switch out modules with the HSF mounted, and I dint know if I would want to chance a re-seat of the HSF, I have unbelievably sweet temps with this thing. A test run at 3.6GHz @ 1.5 vcore and my core temps never went above 48c. The test did crash eventually. Not errored, but crashed.
I reiterate the part where I make the distinction between an error and an instant-reboot. I feel that information may be important in diagnosing which component needs attention.
I was 1 hour Prime stable at 3.6 GHz but that was only 3 cores and it needed more vcore then what I need on all 4 cores, but here is a snapshot of my settings at that time:
Base: 206
vCore: 1.4875
DRAM: 1.5
CPU/NB: 1.3875
I stopped it after an hour, excited about finding a stability. But I became worried when reading post's referencing a similar setup stating that my vcore was way too high for that speed on only 3 cores, and that something else was wrong. They also told me that if the 4th core unlocked, and that if it was stable at stock, then the core was most probably good and that I had a good processor. I'm kind of believing it, and I can see that it's a beast, just can't seem to find a sweet spot, and now I'm looking for recommendations or guidance.
Thanks in advance