Overclocking a 955 BE.

[Whitefang]

So, when I first got my chip and my Hyper 212+, I spent a long time tweaking my overclock to get it stable. I managed to get it up to 3.8GHz at 1.4V (temperatures capped at 56C). It was stable for 2-3 months, before it started crashing on me. I wasn't sure why, so I dropped it back down to stock settings - it worked fine.

I then got it to 3.6GHz at stock voltage (1.3v, I believe). It was, again, stable for about a month before it started crashing again. I then dropped it to stock, 3.2GHz at 1.3V, and it was fine.

So I decided that, instead of overclocking (as it seemed to be hating me), I would just undervolt it as much as I could, so I could drop temperatures and lower my fan settings, so it wasn't so much like a wind tunnel. I got it down to 3.2GHz at 1.25V, dropping temperatures by something like 5-8C, maybe a bit more (idle at 34C). It hasn't started crashing again, so I'm not sure what the problem was.

Does anyone have anything I could try? I quite miss the processing power it had at 3.8GHz - I can no longer stream above 30FPS 720p while playing a game like League of Legends - it's too processor heavy. It's not that important, but I'd like to have it, you know? .

Current rig:

Radeon HD 4890.
Phenom II 955 BE (3.2GHz at 1.25V, cooled by Hyper 212+ with MX-4 thermal paste).
Asrock 970 Extreme3.
2x4GB G.Skill DDR3 1600 (at 1333) RAM.
NZXT Source 210.
Edit: Corsair 600W Builder PSU.

Edit: I want to state that all my overclocks passed at least a 24 hour Prime95 blend, and then started crashing later.

 

[manu2b]

That's weird. With 1.4v and 56°C max, you're far from anything dangerous for your CPU.

I assume nothing changed in your rig (more ram modules?).

What type of crash do you get?

 

[Whitefang]

Nothing changed, no. Sometimes it was a bluescreen. Other times, my computer would just sort of... die. Spotify would keep playing music, but every window would start displaying a solid color.

 

[manu2b]

Do you remember the bsod error code?

 

[Whitefang]

No, this was a few months ago. I could work the overclock back up to their and see if I can get another BSOD, but it would never shut off when I was there (I'm just assuming it BSOD'd, but maybe it just shut off).

 

[storm-chaser]

First of all, dont worry about passing prime for 24 hours. Thats not needed for a personal use PC. More like 30 minutes or an hour should be fine. Really it comes down to the programs/games you will be running. So long as its stable in the desired configuration, dont waste hours and hours running at at full throttle just to make sure its 300% stable. For example, my 970T overclocked at 4.2 only passes 10 minutes of prime and then BSODs, but I can play all the latest games at the highest resolution or render cad drawings without any issues at all in a stable configuration 24/7. Or just browse the web. Hope you get what Im saying ie you can take advantage of an overclock and remain stable EVEN if you start failing in prime after 20-30 minutes. I have lots of tests and documentation to back this up. Take advantage of the hardware you have. The key is stability for YOU, not stability for PRIME. Stability for you means no BSODs and a stable 24/7 config, and this can be achieved even if you are failing prime tests. Think of a CPU as an engine. Stress testing an engine is done to measure HP and torque. The engine is driven from approx 2000rpm-7000rpm and output is measured using a dyno. Prime is like a dyno for your CPU. If you ask most automotive experts, durability testing is not just a matter of holding the gas pedal to the floor (IE prime), and holding it there for 24 hours. its more of engineering a durable engine to suite the demands of whatever car it is going to be install in. Much the same applies to a CPU. SQL servers or domain controllers see lots of IO traffic and heavy CPU usage and in that case, prime would be more applicable. For you, download AIDA64, and be content if you can pass the Benchmark performance suite. It also has a built in stability test.


What you are working with:
V core = 0.85V - 1.425V
Maximum operating temperature = 55°C - 62°C
Thermal Design Power = 125 Watt
C2 stepping = HDZ955FBK4DGI
C3 stepping (better proc) = HDZ955FBK4DGM
Check your stepping using CPU-Z (and post a pic)
The older C2 stepping chips dont OC as well.

Here is what you need to do to get a stable overclock:
-Something is wrong with the cooling system - you need to get that under control before attempting an OC. The hyper 212+ should keep that CPU around 110-120 under FULL load. Reseat or re-apply heat sink grease.
-Set FSB to 200mhz
-Set core voltage to 1.3900
-Set CPU-NB voltage to 1.37-.138
-Set NB voltage to 1.22
-Set SB to stock 2000mhz and stock voltage
-Set memory multi to achieve 1600mhz clock. (if your controller doesnt support 1600, look at the XMP profile in CPU Z and tighten up timings in the bios.
-Set proper memory voltage for mem clock speed
----------------------------------------------
Avoid raising the FSB, do all your CPU tuning via the CPU multi.
RUN PRIME or stress tests initially to find out the max stable limit of the CPU. In my case, its 3.9ghz with all 6 cores running. Above that, it bsods every 4=6 hours no matter what tweaking I've done.
-Find the max stable CPU clock (IE 3.7-3.8)
-Increase voltage to the processor as needed. Your temps should not be anywhere near maximum if you keep within the voltage lines.
Once you have found the processor sweet spot, start tweaking the NB. In my case, on one board I could only get 2800mhz. On my current systems, I've reached 3000mhz simply by using the above provided voltaged, with massive throughput gains. let us know how it turns out. if the chip is c2, you may be limited.

 

[manu2b]

A CPU and a car engine are two very, very different things.

The OP rig do not crash while running Prime95. It crashes during day to day use.

EDIT: Yes, running p95 is needed to make sure the rig is stable. You are not helping the OP with your statements and "advice"
It's a Deneb HT must stay around 2000mhz...

 

[BeepBeep2]

30 minutes to an hour on any Phenom II CPU in Prime95 is not enough to ensure stability. Under 1 hour with a Phenom II X4 (on edge, one that will fail in 1-8 hours) is guaranteed to BSOD during daily use.
CPUs often give up between 1 and 8 hours. I'd consider more than 8 hours completely useless.

Storm chaser has a lot of incorrect input in his post and his car analogy makes no sense.

Also manu2b, Deneb CPU_NB is default at 2 GHz but that doesn't mean that most don't do about 2.6-3.0. Anyway, it will be simple to diagnose the problem by just knowing the BSOD hex code. I've seen hundreds of these while overclocking/benchmarking

0x00000124 is most common with Phenom II, CPU cores need more voltage in that case. Anything "IRQL..." "PAGE_FAULT" ... etc are memory related/CPU_NB

 

[storm-chaser]

30 minutes to an hour on any Phenom II CPU in Prime95 is not enough to ensure stability. Under 1 hour with a Phenom II X4 (on edge, one that will fail in 1-8 hours) is guaranteed to BSOD during daily use.
CPUs often give up between 1 and 8 hours. I'd consider more than 8 hours completely useless.

Storm chaser has a lot of incorrect input in his post and his car analogy makes no sense.

Also manu2b, Deneb CPU_NB is default at 2 GHz but that doesn't mean that most don't do about 2.6-3.0. Anyway, it will be simple to diagnose the problem by just knowing the BSOD hex code. I've seen hundreds of these while overclocking/benchmarking

0x00000124 is most common with Phenom II, CPU cores need more voltage in that case. Anything "IRQL..." "PAGE_FAULT" ... etc are memory related/CPU_NB

There are no hard set rules to overclocking. If the system is stable for what you do and doesnt crash, so what if doesnt pass a damn stress test for 24 hours? Take the system below for example, it fails prime after about an hour. Yet I can run any game on max settings tab out and have 20 IE windows open, render anything I want and still watch netflix at the same time on my 2nd monitor without any issues or hickups. I've never seen a BSOD in three years short of running prime tests. Does that mean I should re-tool everything set my NB to 2400 and my cpu clock back to 3.6 for the sake of passing a 24 hour prime marathon? does it give me a stronger sense of accomplishment? I dont really think so, but I guess to each his own.

Not for the performance gains to be had, not for the higher hash benchmarks and certainly not for the 35 ns memory response time. No, 24 hours of prime is not worth that much to me, nor is it to the exteme overclockers going for record breaking clock attempts. Maybe for a network administrator who requires near 100% uptime on a new server build. Perhaps for a top gun who has thousands to spend on cooling and components. Truth is, sometimes you have to challenge the status quo, and in this case I'll take this OC any day of the week, prime can go in the recycle bin. Its been my work machine for three years and has yet go down in flames, been overclocked since day one 100%without missing a beat. That fits the definition of "stable", doesnt it?

I do not mean to devalue Prime and its uses. I use prime when stress testing new servers for our clients and hospital networks. I use prime on $3000 I7 based custom builds, I use prime to check for bad cores and other flawed hardware. Prime is a very valuable tool for finding the limits of system components, certainly not the "be all", but in the overall scheme just one of many tools to make an overclock "stable", and in this case one I didnt need.

 

[BeepBeep2]

There are no hard set rules to overclocking. If the system is stable for what you do and doesnt crash, so what if doesnt pass a damn stress test for 24 hours? Take the system below for example, it fails prime after about an hour. Yet I can run any game on max settings tab out and have 20 IE windows open, render anything I want and still watch netflix at the same time on my 2nd monitor without any issues or hickups. I've never seen a BSOD in three years short of running prime tests.

Okay?...playing games (especially) on max settings means nothing for CPU stability...it is a very low % CPU load most of the time. Surfing the internet too, you can do that well enough on a Sempron 140 even. ...watching netflix, I'm not sure if it is GPU accelerated or not but I'll guess that it is...(~5% CPU load)

I'm glad you haven't seen a BSOD. You also haven't been running any type of CPU-intensive task so that is a given...

Does that mean I should re-tool everything set my NB to 2400 and my cpu clock back to 3.6 for the sake of passing a 24 hour prime marathon? does it give me a stronger sense of accomplishment? I dont really think so, but I guess to each his own.

Uhh, no? If your CPU passes 1 hour of Prime95 and "fails right after", you could probably go 1-2 MHz HT Ref. Clock back, lower the CPU_NB multiplier (useless overclock anyway for the types of things you are doing, you're gaining nothing...) or bump voltages a tad and be stable...(3868 MHz and 3094 MHz CPU_NB)

Not for the performance gains to be had, not for the higher hash benchmarks and certainly not for the 35 ns memory response time. No, 24 hours of prime is not worth that much to me, nor is it to the exteme overclockers going for record breaking clock attempts. Maybe for a network administrator who requires near 100% uptime on a new server build. Perhaps for a top gun who has thousands to spend on cooling and components. Truth is, sometimes you have to challenge the status quo, and in this case I'll take this OC any day of the week, prime can go in the recycle bin. Its been my work machine for three years and has yet go down in flames, been overclocked since day one 100%without missing a beat. That fits the definition of "stable", doesnt it?

What?
Who cares about how many ns AIDA measures for latency? That even changes ~5ns everytime you run it, not totally accurate. IMO again, over 8 hours Prime95 or any other stability program is useless however a "1 hour or less" stable Phenom II X4 CPU I guarantee you I could make BSOD with something as simple as Windows Movie Maker (uses up to 4 threads)

No need to exaggerate on the price of cooling components etc etc, "challenging the status quo"...

I do not mean to devalue Prime and its uses. I use prime when stress testing new servers for our clients and hospital networks. I use prime on $3000 I7 based custom builds, I use prime to check for bad cores and other flawed hardware. Prime is a very valuable tool for finding the limits of system components, certainly not the "be all", but in the overall scheme just one of many tools to make an overclock "stable", and in this case one I didnt need.

//img.photobucket.com/albums/v423/smithsil/3900oc.jpg[IMG][/QUOTE]
Well...

 

[storm-chaser]

Okay?...playing games (especially) on max settings means nothing for CPU stability...it is a very low % CPU load most of the time. Surfing the internet too, you can do that well enough on a Sempron 140 even. ...watching netflix, I'm not sure if it is GPU accelerated or not but I'll guess that it is...(~5% CPU load)

I'm glad you haven't seen a BSOD. You also haven't been running any type of CPU-intensive task so that is a given...

Uhh, no? If your CPU passes 1 hour of Prime95 and "fails right after", you could probably go 1-2 MHz HT Ref. Clock back or bump voltages a tad and be stable...

What?
Who cares about how many ns AIDA measures for latency? That even changes ~5ns everytime you run it, not totally accurate. IMO again, over 8 hours Prime95 or any other stability program is useless however a "1 hour or less" stable Phenom II X4 CPU I guarantee you I could make BSOD with something as simple as Windows Movie Maker (uses up to 4 threads)

No need to exaggerate on the price of cooling components etc etc, "challenging the status quo"...

As we can see, stability can be measured in many different ways.

-You are right, I was only talking fun and gaming stuff. I Forgot to mention the OS test components it has handled from time to time. How about that Hyper-V server I had set up on it for a while, back when I had 16GB of ram in it @ 1784mhz, 3121nb and 3.9cpu with a couple guest vms including a sql server. Now do you want to talk CPU intensive? naww, thats just kid stuff. Again, no flaws were found out. Hyper 212+ keeps it cool. Guess windows movie maker would still probably make it BSOD? since it hasnt passed 24 hours of prime. How about sbs 2011 w/ exchange running full bore? Good enough for me...

I was referring to a complete I7 build up, not simply cooling components.

rebuttle? haha

 

[Whitefang]

So I love how you're arguing about the test of stability I've personally chosen to use, because I would like to be able to stress my processor for hours at a time (which Prime95 does) in my thread. I'd like to kindly ask you to take it to PM, or to your own thread, please .

 

[BeepBeep2]

LOL...Hyper-V/SQL server? 100% load for extended period? Please...try WCG.

Anyway, Whitefang...
Is your computer set up to do minidumps (memory dumps containing the info on the BSOD)?

If so, what does WhoCrashed say?

 

[Whitefang]

Uhh, I don't know. How do I tell if it is? xD.

 

[BeepBeep2]

WhoCrashed will either find them or won't

 

[trents]

I agree in theory with Storm-chaser in that what is stable is relative to the kind of use you put the machine to. And I definitely do not believe 8 hrs. is necessary to ensure stability in all but the most mission critical and demanding tasks. Running a system at 100% load for many hours of stress testing does put more wear and tear on a system than does a lesser amount of time and when you think that may be multiplied by a number of "almost but not quite" failed 8 hr. tests that really adds up. Heat is the enemy of electronics and even when we keep a CPU well-cooled there are all those motherboard components that are heating up.

Where I disagree with Storm-Chaser and others is the amount of stress testing time needed to reasonably ensure your overclock is stable. Thirty minutes is certainly not enough but IMO, neither is 8 hours of Prime95, OCCT, Intel burn test, or whatever 100% stress test method people use necessary to reasonably ensure stability. In fact, there are those who say even 8 hrs. is not enough. Where do you draw the line, anyway?

Personally, I have settled on two hours. I have never had an overclocked computer crash or BSOD on me in regular app use if it will pass two hours of Prime95 blend. And I keep my computer on 24/7.

 

[BeepBeep2]

They don't really add up...unless you are running Prime95 @ 80+c with 1.7v on air you won't really be doing much damage. Phenom II X4 if temps are within reason can usually take 1.6v fine, no matter what you're doing...be it web surfing or crunching/folding.

You guys should know that Prime95 isn't just a stress tester but a research program to find new Prime numbers...people run it 24/7 for years on end.

"chew*" (overclocker that set previous AMD CPU frequency WR @ AMD Austin) recommended minimum 4 hours, especially if overclocking ram as well (4 hours regardless though)...

Again, I've had overclocks fail after 3 hours, but usually 4 hours is enough to ensure there won't be future issues.

 

[trents]

They don't really add up...unless you are running Prime95 @ 80+c with 1.7v on air you won't really be doing much damage. Phenom II X4 if temps are within reason can usually take 1.6v fine, no matter what you're doing...be it web surfing or crunching/folding.

You guys should know that Prime95 isn't just a stress tester but a research program to find new Prime numbers...people run it 24/7 for years on end.

Yep, and I know people who subject themselves to a lot of stress for years and may get away with it for years but maybe not. And it still shortens their life span to some degree and might even make it very short. And I think its true also that overclocked computers will be affected by extra stress more than non overclocked ones. And it might surprise you that a lot of us guys did already know that Prime95 was originally developed as a research tool; to find new prime numbers as a matter of fact. But that doesn't make it any less stressful.

"chew*" (overclocker that set previous AMD CPU frequency WR @ AMD Austin) recommended minimum 4 hours, especially if overclocking ram as well (4 hours regardless though)...

Again, I've had overclocks fail after 3 hours, but usually 4 hours is enough to ensure there won't be future issues.

 

[manu2b]

@Storm Chaser: I meant HT (freudian slip). Edited...

 

[Whitefang]

Crash Dump Analysis
--------------------------------------------------------------------------------

Crash dump directory: C:\Windows\Minidump

Crash dumps are enabled on your computer.


On Sat 3/17/2012 6:31:13 PM GMT your computer crashed
crash dump file: C:\Windows\Minidump\031712-26176-01.dmp
This was probably caused by the following module: ntoskrnl.exe (nt+0x7CD40) 
Bugcheck code: 0xA (0xFFFFFFFFFFFFFFFF, 0x2, 0x0, 0xFFFFF80002A82366)
Error: IRQL_NOT_LESS_OR_EQUAL
file path: C:\Windows\system32\ntoskrnl.exe
product: Microsoft® Windows® Operating System
company: Microsoft Corporation
description: NT Kernel & System
Bug check description: This indicates that Microsoft Windows or a kernel-mode driver accessed paged memory at DISPATCH_LEVEL or above.
This appears to be a typical software driver bug and is not likely to be caused by a hardware problem. 
The crash took place in the Windows kernel. Possibly this problem is caused by another driver which cannot be identified at this time. 


On Sat 3/17/2012 6:31:13 PM GMT your computer crashed
crash dump file: C:\Windows\memory.dmp
This was probably caused by the following module: atikmdag.sys (atikmdag+0x4DF97) 
Bugcheck code: 0xA (0xFFFFFFFFFFFFFFFF, 0x2, 0x0, 0xFFFFF80002A82366)
Error: IRQL_NOT_LESS_OR_EQUAL
file path: C:\Windows\system32\drivers\atikmdag.sys
product: ATI Radeon Family
company: Advanced Micro Devices, Inc.
description: ATI Radeon Kernel Mode Driver
Bug check description: This indicates that Microsoft Windows or a kernel-mode driver accessed paged memory at DISPATCH_LEVEL or above.
This appears to be a typical software driver bug and is not likely to be caused by a hardware problem. 
A third party driver was identified as the probable root cause of this system error. It is suggested you look for an update for the following driver: atikmdag.sys (ATI Radeon Kernel Mode Driver, Advanced Micro Devices, Inc.). 
Google query: atikmdag.sys Advanced Micro Devices, Inc. IRQL_NOT_LESS_OR_EQUAL




On Fri 3/16/2012 12:53:57 AM GMT your computer crashed
crash dump file: C:\Windows\Minidump\031512-27783-01.dmp
This was probably caused by the following module: atikmdag.sys (atikmdag+0xF8969) 
Bugcheck code: 0x1000007E (0xFFFFFFFFC0000094, 0xFFFFF88004B15969, 0xFFFFF88006BECFC8, 0xFFFFF88006BEC820)
Error: SYSTEM_THREAD_EXCEPTION_NOT_HANDLED_M
file path: C:\Windows\system32\drivers\atikmdag.sys
product: ATI Radeon Family
company: Advanced Micro Devices, Inc.
description: ATI Radeon Kernel Mode Driver
Bug check description: This indicates that a system thread generated an exception which the error handler did not catch.
This appears to be a typical software driver bug and is not likely to be caused by a hardware problem. 
A third party driver was identified as the probable root cause of this system error. It is suggested you look for an update for the following driver: atikmdag.sys (ATI Radeon Kernel Mode Driver, Advanced Micro Devices, Inc.). 
Google query: atikmdag.sys Advanced Micro Devices, Inc. SYSTEM_THREAD_EXCEPTION_NOT_HANDLED_M




On Fri 3/16/2012 12:16:06 AM GMT your computer crashed
crash dump file: C:\Windows\Minidump\031512-27019-01.dmp
This was probably caused by the following module: atikmdag.sys (atikmdag+0xF8969) 
Bugcheck code: 0x1000007E (0xFFFFFFFFC0000094, 0xFFFFF88004B60969, 0xFFFFF88006BEAFC8, 0xFFFFF88006BEA820)
Error: SYSTEM_THREAD_EXCEPTION_NOT_HANDLED_M
file path: C:\Windows\system32\drivers\atikmdag.sys
product: ATI Radeon Family
company: Advanced Micro Devices, Inc.
description: ATI Radeon Kernel Mode Driver
Bug check description: This indicates that a system thread generated an exception which the error handler did not catch.
This appears to be a typical software driver bug and is not likely to be caused by a hardware problem. 
A third party driver was identified as the probable root cause of this system error. It is suggested you look for an update for the following driver: atikmdag.sys (ATI Radeon Kernel Mode Driver, Advanced Micro Devices, Inc.). 
Google query: atikmdag.sys Advanced Micro Devices, Inc. SYSTEM_THREAD_EXCEPTION_NOT_HANDLED_M




On Thu 2/23/2012 9:19:15 PM GMT your computer crashed
crash dump file: C:\Windows\Minidump\022312-13150-01.dmp
This was probably caused by the following module: ntoskrnl.exe (nt+0x7CC40) 
Bugcheck code: 0x101 (0x31, 0x0, 0xFFFFF88002FD5180, 0x3)
Error: CLOCK_WATCHDOG_TIMEOUT
file path: C:\Windows\system32\ntoskrnl.exe
product: Microsoft® Windows® Operating System
company: Microsoft Corporation
description: NT Kernel & System
Bug check description: This indicates that an expected clock interrupt on a secondary processor, in a multi-processor system, was not received within the allocated interval. 
This appears to be a typical software driver bug and is not likely to be caused by a hardware problem. This problem might be caused by a thermal issue. 
The crash took place in the Windows kernel. Possibly this problem is caused by another driver which cannot be identified at this time.