SOLVED Brick wall at 3.9GHz

I'll admit now that I am not the most well versed at overclocking. Only ever done a handful of older Intel systems (P4 anyone?) I've been trying to get this Phenom to 4.0Ghz and Beyond <insert buzz lightyear> and I seem to have brick walled at 3.9.

I got it working at 4.01GHz by pushing the Multipler to 19.5 and the HT Ref from 200 to 206MHz (apps started crashing at 207) with the CPU-NB at 2.6GHz. RAM was left at the default (crashed above stock). AMD Overdrive tests and some video games ran perfectly.

However, cracking open Prime95 the system started to choke and it fell over with two Panics (Memory Management and IRQ_NOT_LESS_OR_EQUAL) within seconds of each test starting.

I lowered HT Ref back down to its default of 200MHz, but by doing this I was able to push the memory to 1600MHz at 1.57v. Prime95 ran without question and the system remained responsive. It wasn't a 2 hour test, only 45 minutes as I had to dash to work and I didn't want to be out if it collapsed but I'm confident that its stable now.

The HT Ref falling over at only 207MHz seems very, very low to me when others report 225+. Too high a multiplier set on the CPU for it to work?

How would the core voltage factor into this? Set it to 1.5v in overdrive, but speed fan and HWMonitor showed that it never went anywhere above 1.46v under stress (0.01v increase over stock). It may be a BIOS only tweak but I'll test that when I get home.

Anything tips?

Cheers
-Ben

On same page...

...With these CPUs you would need to be concerned with and deal with certain things when overclocking:

The first thing would be to download the tools we use when overclocking: CPU-z, HWMonitor and Prime95. All are freeware.

The next thing to do would be to check core/"package" temps and CPU temps at stock frequencies and voltages to get an idea of how much overclocking headroom you have from a temperature standpoint, which is the main limiting factor.

To check those stock condition temps: Open HWMonitor on your desktop and adjust the slider and frame so you can see the core temps section and the voltage section. Leave it open while you run the Prime95 blend test for 20 minutes. When that is done, attach a pic of the HWMonitor interface with your next post. To attach a pic, first crop and save the image to disc using Snipping Tool in Windows Accessories. Then click on the Go Advanced button at the bottom of any new post window. When the Advanced Post window appears, click on the little paperclip icon which will load the file browser and upload tool. The rest is obvious.

Next you need to be able to:

1. In bios, first disable: Cool N Quiet, Turbo, C1E and C6 (BD/PD cpus). Also new with FX-series processor is the APM setting in bios. Locate and disable APM, so the high current draw thru the VRMs will not cause cpu throttling.
2. In Windows Control Panel Power Options configure it to High Performance.

These first two measures should disable all the "green" power saving stuff that cause erratic frequencies and voltages.

Then In bios, you will need to be able to manipulate:
1. CPU core voltage
2. CPU multiplier
3. CPUNB voltage
4. CPUNB frequency (may be expressed as a multiplier)
5. And perhaps, memory voltage

If you can locate those controls in bios you will be off to a good start and we can advise more specifically with regard to their adjustment. The terminology you encounter in the particular bios you are working on may vary somewhat from what I have used so be aware of that.

If you have a CPU with an upward unlocked multiplier, now just start increasing your CPU multiplier by .5x.

After each increase run a 20 minute Prime95 blend test to check for stability.

Always have HWMonitor open on the desktop to monitor core temps. Max stable core/"package" temp is typically 55-58c, somewhere in there. CPU Temp should not go higher than 70c.

When you first fail the 20 minute Prime blend test, increase your CPU core voltage by .025 and retest. If you still fail, add another .025 vcore. Then retest, repeating the pattern outlined. Stop adding vcore when your reach 1.5 or core temps exceed the parameters mentioned above.

Failing the Prime test can mean blue screen, spontaneous restart, lockup or one of the Prime core workers dropping out. Post back after you have added two increments of vcore or hit the mid 50's core temp wall. When you post, attach pics of HWMonitor; CPUz CPU Tab and CPUz Memory Tab from your last Prime95 blend run and give us a report.

I should actually be able to clock this system relatively well given the current temperature the system is running at when clocked to 3.9GHz; max Core temp only reaches 51 under Prime 95 and the fans barely tick over 1k RPM (with a 1.5k limit) so there's plenty of headroom.

Just to clarify for my own naïveté: When increasing the clock, an increase in voltage is also required to drive the higher clock. Yes?
My system at stock voltage of 1.45v crashes at x20 multiplier, but is stable at x19.5 - would an increase in voltage allow me to reach x20 and higher and remain stable?

4C or so is plenty? I recall people saying to keep these around 55C... 60C at the most (as RgoneSTER! posted already) as you tend to run in to instability with temps above 55-60C...

You may have lots of headroom on CPU temps and voltage but you don't have lots of headroom because of the motherboard.

I know it has overclocking options but it is not for overclocking. You're actually pretty lucky to get 3.9GHz out of it and even then I'd suspect it's throttling.

It all boils down to the VRM. Your motherboard has a very skimpy one likely made with low quality parts as well. When they say 125W CPU support on a cheap board they mean it will handle the average user's computer demands with a 125W CPU and be fine. They're really playing to the literal meaning of what they say on cheap motherboards, where as the more expensive boards will often say 140W CPU support but can deal with 350W of CPU draw all day long.

Here are 2 things that were helpful for another user to get what I'm talking about. He had the next up motherboard from yours and found it hard to hit 3.8GHz.

This is the power draw scaling of your processor. I drew in the 140w line which is more than what ASUS intends your board to deal with already(but was what the other user's board was for). The spike at 3.9GHz is because they upped voltage from 1.4V to 1.5V.


This was the VRM on his board(4+1 with no heatsink) I think yours is 3+1 with no heatsink.


This is a high quality VRM designed for overclocking. 8+2 phase and it sits under a heatsink.


Hopefully that will help you understand a bit. Most motherboards have features on them to protect themselves if CPU power draw is too high(they'll decrease CPU speed without telling you).

Edit: Oh I didn't read you whole post....51C isn't lots of headroom as Earthdog has said.

EarthDog said:

4C or so is plenty? I recall people saying to keep these around 55C... 60C at the most (as RgoneSTER! posted already) as you tend to run in to instability with temps above 55-60C...

Click to expand...

It is when you consider the temperature increased by a grand total of 1/2C when I clocked it from 3.4 to 3.9. I should, in theory, be able to hit 4.

Hicksimus said:

You may have lots of headroom on CPU temps and voltage but you don't have lots of headroom because of the motherboard.

I know it has overclocking options but it is not for overclocking. You're actually pretty lucky to get 3.9GHz out of it and even then I'd suspect it's throttling.

It all boils down to the VRM. Your motherboard has a very skimpy one likely made with low quality parts as well. When they say 125W CPU support on a cheap board they mean it will handle the average user's computer demands with a 125W CPU and be fine. They're really playing to the literal meaning of what they say on cheap motherboards, where as the more expensive boards will often say 140W CPU support but can deal with 350W of CPU draw all day long.

Here are 2 things that were helpful for another user to get what I'm talking about. He had the next up motherboard from yours and found it hard to hit 3.8GHz.

This is the power draw scaling of your processor. I drew in the 140w line which is more than what ASUS intends your board to deal with already(but was what the other user's board was for). The spike at 3.9GHz is because they upped voltage from 1.4V to 1.5V.

This was the VRM on his board(4+1 with no heatsink) I think yours is 3+1 with no heatsink.

This is a high quality VRM designed for overclocking. 8+2 phase and it sits under a heatsink.

Hopefully that will help you understand a bit. Most motherboards have features on them to protect themselves if CPU power draw is too high(they'll decrease CPU speed without telling you).

Edit: Oh I didn't read you whole post....51C isn't lots of headroom as Earthdog has said.

Click to expand...

Good to know.

I did build this on the cheap (<£500) and was just seeing how far I could push it. High quality anything wasn't on the menu (apart from PSU, of course). I am still going to try for 4.0GHz because 3.9GHz was acheived with nothing but the Multiplier and I've yet to touch the voltages, with the jump from 3.4 to 3.9 increasing the temperature by 1/2C at most with no throttling from what I can see in CPU-Z and HWMonitor. If a voltage increase got your friends board from 3.8 to 3.9, it may do the same to get me from 3.9 to 4. Only if its stable with no throttling with I actually try for 4.1+. Only three hours left (of a brain smashing 12) in my shift so not long to wait before I can give the voltage a whirl.

If you left the voltage on AUTO, YOU didnt touch it but its likely still adjusting the voltage. Get off AUTO.

EarthDog said:

If you left the voltage on AUTO, YOU didnt touch it but its likely still adjusting the voltage. Get off AUTO.

Click to expand...

It is left on Auto, true. But the MoBo seems to limit auto to the maximum value as set by AMD (1.45v - although CPU-Z reports 1.46 which is different from HWMonitor, AMD Overdrive and SpeedFan). I need to set the over-voltage value manually for it to go beyond stock, it appears.

And what is the stock load voltage? Again, leaving it on Auto adjusts voltage according to clockspeed/multiplier.

EarthDog said:

And what is the stock load voltage? Again, leaving it on Auto adjusts voltage according to clockspeed/multiplier.

Click to expand...

All the numbers I'm quoting are for heavy load when running Prime95 or AMD Overdrive. Googling around, different people report different stock voltages ranging from as low as 1.32 to 1.45 - the latter being the highest reported stock voltage I can find.

At stock settings I constantly flicked inbetween 1.416 and 1.45 according to CPU-Z and Hardware Monitor. It sticks at 1.45 now that I've clocked it to 3.9GHz.

I will, of course, set a manual voltage, starting at 1.40 and heading up from there in 0.025 steps (if possible, I think my BIOS has far more servere stepping than this).

Well that was fun!

I increased the core voltage from 1.37v (actual default, Windows reported incorrectly against the BIOS) to 1.476v and bumped the multiplier up to x20 (4.0GHz!).

Tweaked my RAM timings to:
CAS Latency: 9
RAS to CAS: 9
RAS Precharge: 9
tRAS: 25
Bank Cycle: 34

and upped the DIMM voltage to 1.55v. I also switched the Q-Fan mode to Turbo.

Ran Prime 95 for a solid hour and ended up with a grin on my face.




The interesting thing is, although it had an initial spike of 129F, once the fans kicked in the temperature actually dropped! Hovered around 52/53C with ocassional spikes to 54C (128F).

The fans also got no where near their maximum RPM (1500) and, bizarrly, running Prime95 the voltage never went over 1.440v - opening applications like Firefox are what caused the spike to 1.476v. Sadly, reducing the voltage to 1.45 in the BIOS caused major instability and blue screens, even though prime never hits that value. The extra overhead must allow for the likes of the LLC to run effectively or something.

But either way, 4GHz system well within safe temperatures. I'm going to try and push the HT Ref clock now, to see if I can hit 4.1GHz.

Another 1 hour burn in. Increased the multiplier to x20.5, the VCore to 1.53v and the RAM voltage to 1.57v.

Once settled, the temp twitched between 54/55C after the initial spike to 56 during stress testing.



I need to do a 2 hour test but that can wait for another day. So for now, I'm considering the system stable at 4.1GHz.

benanderson89 said:

IRQ_NOT_LESS_OR_EQUAL

Click to expand...

Looks like a core wall. Reminds me of old school overclocking, my T'bird 900 did the same thing at 1050 Mhz.

Always fracking "STOP: IRQL_NOT_LESS_OR_EQUAL" within 10 minutes of Need For Speed III, back in 2002, lol.


(Even though it may be a PSU issue.)

Even though I never gotten a BSOD when Prime95 failed, even when it failed hard.


Intel Burn Test however, has failed with that message before with one of my early E2180 OCs under XP 32-bit. Vcore increase solved the problem.

"IRQL_NOT_LESS_OR_EQUAL" means you need more Vcore or you reached a core brick wall.

If the PC just froze on you instead of an error message, that can mean a core brick wall.

-> Linpack is the best test for multi-only OCing! If you get real high temps with LinX and the like, then you need a better cooler or more air flow into the case.

with my denebs, 955's and 965's i have, as a rule found that they have all worked best at around 3.8.
I have pushed them much higher but only as moon shots.

I have never seen a "brick wall" with them, I have pushed my 955 to 4.5 and that was heat limited at that time.

found it, http://www.overclockers.com/forums/showthread.php?t=717371&highlight=955be

Looks like a lot of AMDs have a wall at 3.8 Ghz or roughly that.

I've been sitting in a sweet spot at 3.9 for a while. Ticked it up there to this speed without a hiccup. Tried a couple of ways to get to 4.0+ x20 multi-nope. Upped the bus to 206-nope. Workers stopping with the x20 and without a voltage jump on the memory, got a blue screen when I upped the FSB.

I know I'll eventually get impatient and have to run at 4.0+ For now I'm pretty happy with a $100 CPU.

3.9 is the brick wall for my chip even with the huge amount of temp and voltage headroom I have left over, soon as I hit that x20 multiplier I can't safely feed it whatever voltage it would take to hold 4ghz and i've tried everything, even had it 10 degrees F in my apt and it still instantly crash on prime95.

I learned to live with the fact that my chip is a below avg clocker, even bumping my cpu/nb to 2600 takes a boost from 1.17 to 1.35v, for gaming those extra mhz don't amount to anything which is why I can tolerate not sitting at 4.0ghz for a little while longer until my 8350 is ordered. When I 1st got this chip I was using an asus m4a77td mobo which had no cooling on the mosfets and the voltage fluctuations were so out of whack it was impossible to find a happy medium which is why I figured I wouldn't see 4ghz but a crosshair iv forumula and a sabertooth 990fx later I'm still sitting at the same settings as I was when I started lol This is probobly the longest I've ever stuck with the same cpu and it has served me well but it'll be happy sitting in my dedicated server, I just hope its replacement isn't a let down.