Rock solid 4.7GHz 4770K OC - Until I Enable Adaptive Voltage

[Hurin]

EDIT: This has turned into more of a standard OCing thread because it turns out I wasn't as stable as I thought with manual voltages.

Hi All,

The Problem: My 4.7GHz OC is rock solid with manual voltage set. But it crashes when I try to migrate those settings into an adaptive voltage environment.

Please see sig for full system specs.

Rock Solid - Manual Voltages
Hyperthreading OFF
4.7GHz (47 multi)
Cache 4.4GHz (44 multi)
vcore = 1.340
Cache = 1.300v
input = 1.8v
Passed Prime 95 24 Hours Torture Test with 512 minimum FFT while simultaneously looping Valley Benchmark. Max temp is 86c.

Not Stable
When I enable Adaptive voltage, like so. . .
Vcore "Additional Turbo Mode CPU core voltage" = 1.340v
Offset = +.001v (also tried auto)
Cach "Additional Turbo Mode Cache voltage" = 1.300v
Offset = +.001v (also tried auto)
input = 1.8v (tried 1.9v too)

. . . Prime95 set exactly as above fails within five minutes despite temps not getting much worse (88c max, again my min FFT is 512k). Sometimes with a reported error and sometimes with "this program needs to close". . .

Also tried setting cache back to auto multiplier so it would (I think) go to 3.9GHz. No effect.

Any advice on dialing in adaptive voltages? I would prefer not to have 1.34v going through my CPU at all times if it can be avoided.

--H

 

[Justintoxicated]

I think that is just the way it goes man. Make sure you test prime with v28.1 since it is made for Haswell and is much more stressful.

http://www.sendspace.com/file/mtfcor

That's a great overclock though. Mine took 1.4v to do 2700 and I'm not sure how stable since I only tested with v27.9.

I'll let others chime in about adaptive voltage as it sounds like something I might want to play with eventually as well.

 

[Hurin]

Sigh. . . I'm hoping I don't have to live without adaptive voltage. The more I see people trying to stay below 1.3 vcore, the more I'm second-guessing my decision to go with 1.34v and 4.7GHz. Especially if it's going to be coursing through my CPU 24/7 with only a Corsair H100i to protect its poor little circuits (heh). Yet, I can't really spend the time now to properly stress test and validate a more conservative OC.

If anyone has any particular insight into why these voltages work at manual but fail with adaptive voltage enabled, please. . . share your wisdom! =)

 

[trick0502]

What cooling are you using?

 

[Hurin]

What cooling are you using?

Corsair H100i. Keeps it about 85-88c while running Prime95 with Min FFT set to 512 (it will throttle and get out of hand if I go small FFT -- around 8k-16k).

But that seems par for the course with Haswell under just about any cooling save exotic water cooling and/or de-lidded like Justin above.

--H

 

[Hurin]

Make sure you test prime with v28.1 since it is made for Haswell and is much more stressful.

I think you may have just helped me quite a bit. I hadn't bought into the hype about v28.1 vs 27.9. But I just tried my manual voltage and 512k fft prime settings on v28.1 and sure enough, I'm either getting reported hardware errors or the computer is freezing. So it's likely weeding out some instability that the prior Prime95 was missing. Hopefully this will be the new tool I need to get this straightened out and get adaptive voltage working.

Thanks! Now I just hope I don't have to up the vcore much more. . . or drop down to a more conservative multiplier. And I'm doing this on my full install of Windows. . . yikes (would rather do it on a testing drive so I don't corrupt anything important).

 

[Hurin]

Following up. . .

BSOD even at vcore of 1.350 and input of 1.95.

First things first, time to get everything back to stock speeds and make sure I have a stable system according to p95 v28.1 at stock speeds. Otherwise, without that known for sure, I'm possibly chasing my tail.

Incidentally, virustotal shows the zip file for v28.1 of prime95 as malware (though only via one av engine). Yet, when you upload the actual EXE it contains, it comes up clean. Certainly gave me a moment of pause though.

 

[Justintoxicated]

Following up. . .

BSOD even at vcore of 1.350 and input of 1.95.

First things first, time to get everything back to stock speeds and make sure I have a stable system according to p95 v28.1 at stock speeds. Otherwise, without that known for sure, I'm possibly chasing my tail.

Incidentally, virustotal shows the zip file for v28.1 of prime95 as malware (though only via one av engine). Yet, when you upload the actual EXE it contains, it comes up clean. Certainly gave me a moment of pause though.

Make sure your memory is at stock speeds. is that 1.350 for 4700?

 

[Super_Unleaded]

I don't think your OC is stable. Try lowering your memory to 1333mhz and go from there also work one thing at a time. I have 4770k that's stable at 4.5ghz - 1.276 volts. My processor wont overclock past 4.5 without adding tons of voltage. When I had it on manual it was stable at 1.275 for adaptive I had to add +.01. Also I recommend running OCCT if you get any error it will tell you in the first 30 minutes. Its an amazing OC stability tool.

 

[Hurin]

Make sure your memory is at stock speeds.

For now I'm running the memory at its XMP profile (1600).

is that 1.350 for 4700?

It was. But as we've learned, it's not as stable as the earlier version of Prime led me to believe.

Coming up on four hours at stock speeds. Just getting a good baseline w/ new Prime95.

 

[Justintoxicated]

For now I'm running the memory at its XMP profile (1600).


It was. But as we've learned, it's not as stable as the earlier version of Prime led me to believe.

Coming up on four hours at stock speeds. Just getting a good baseline w/ new Prime95.

"If at first you don't succeed, lower your standards"

 

[Bluesun311]

Wait. No one has mentioned that prime95 and other avx stress tests should never be run in adaptive voltage? The IVR will call for what voltage it thinks it needs in adaptive mode and ask for too much (engineering flaw). You onlu benchmark cpu in manual voltage mode. Period.

Both 27.9 and 28.1 place unrealistic loads on the processor so why go for the one that is even hotter? Realistic use will not require the same voltages needed to pass tests like p95 for 24 hours.

 

[WorldExclusive]

Do you run stress tests 24/7/365?

If not, I wouldn't be worried stress results. Get it stable and move on.
Run 4.7 with manual volts and enjoy your PC.

 

[Hurin]

Do you run stress tests 24/7/365?

If not, I wouldn't be worried stress results. Get it stable and move on.
Run 4.7 with manual volts and enjoy your PC.

Sorry, but that's not the way I roll.

Wait. No one has mentioned that prime95 and other avx stress tests should never be run in adaptive voltage? The IVR will call for what voltage it thinks it needs in adaptive mode and ask for too much (engineering flaw). You onlu benchmark cpu in manual voltage mode. Period.

Both 27.9 and 28.1 place unrealistic loads on the processor so why go for the one that is even hotter? Realistic use will not require the same voltages needed to pass tests like p95 for 24 hours.

v28.1 doesn't actually seem to be any hotter than 27.9. It does, however, seem to weed out whatever instability was still lurking in my system while overclocked to 4.7 w/ manual voltage. Though I hadn't BSODed or otherwise seen something relatively definitive, even at stock voltage I'd seen a couple programs quit unexpectedly. More than I would normally see. . . so I was suspicious. That v28.1 causes my system to crash even at static volts, tends to confirm those suspicions.

When I have adaptive voltage enabled, I test with custom FFT lengths (min 512k) to avoid the worst of that heat you mention. Still seems to weed out instability fine. But while finding my OC, I'm at manual voltages.

Thanks for your concern though.

 

[Gadfly]

Hey Hurin,

I too am fighting a 4770k right now, but I am not willing to push V-core past 1.3; It has been a long time since I bought a new cpu, and I have to admit, I don't like all this auto OC BS.

I am having trouble getting past 4.5 @ 1.28 with H100 push/pull

 

[Jalseng]

Hey Hurin,

I too am fighting a 4770k right now, but I am not willing to push V-core past 1.3; It has been a long time since I bought a new cpu, and I have to admit, I don't like all this auto OC BS.

I am having trouble getting past 4.5 @ 1.28 with H100 push/pull

lol auto OC.. something to market for the beginners..

 

[Gadfly]

If you have never overclocked with some micro-drill bits, and hand drill and a soldering iron, your are a beginner.


Even the "bios" over-clocking on these wiz bang boards have a lot of pre-sets, and auto-built rules. At least the Asus ROG boards have a full memory timing suite...


I just don't seam to be able to push my chip(s) like most. First 4770k was 4.4 @ 1.3; 4200 cache with 2400mhz memory; second 4770k is a big better... can do 4.4 @ 1.25, 4200 cache with 2400mhz memory

Running an H100 push + pull with upgraded fans.

Are you guys locking your cores and pushing all of them together? (I assume you are). The heat is just off the charts... I am seriously thinking about busting out one of the old vaporchills and seeing if it will run better on phase.
.

 

[Hurin]

Hey Hurin,

I too am fighting a 4770k right now, but I am not willing to push V-core past 1.3; It has been a long time since I bought a new cpu, and I have to admit, I don't like all this auto OC BS.

I am having trouble getting past 4.5 @ 1.28 with H100 push/pull

That's exactly where I settled last night. I thought setting RAM to defaults (despite XMP calling for 1600 8-8-8-24) seemed to get me over a hump into 4.6GHz @ 1.30 vcore. But I wasn't stable in Prime95 (I think it errored at 2.5 hours). 1.31 vcore didn't help. At which point, I decided to go back to 4.5GHz.

Settings:

cores synced @ 45
cache max @ 42
1.28v vcore manual (will switch to adaptive 1.28 +.001 after stress testing)
1.25v cache (will switch to adaptive 1.25 + .001 after stress testing)
1.8v input
HT Off

Cruising along at 5 hours in Prime95 now. May not go the full 24 hours until/unless I see any strange behavior. Max temps are 75-78c across cores. Which is 10c lower than my previous (not stable) OC of 4.7 GHz @ 1.34 vcore.

 

[Justintoxicated]

That's exactly where I settled last night. I thought setting RAM to defaults (despite XMP calling for 1600 8-8-8-24) seemed to get me over a hump into 4.6GHz @ 1.30 vcore. But I wasn't stable in Prime95 (I think it errored at 2.5 hours). 1.31 vcore didn't help. At which point, I decided to go back to 4.5GHz.

Settings:

cores synced @ 45
cache max @ 42
1.28v vcore manual (will switch to adaptive 1.28 +.001 after stress testing)
1.25v cache (will switch to adaptive 1.25 + .001 after stress testing)
1.8v input
HT Off

Cruising along at 5 hours in Prime95 now. May not go the full 24 hours until/unless I see any strange behavior. Max temps are 75-78c across cores. Which is 10c lower than my previous (not stable) OC of 4.7 GHz @ 3.4 vcore.

I hope you meant 1.34v core

 

[Hurin]

I hope you meant 1.34v core

Yes indeed. Thank you. Fixed.

 

[jhatfie]

Just getting started on my 4770k OC, so I figured I'll throw in my experience thus far to add an additional perspective of my limited testing thus far. Using MSI Z87-GD65 motherboard and Cooler Master Glacer 240L for cooling.

Set in bios:
CPU Ratio @ 46
Ring Ratio @ 42
1.295v vcore (Override setting)
1.21v Ring (Override setting)
1.9v VCCIN
CPU Analog/Digital I/O +.05v
Ram @ 2400Mhz

Aida64 cpu/fpu stress test will run for about 2 hours then fail. Max temp is core 0 @ 75C

Tried adaptive setting mode for CPU and Ring voltages when testing 4.5Ghz @ 1.25v vcore, but cpu voltage shoots up to 1.32v when Aida64 is running which is higher than the 1.3v I was hoping to max out at, so stopped stability test after about an hour. Max temps hit 78C.

4.7Ghz @ 1.3v (override) will crash system after about 10 seconds of stress test.

Will perform additional tests tonight.

 

[Gadfly]

Here is where I ended up:

4.5ghz across all cores
4.4ghz cache
2400mhz memory @1.65v
1.265v vcore
1.265v ring v

left everything else at default.

Run some 3d test looping most of the day (9 hours), ran prime for 6 hours and manually stopped it. Played some ARMA 3 for a few hours, had no issues.

 

[Hurin]

Failed prime at twelve hours.

Of course, some will "call that stable". . . but since I pass prime95 indefinitely at stock, my overclocking has affected something

Upped to. . .

vcore 1.290
cache voltage: 1.260
CPU input voltage: 1.85

This also failed (even sooner)

still shooting for 4.5GHz and 4.2GHz cache.

Reconfirming that I can finish 24-hour loop of Prime95 and Valley simultaneously at stock. Then I'll nudge all three up by .1 yet again.

I thought my CPU was a champ there for a while. Now beginning to wonder if it's a dud. =(

 

[Super_Unleaded]

Raise the Cpu Input Voltage to 1.90 and you should be good to go. Also cache voltage seems too high, i'm running mine at 4.2GHz and 1.17 volts. Forget about prime for a second and run OCCT much faster results than waiting for Prime. Two hours of OCCT is considered stable in my book.

 

[Justintoxicated]

Raise the Cpu Input Voltage to 1.90 and you should be good to go. Also cache voltage seems too high, i'm running mine at 4.2GHz and 1.17 volts. Forget about prime for a second and run OCCT much faster results than waiting for Prime. Two hours of OCCT is considered stable in my book.

I'm going to get OCCT soon too, I was under the impression that prime 28.1 was best for haswell though. 27.9 not so much. Playing with cache voltage myself, but I think I'm going to just forget about adaptive voltages since it seems to make everything harder. CPU is rock solid at 4400, but I want 4500!

 

[Super_Unleaded]

I'm going to get OCCT soon too, I was under the impression that prime 28.1 was best for haswell though. 27.9 not so much. Playing with cache voltage myself, but I think I'm going to just forget about adaptive voltages since it seems to make everything harder. CPU is rock solid at 4400, but I want 4500!

Do it! its worth it. When you test make sure your voltages are set to Manual so the voltage dosen't shoot up. Once you find the stable voltage switch to Adaptive mode and then add 0.01 volts. It literally takes OCCT 10-15 minutes to find an error or restart your system. Much better than waiting 6+ hours of prime. Don't give up on Adaptive, I literally spent weeks to find the right settings. What are your current settings for 4.4ghz?

 

[Darksword]

Why even both with adaptive voltage at all? My 4770k is running 4.5Ghz @ 1.225 w/ 4.5Ghz cache speed. All manual overclocking and I have no issues so far.

H110 push/pull setup.

 

[Super_Unleaded]

Why even both with adaptive voltage at all? My 4770k is running 4.5Ghz @ 1.225 w/ 4.5Ghz cache speed. All manual overclocking and I have no issues so far.

H110 push/pull setup.

Because it saves power and also extends cpu life. Its not necessary but its beneficial.

 

[Gadfly]

Pretty sure Prime95 is pretty worthless for these chips, even the the newer version.

I can run at 100% load over 8 threads with multiple applications, load testers, benchmarks, etc. and temps always hang around at 60-65'c; I fire up Prime (any version) and it jumps to 80'C +.

Not really sure I understand the dynamic going on with prime on these Chips.. but something is not right.

 

[Gadfly]

Failed prime at twelve hours.

Of course, some will "call that stable". . . but since I pass prime95 indefinitely at stock, my overclocking has affected something

Upped to. . .

vcore 1.290
cache voltage: 1.260
CPU input voltage: 1.85

This also failed (even sooner)

still shooting for 4.5GHz and 4.2GHz cache.

Reconfirming that I can finish 24-hour loop of Prime95 and Valley simultaneously at stock. Then I'll nudge all three up by .1 yet again.

I thought my CPU was a champ there for a while. Now beginning to wonder if it's a dud. =(

It still might be, What the combination of your overclock and prime95 changed was producing an outrageously high and artificial heat load that is proving to be difficult to replicate; the fact that it failed at 12 hours is further indication of an unintended consequence of the two. If it runs for 12 hours, it should run for 24, or 36. The length of time the CPU sustains load at a constant temperature is generally irrelevant; once you break into hours, and the cooling system is completely heat soaked, and equilibrium has been reached, and the load maintained reliably; then future results are predictable.

Though I have used it for years, in this rare case I am pretty sure the flaw is in prime95, not your chip. in all my previous builds and overclocks I would get similar thermal loads no matter how I loaded the CPU, be it via a large compile, number crunch, floating point, benchmarks, loops, burn in tools, etc. etc. All would be within a few degrees of each other, if not absolutely identical.

That is not the case with the new version of prime95 and my 4770k. I can run at 100% CPU load for hours on end at not break 65'C with every other app; I fire up prime95, and within seconds the CPU jumps to 85'C+ and the CPU starts to throttle back. It is the only application that I have in my library that produces such a result. This is an anomaly in my time overclocking and bench-marking CPU's.

That being the case, for now, I would test your stability with a several other burn-in / stability tests that produce repeatable results, and skip Prime95 until a future build is released. It appears to have something going on with it that produces artificially high temperatures and observable spikes in draw on the Haswell chips.

 

[Hurin]

It's no mystery what is going on. . . it's been well-documented in the Asus overclocking guide and elsewhere. . . and it's not just Prime95. Even AIDA64 will behave the same way if you ask it to only perform the FPU testing. Your temps will spike and you'll get throttling almost instantly.

But heat isn't the issue in my case. There is no sudden spike in heat after many hours and my temps are reasonable for Haswell (peaking at ~80c occasionally). Setting voltages to manual, and setting minimum FFTs to 512 keeps Prime95 from asking for the small FFTs that cause the artificially high temperatures (which, again, can be replicated in other programs).

I use IBT as well as other tools to quickly show me that my overclock is too aggressive. But when those tools stop showing me useful results (ie., I start passing them) and yet Prime95 continues to give me trouble, I have found through my own long experience that I ignore the Prime95 results or fail to pass my 24-hour torture test at my own peril. Because Prime95 often detects instability that the other "quickie" tests miss. It's simply not true that "if it runs for 12 hours, it should run for 24, or 36" in the case of an overly aggressively OCed CPU that is just barely short of voltage. Again, if it can run Prime95 indefinitely at stock, but fails at hour 22 while OCed where heat and voltages have remained reasonable, that's an indication of something other than a "problem with Prime95."

When a test fails to give us the results we would like, many of us seem too ready to fault the test itself.

 

[Hurin]

Raise the Cpu Input Voltage to 1.90 and you should be good to go. Also cache voltage seems too high, i'm running mine at 4.2GHz and 1.17 volts. Forget about prime for a second and run OCCT much faster results than waiting for Prime. Two hours of OCCT is considered stable in my book.

Is the free version of OCCT at ocbase.com the preferred version? I'm a bit wary of trying it considering virustotal has one set of definitions reporting it as malware (both the EXE and the ZIP). I realize that's probably a false positive, but better safe than sorry is usually my policy.

 

[Hurin]

Pretty sure Prime95 is pretty worthless for these chips, even the the newer version.

I can run at 100% load over 8 threads with multiple applications, load testers, benchmarks, etc. and temps always hang around at 60-65'c; I fire up Prime (any version) and it jumps to 80'C +.

Not really sure I understand the dynamic going on with prime on these Chips.. but something is not right.

Try AIDA64 torture test w/ only FPU selected. It'll produce similar results (nearly instant 95c+ temperatures and throttling). Especially if your voltages are set to be "adaptive."

IBT will generally do so too, though not as abruptly/quickly.

As I've said, Prime95 isn't alone in causing these abnormally high temperatures. I have found that setting FFT minimum to 512k works around the issue while still stressing the chip enough to be a meaningful torture test.

 

[Super_Unleaded]

Here is the official website http://www.ocbase.com/ click on download and click either the .exe or .zip file. I got the zip file and had no problems with virus or anything like that. They are legit. Also make sure when you test your voltages are set to "manual" in bios or else your voltages will spike to dangerous levels and you will reach 90+ degrees in a second. Its a great tool because it also has a built in temp monitor and also works with other temperature software and you can set a temp limit for the test to stop. Also make sure to test with CPU:OCCT and Large Data Set. Let me know if you have any questions. Also another great reason to use OCCT because with each minute it will down clock for a second and go back up, testing the stability. In prime it always keeps it constant which is easier to pass.

 

[Gadfly]

First, Merry Christmas!

It's no mystery what is going on. . . it's been well-documented in the Asus overclocking guide and elsewhere. . . and it's not just Prime95. Even AIDA64 will behave the same way if you ask it to only perform the FPU testing. Your temps will spike and you'll get throttling almost instantly.

I tested AIDA64 and even when only doing FPU with small FFT's that fit into cache it was still a full 14'C cooler.

But heat isn't the issue in my case. There is no sudden spike in heat after many hours and my temps are reasonable for Haswell (peaking at ~80c occasionally). Setting voltages to manual, and setting minimum FFTs to 512 keeps Prime95 from asking for the small FFTs that cause the artificially high temperatures (which, again, can be replicated in other programs).

I still can not replicate it... but it did help.

It's simply not true that "if it runs for 12 hours, it should run for 24, or 36" in the case of an overly aggressively OCed CPU that is just barely short of voltage. Again, if it can run Prime95 indefinitely at stock, but fails at hour 22 while OCed where heat and voltages have remained reasonable, that's an indication of something other than a "problem with Prime95."

When a test fails to give us the results we would like, many of us seem too ready to fault the test itself.

Yes it is , and that has been an issue with the "24 hour prime" test for the past 20 years. It is flawed in it's very concept, and implementation. Prime95 has been shown to be flawed many times in the past, and in it's current version, it appears once again has issues. So we will just have to agree to disagree on it's actual usefulness, especially in a modern context. Which is 100% fine by me, we are both obviously both passionate about our life long hobby, so such a small difference of opinion on a single application is really not worth arguing over.

All of that said, I could use another set of "eyes" so to speak on my overclock. Your chip appears to be a better sample than mine. I cannot get windows to even boot at 4.7, even at 1.34 vcore (1.25v @ 4.2ghz ring); I did up the input voltage to 1.8v, and added 0.05v to the I/O voltage, disabled spread spectrum, and turned the memory back down to 1600mhz. So 4.6 may be the limitation of my chip, but I would like to ask if anyone has good "check list" of setting to verify, I would appreciate it, here is the build:

4770k
Asus Maximus VI Formula
2x8GB 2400mhz DDR3 (off brand, Silicone power I think? Tested to be stable 2400mhz)
bunch of Samsung Evo 250gb SSD's (Raid 0)

 

[Hurin]

For your 4.7GHz attempt (I was never able to get it completely stable at 4.7 -- but I feel like I got close before getting nervous about voltages above 1.34vcore!), I'd leave cache at default/auto (3.9GHz) for now until you isolate the issue with your core.

I'd definitely go with 1.9v for CPU input.

Is HT off? (I would turn it off -- and have it off on mine).

Might try running RAM at 1333MHz if you're going for anything above 4.5GHz. The integrated memory controller seems to suffer at higher core clock speeds. I noticed an improvement in stability when I decreased my RAM speed from 1600 (via XMP profile) to the default of 1333. I think I was trying to get 4.6GHz stable at that point though. I'm now back at 1600MHz while dialing in 4.5GHz.

P.S. Merry Christmas to you as well! My daughter is playing with her X-Wing while I type this and my one year-old little guy is napping. Good times. But I also hope Santa brings me a belated gift of stability at 4.5GHz for Christmas.

 

[Gadfly]

I disabled HT, and set the input voltage at 1.9, set memory to 1333, and left the cache at auto, I was able to boot into windows, but it was at 1.35v core eek, and was not stable at all. It made it though @ 5min of Aida64 and BSOD'd; I am going to guess that I will need at least 1.37v to run even "semi-stable"... which is more heat than my 280mm push/pull h100 can take; even with 100+ cfm fans.

Thier is just no way this CPU is going to run that high without a de-lid or phase.

Here is where I have ended up:

4.6ghz
1.285v vcore
4.2 cache
1.215v ring

HT enabled
PLL over-volt enabled
+ .05v offset to the I/O voltages
spread spectrum disabled
1.9 input voltage
low PLL selected

2400mhz on the memory via XMP @ 1.65v

Starting a stability test loop now, but initial results and temps look promising with these settings; then I will start to configure the adaptive voltage.

I have a vice, hammer, and a bunch of painters tape.... which is very tempting at this point.

 

[Justintoxicated]

Do it! its worth it. When you test make sure your voltages are set to Manual so the voltage dosen't shoot up. Once you find the stable voltage switch to Adaptive mode and then add 0.01 volts. It literally takes OCCT 10-15 minutes to find an error or restart your system. Much better than waiting 6+ hours of prime. Don't give up on Adaptive, I literally spent weeks to find the right settings. What are your current settings for 4.4ghz?

1.275v CPU 1.8v Input

Trying to stabilize 4500mhz without going over the mythical 1.3v for CPU. I seem to keep getting computer restart after 6-8 hours of prime95 small FFT. I have bumped cou voltage to 1.3v (so its over 1.3v when running prime) and CPU Cache voltage to 1.295 now Vin @ 1.9v

Bumping the cache seemed to help some, just increased Vin up to 1.9v to see if that is the real issue. Next step is to boost CPU voltage to 1.31v and maybe even 1.32v I think it will stabilize around there. Then I just need to decide if that is really a dangerous voltage with the CPU running so cool (30c) drop after delidding. @ 1.3v I seem to max at 64c now. @ 1.275 my max is closer to 59-60c with 65-70F ambients.

Merry Christmas!

 

[Hurin]

Failed prime at twelve hours.

Of course, some will "call that stable". . . but since I pass prime95 indefinitely at stock, my overclocking has affected something

Upped to. . .

vcore 1.290
cache voltage: 1.260
CPU input voltage: 1.85

This also failed (even sooner)

still shooting for 4.5GHz and 4.2GHz cache.

Reconfirming that I can finish 24-hour loop of Prime95 and Valley simultaneously at stock. Then I'll nudge all three up by .1 yet again.

I thought my CPU was a champ there for a while. Now beginning to wonder if it's a dud. =(

Final results. . .

Stable for 24 hours at stock.

Stable for 24 hours (Prime95 512k FFT min + Valley) at. . .

4.5GHz
4.2GHz cache
vcore 1.290v
cache/uncore 1.250v
cpu input 1.900v

Going to enable adaptive voltage and hope it stays that way. When I have more time and can be without my computer for a while, I may shoot for 4.6GHz again. But for now, this is fine for me.

 

[Gadfly]

Here is where I ended up:

Ran AIDA64 for 13 hours, simply need to use my computer before it could run for 24, fire-strike demo looped for 6+ hours

4.6ghz
1.295v vcore
4.4 cache
1.280v ring

HT enabled
PLL over-volt enabled
+ .05v offset to the I/O voltages
spread spectrum disabled
1.9 input voltage
low PLL selected

2x8GB 2400mhz 11-11-11-30 memory via XMP @ 1.65v


GTX 780
1201mhz core, 6400mhz memory

A very nice upgrade from my i920 @ 4.2ghz, I am a little disappointed that I could not get 4.7 or 4.8 out of it, it seams that finding chips that exceed 4.5/4.6 are pretty rare. I am very seriously thinking about de-lidding the cpu and attempting to get another 100mhz out of it.