Hit a 4Ghz wall @1.35v core shutdown, ideas?

Printable View

06-15-2010, 03:00 AM
R37ribution

1 Attachment(s)

Hit a 4Ghz wall @1.35v core shutdown, ideas?

With the setup below I can get a 3.9Ghz with a pretty low CPU-Z vcore of 1.24v. I used the 3 step guide to overclock your i3/i5/i7 on overclockers.com to get to this point. Before I got to these voltages I hit a bclock wall at 192 in step one.

Once I got to 3.9Ghz I tried starting over on step one with my current 3.9Ghz voltage settings and disabled SpeedStep and Hyper-Threading. My max bclock was 220 in step one this time with prime95 for 10 minutes with a QPI voltage of 1.39v. I dropped the bclock 2Mhz and kept the 2:8 ratio for a 872Mhz DRAM frequency w/ 7-8-7-20 timings. I popped in MemTest86+ and was able to make 3 passes w/out changing the DRAM voltage of 1.53756v. I then moved on to step three and raised the multi back to 21 and lowered the bclock. Again I hit a wall at 185 until I raised the vcore to 1.35v which allowed for 190. I loaded up prime95 and it reached temperatures of 87C on core 1 dropping in order to core 4 which hit a max of 83C. After about three minutes core 4 shutdown. I could tell because the core temp dropped to 63C and the utilization in task manager went to 0% on virtual cores 7 and 8. I tried dropping the vcore down a little lower but it wasn't stable. I tried going to 205 or 215 because I remembered about bclock holes but it still failed.

So here is my question: Why such a huge vcore difference just to go up 100Mhz from 3.885Ghz to 4Ghz? Do any of my settings look wrong? I ordered two Delta 210cfm 120mm fans to upgrade from my 78cfm Noiseblocker to a push pull config on the radiator to help with my cooling.

Also, does the fact that my core temps drop in sequential order from core 1 thru core 4 indicate a problem with the seating of the H50? I am currently running an H.264 encode of an HD mpg that includes deinterlacing using Handbrake at the 3.9Ghz settings below and have had RealTemp running the whole time. The max core temps are 76, 74, 72, 72 and are currently 73, 72, 70, 70 and now 73, 72, 70, 68.

I run both CPUID HW Monitor and RealTemp. CPUID reports core temps 3-4 degrees below that of RealTemp. I understand that CPUID updates less frequently but core 1 for example is consistantly at 73-74C but in CPUID it never goes above 71C. Is one more accurate than the other?

My setup:
Xeon W3520 - Corsair H50 w/ 120mm 78cfm Noiseblocker
Asus Rampage II Gene - Spotcool on the NB, removed NB & SB thermal paste and replaced w/ Arctic Silver 5
Corsair Dominator GT 2000 7-8-7-20
2x Geforce 9800GT in SLI
PC Power & Cooling TurboCool 860 ESA

My prime95 (1 hour max core temp 78C) and memtest86+ (2 pass) settings:
bclk - 185
multi - 21
C1E - disabled
Hyper-Threading - enabled
SpeedStep - enabled
TurboMode - disabled??? (doesn't show up under SpeedStep and multi never hits 22)
C-state - disabled
CPU LLC - disabled
QPI LLC - disabled

Voltages:
CPU - 3.895Ghz @ Auto (1.29375v in TurboV) to support SpeedStep (CPU-Z 1.24v at load and 0.96 at idle)
QPI - 3340Mhz @ 1.35000v
DRAM - 927Mhz @ 1.53756v
Uncore - 3710Mhz
PLL - 1.81592v
IOH - 1.33866v
IOH PCIe - 1.51106v
ICH - 1.11341v
ICH PCIe - 1.51106v
06-15-2010, 12:06 PM
famish99

Your prime95 temps are far too high, that's what's causing the thermal shutdown. If I had to guess what your number one reason for failing, it would be due to thermal reasons. As you increase temperature, higher voltages are necessary for stability. Unfortunately, higher voltage produces more heat, and the vicious cycle continues. Even although the die is spec to 100C, you can start seeing instability in the high 80's because you can't account for spikes that the thermal sensor isn't quick enough to notice. Although it may reduce your temperatures may drop a little, beefing up the fan will do little to adding stability; you most likely need to try reapplying thermal paste or reseating it. I hesitate to say that it's inadequate because many people have gotten higher frequencies on air, but mileage always varies. I've tested chips on my rig that don't break 4ghz under any reasonable amount of voltage on my custom water loop. Just because somebody got x frequency on the 930, doesn't guarantee yours always will. In the end, Intel is just trying to make money, as long as it passes the spec, they're gonna sell it. They aren't obligated to sell you a part that goes 4ghz if you're only paying for 2.8 ;)

Other than that, your DDR voltage seems a bit low as well, but if it passes memtest, then it's most likely not the cause. That doesn't mean your memory subsystem is broken (ie your cache, imc could be failing). But most of what you've said point towards thermal instability.
06-15-2010, 12:41 PM
OC Nub

I agree with famish99, your temps are to high and may cause instability. Your ram is rated at 1.65v, try 1.640v, I have the 2000GT's also, I think 1.53v is to low. Also your IOH is high, I have a gigabyte board but if that is the same as IOH core try 1.140-1.160v and link that with your QPI PLL so both are running at the same voltage. Try 1.40v-CPU PLL, that may help lower temps.
06-15-2010, 01:04 PM
famish99

Quote:

Originally Posted by OC Nub

Also your IOH is high, I have a gigabyte board but if that is the same as IOH core try 1.140-1.160v

It doesn't sound like his system is shutting down due to BSOD which suggests that his platform stability is fairly solid. But I do agree with increasing the DDR voltage. Most of the boards I've used required 1.3v IOH to get any sort of stability on bclk > 190MHz. But I've never tried reducing CPU PLL voltage as effort to bringing down heat. I always figured that would increase the amount of jitter and reducing stability, but if you've gotten it to work then that's interesting to know.
06-15-2010, 02:12 PM
OC Nub

These settings were suggested to me by a couple guys in the forums and have worked very well. I am currently running a very stable 210x21-4400mhz HT enabled with 1.140v IOH core, 1.140v QPI PLL, and 1.40v CPU PLL.

Here is a very short LinX run with ET6 installed to show current settings.

http://img535.imageshack.us/img535/4...ll1140linx.png

Uploaded with ImageShack.us
06-15-2010, 02:58 PM
famish99

I'm gonna have to try that then.
06-16-2010, 11:58 AM
famish99

So I have the same board as the OP, which for some reason doesn't allow setting of the CPU PLL below the default 1.8v and the IOH PLL option doesn't even exist. So regarding those recommendations, the OP is probably out of luck concerning setting voltages that way.

Hopefully you can find another solution to your heat issues.
07-12-2010, 04:13 PM
R37ribution

Well, great news!

I reapplied the Arctic Silver 5 and added the Delta 210cfm fan and now I am creating this post while running Prime95 (large FFTs) for over 13min wth temps never going above 71C on the hottest core (max temps 71, 69, 67, 66) at 4Ghz!!!.

I still have the core temps dropping a degree or two in sequential order like it was in my first post before reapplying the thermal paste so I'm not sure how which made the most improvement - the reapply or the fan upgrade.

I've seen people replace the stock radiator on the H50 with 240mm ones. I noticed that swiftech has 120mm stackable brass/copper radiators that I might replace this one with. All the reviews I have read on the H50 indicate that the pump/water block is more than adiquate for an upgraded radiator and that the radiator is in fact the weakest link other than the stock fan.

http://img541.imageshack.us/img541/8...ta120mmfan.jpg
07-12-2010, 05:33 PM
Johnny87au

geebus, with that delta fan is your case hovering?
07-12-2010, 08:16 PM
R37ribution

Quote:

Originally Posted by Johnny87au

geebus, with that delta fan is your case hovering?

It sure as hell sounds like it's tryin to take off lol. I was expecting it to be loud and it is... when its only running at 3500rpm! When it revs all the way up to 5800-6000rpm while running a stress test in Prime95 its insane haha.

Quote:

Originally Posted by famish99

So I have the same board as the OP, which for some reason doesn't allow setting of the CPU PLL below the default 1.8v and the IOH PLL option doesn't even exist. So regarding those recommendations, the OP is probably out of luck concerning setting voltages that way.

Hopefully you can find another solution to your heat issues.

I agree, this board is in desperate need of lower than Auto voltages. Asus needs to add this support in addition to the CPU PLL and IOH PLL. I would really love to try voltages near that which OC Nub suggests.

Caution: very important for anyone buying the Delta 120mm 210cfm fan w/ a 3 or 4 pin motherboard fan header connector. This baby draws 2.7 amps! I damn near fried my motherboard when I first fired up Prime95.

I was actually walking my girl friend thru tweaking the BIOS settins while I was board one night at work (more like anxious to see how much of a difference it made). I had her load up the OC profile I was tweaking in the first post and loaded up Prime95 thru Remote Desktop. About 20 seconds after it loaded up, it blue screened.

I figured it was just because my voltages were off. I had her edit them for me and tried it again. This time she said it smelled like I "...let the smoke out." I immediately thought of the power this thing pulls and had her move it to a chassis header that allowed a duty setting - we set it to 50%. It still blue screened, but not before I noticed that the voltage in CPU-z was way lower than normal (1.20v vs. 1.26v). I ordered a PWM Splitter which allowed me to pull power from the power supply while still allowing for the PWM and RPM wires. The PWM wire goes to all 3 fans and the RPM wire is connected to the fan cable labeled CPU.
10-09-2010, 09:31 AM
R37ribution

Alright, I'm at 3.9Ghz with a 185blck and 927Mhz memory @ 7-8-7-20

Whenever I plug in my UPS and test with Prime95 I get a blue screen within a minute. If I don't have it plugged in I can run Prime95 AND 3DMark06 similtaneously as long as I want and it will never blue screen.

So at this point do I need to OC my IOH, ICH, PCI or CPU PLL? I haven't done much in that area and I'm not sure how to tell where I should start.

Thanks for the comments.

http://valid.canardpc.com/show_oc.php?id=1425910

http://img697.imageshack.us/img697/4117/39ghz.jpg
10-09-2010, 09:40 PM
Neova

I recently went through the same process using the same excellent guide but tweaked a i5-760 to 4.2 GHZ Linpack, Intel Burn Test stable with max settings.

When tweaking, you should not use auto because it may be too much or too little for 100% load stability.

My system settings - should be similar but without HT:

Asus P7P55D-Deluxe board, i5-760, 8 GB 1333 ram at 1200 8-8-8 settings. Corsair H50 water cooling just like you but with exhausting setup and case fan pulling, stock fan pushing air out:

Blck is locked at 200 to max out my ram - I have holes from 201-209, then 211-219. I can set blck at 210 and 220 but is not stable.

cpu @ 4 ghz 100% IBT MAX stable - 1.334 volts
cpu @ 4.2 ghz 100% IBT MAX stable - 1.464 volts <--- over 0.13 volts more!
cpu @ 4.3 ghz NOT IBT stable but plays games with no issues - 1.424 volts
IMC - 1.4 volts but jumps to 1.42 on full load
Dram - auto @ 1.5 - my ram is underclocked from 1333 to 1200 and makes no difference when raising ram voltage.
PLL - auto @ 1.8 volts
LLC is ENABLED to help with vdroop supposedly
PCH - 1.1 volts for stability - P55 southbridge to improve stability for streaming content during full load use
Turn off QPI and PCIe spread spectrum and all power saving modes
Everthing else on auto for voltages which is near stock.

@ 4.2 get near 100 C at 100% IBT load but never gets above 80 c when playing games
@ 4.0 gets near 80 C at 100% IBT load but never gets above 70 c when playing games

I had tested with GTA 4 at max settings which shows it uses all 4 cores, 4 GB of ram and sets my GPU at 100% load.
I had also test with Metro 2033 but it seems to max the gpu, and uses maybe 2.5 cores at less than 80% cpu load.

I'm setting my cpu back down to 4 GHZ since no games uses all 4 cores at 100% yet.

Once you find all your MAX stable settings, then you can dial down the voltages.
11-12-2010, 04:28 AM
R37ribution

Linpack (IntelBurnTest) 30x stable @ 4.2Ghz DRAM 1.0Ghz

I've made some pretty good headway...

I have just successfully ran thru 30 iterations of Linpack via IntelBurnTest w/ Stress Level set to "Very High" 4096MB:

Code:

---------------------------- IntelBurnTest v2.50 Created by AgentGOD ---------------------------- Processor: Intel(R) Xeon(R) CPU W3520 @ 2.67GHz Clock Speed: 4.00 GHz Active Physical Cores: 4 Total System Memory: 6135 MB Stress Level: Very High (4096 MB) Testing started on 11/11/2010 8:57:43 PM Time (s) Speed (GFlops) Result [21:00:16] 131.628 58.5566 3.161416e-002 [21:02:46] 131.291 58.7071 3.161416e-002 [21:05:17] 130.787 58.9335 3.161416e-002 [21:07:47] 130.836 58.9112 3.161416e-002 [21:10:17] 130.666 58.9878 3.161416e-002 [21:12:49] 132.378 58.2249 3.161416e-002 [21:15:21] 132.218 58.2955 3.161416e-002 [21:17:51] 131.059 58.8108 3.161416e-002 [21:20:21] 130.842 58.9084 3.161416e-002 [21:22:53] 131.695 58.5269 3.161416e-002 [21:25:23] 131.048 58.8160 3.161416e-002 [21:27:54] 131.356 58.6782 3.161416e-002 [21:30:25] 130.922 58.8724 3.161416e-002 [21:32:55] 131.194 58.7506 3.161416e-002 [21:35:27] 132.942 57.9780 3.161416e-002 [21:37:58] 131.329 58.6900 3.161416e-002 [21:40:28] 130.628 59.0052 3.161416e-002 [21:42:59] 130.801 58.9271 3.161416e-002 [21:45:30] 132.298 58.2604 3.161416e-002 [21:48:01] 131.033 58.8226 3.161416e-002 [21:50:31] 130.690 58.9773 3.161416e-002 [21:53:02] 131.386 58.6648 3.161416e-002 [21:55:32] 131.183 58.7553 3.161416e-002 [21:58:03] 131.701 58.5243 3.161416e-002 [22:00:34] 130.850 58.9049 3.161416e-002 [22:03:04] 130.813 58.9217 3.161416e-002 [22:05:34] 130.723 58.9622 3.161416e-002 [22:08:04] 130.616 59.0104 3.161416e-002 [22:10:34] 130.176 59.2098 3.161416e-002 [22:13:04] 130.675 58.9838 3.161416e-002 Testing ended on 11/11/2010 10:13:04 PM Test Result: Success. ----------------------------

I don't understand why the report shows my CPU at 4.0Ghz when its running at 4.2Ghz. Perhaps it doesn't count the 21st multiplier because its for turbo?

After running 30 iterations my max core temps are 71/67/67/62 C. I'm still using a modded H50 for cooling. The only thing that's stock is the water block. I'm running a single 120mm Black Ice GTX Xtreme rad with that Delta 210CFM fan I described earlier. I wish I would have got a screen shot right after Linpack finished for you guys but oh well. :shrug:

The key to getting this stable was the IOH voltage. I tried the 1.3v and 1.35v that the 3 Step Guide to Overclocking recommended but that was WAY to high for this board - It made it completely unstable. The JMicron and Intel Option ROM were constantly acting up. At first I thought I didn't have enough voltage but when I hit 1.37841v on the IOH and it still wasn't stable I figured it was time to try something different.

Oh, and I have Intel SpeedStep turned on so it drops to ~1.0v at idle. My CPU voltage offset is 0.15v. My next step will be an attempt at enabling Hyperthreading... Or maybe the 4.5Ghz club :)

Anyways, I'm pretty happy!

http://img51.imageshack.us/img51/809...hzdram1000.jpg
11-12-2010, 06:27 AM
SandyTunnel

Quote:

Originally Posted by R37ribution

I reapplied the Arctic Silver 5 and added the Delta 210cfm fan and now I am creating this post while running Prime95 (large FFTs) for over 13min wth temps never going above 71C on the hottest core (max temps 71, 69, 67, 66) at 4Ghz!!!.

I just checked my temps. Max is 42C @ 5.0 Ghz. Not really running anything stressful right now, Prime1024 on 2 cores only of my 6.
11-12-2010, 07:26 AM
PatRaceTin

i use both Prime95 + LinX
11-16-2010, 01:37 AM
Neova

R37ribution - your new settings are looking good bro! I should turn Speedstep back on to lower vcore during less than 100% use to save power. Did you lock your multi at 21 and blck at 200 in the bios, and speed step still works?
11-16-2010, 03:16 AM
R37ribution

Quote:

Originally Posted by Neova

R37ribution - your new settings are looking good bro! I should turn Speedstep back on to lower vcore during less than 100% use to save power. Did you lock your multi at 21 and blck at 200 in the bios, and speed step still works?

yea, that's exactly what I did. I had to set the "CPU Voltage Control" to a "relative offset" and the vcore to 0.15 (offset from 1.20v). The multiplier drops to 12x and the CPU vcore to ~1.0v
11-16-2010, 09:05 AM
OCBalmang

you shouldn't have any speed step or Ce1 EIST or HT on, only turn those on after you've achieved a stable clock, that way you'll be able to determine where you went wrong easier and tweak around it with minimal confusion... also CPU PLL is way too high should be around 1.65 not 1.8 last i looked at bloomfields... Could you give us a bios update right now complete run down of your settings in a code window... That way we can better assist you with Overclocking, IOH was pretty high too... just reply to this with your New Settings ... also speed step would be why your getting 4GHZ not 4.2GHZ if the clock was fine tuning itself down by .5 ... which happens for my C2D chip occasionally .... just an idea... remember read through my first part about disabling features while OCin, once stable enable one by one

Late post lol! internets screwed up today :(
11-16-2010, 03:21 PM
acebmxer

1.8 is the default voltage for bloomfield cpus cpu pll. So find the system more stable at lower cpu pll volts even as low as 1.3v dont push it more then 1.86. 1.9 i think is MAX.
11-17-2010, 08:26 PM
R37ribution

Quote:

Originally Posted by OCBalmang

you shouldn't have any speed step or Ce1 EIST or HT on, only turn those on after you've achieved a stable clock, that way you'll be able to determine where you went wrong easier and tweak around it with minimal confusion... also CPU PLL is way too high should be around 1.65 not 1.8 last i looked at bloomfields... Could you give us a bios update right now complete run down of your settings in a code window... That way we can better assist you with Overclocking, IOH was pretty high too... just reply to this with your New Settings ... also speed step would be why your getting 4GHZ not 4.2GHZ if the clock was fine tuning itself down by .5 ... which happens for my C2D chip occasionally .... just an idea... remember read through my first part about disabling features while OCin, once stable enable one by one

Late post lol! internets screwed up today :(

Really? Are you guys reading anything in this thread or are you just mindlessly replying to get your post count up? Not two posts ago in my last update you will find I have my setup passing 30 Linpack tests @ 4.2Ghz with SpeedStep enabled; its even in a code window. I'd say I've "achieved a stable clock" as you call it. Oh and I've had HT disabled since I reached 4Ghz. In the same post you overlooked I stated "My next step will be an attempt at enabling Hyper Threading... Or maybe the 4.5Ghz club".

Quote:

Originally Posted by acebmxer

1.8 is the default voltage for bloomfield cpus cpu pll. So find the system more stable at lower cpu pll volts even as low as 1.3v dont push it more then 1.86. 1.9 i think is MAX.

Owners of Gigabyte boards must all be oblivious to the fact that Asus board owners cannot lower the CPU PLL less than 1.81v and don't even have an option for QPI PLL. I'm trying not to sound too irritated about such comments since there were screen shots above from another Gigabyte board owner. He suggested a lower PLL weeks ago.

Asus owners have asked multiple times for QPI PLL to be adjustable and a CPU PLL that allows for lower voltage, we've even asked for at least a table that tells us what the QPI PLL gets set to as we adjust the CPU PLL

I appreciate constructive criticism as I am well aware that I have plenty to learn. Hell, I haven't even entered the realm of "CPU Differential Amplitude", "PCIE Frequency", "CPU Clock Skew", or "IOH Clock Skew"; however, replies suggesting the same thing that has previously been suggested and found not to be a possibility just tells me you're too lazy to even bother reading the thread.
12-06-2010, 01:33 AM
R37ribution

Quote:

Originally Posted by R37ribution

Owners of Gigabyte boards must all be oblivious to the fact that Asus board owners cannot lower the CPU PLL less than 1.81v and don't even have an option for QPI PLL. I'm trying not to sound too irritated about such comments since there were screen shots above from another Gigabyte board owner. He suggested a lower PLL weeks ago.

Asus owners have asked multiple times for QPI PLL to be adjustable and a CPU PLL that allows for lower voltage, we've even asked for at least a table that tells us what the QPI PLL gets set to as we adjust the CPU PLL

Well, I ended up having to RMA my Gene II due to unstable QPI voltage - it would bounce +/-.05v constantly while I was in the BIOS (every time it updated it would be a different voltage at least .02-.04 or more within the +/- .05 of the setting). Convincing Asus Support that this was a problem with the board was quite a chore. They didn't have any stock of the Gene II to do an advance RMA so I went ahead and bought the Gene III which, much to my surprise, supports far lower CPU PLL and DRAM voltages.

With the Gene III I was easily able to OC to 4.2Ghz and even able to turn on HT at that speed. I've hit 4515Mhz 20x LinX stable (RAM set to 4GB) and am working on 4563Mhz with a CPU PLL of 1.53v. The Gene III is finally the mATX board to have. The only pet peeve with the board is that it is the ONLY ROG board Asus offers w/out OC Station support.

acebmxer, I looked at your 4.5Ghz post in the Intel 4.5GHz LinX Stable Club thread. I am curious what your voltage settings are after seeing your very low 1.36v vcore. In order to get my setup 5x LinX stable at 4563Mhz I've had to use a 1.65v core, 1.47v QPI, 1.64v DRAM, 1.54 CPU PLL. The vcore and QPI have a vdrop of about .06 so they read 1.58v and 1.41v respectively and a vdroop puts them around 1.56v and 1.43v. At 4515Mhz a setting of 1.59v vcore is sufficient. I have batch 390xxxxx (I'll have to get back to you on the rest).

At any rate, your BSOD stop codes in your signature have been invaluable in helping me reach this point. The only other ones I keep getting that seem to be IOH related are 0x0000000a 0x0000009c and 0x0000003b