Update: Problem solved

Problem:
I was getting opposite result of overclocking. I first experienced this when going from 2.6 to 2.8GHz - and then again when going from 3 to 3.2GHz. Both CPU and RAM was faster in task specific tests, but 3D related tests (for instance in LFS) would show that the performance was even slower than stock speeds!

Cause:
A glitch in the P5B motherboards causes the PCI-E to drop to 1x rate when overclocking - however not everyone experiences this glitch.

Solution:
No clear solution yet, but a BIOS update that fixes this issue is expected to be released. However, the BIOS can be "tricked" to get back to 16x rate.

Quote :

FAQ :

PCI-E 1x Link Width issue :
Happens when OC-ed. After restart link width is automaticlly set to 1x instead of 16x, which causes significant performace downgrade. Reason unknown. Happens mostly with ATI GPU cards. Possibly a BIOS protection measure, preventing percieved possible damage to GPU or motherboard, or a BIOS problem with proper video card initaition.

Symptoms :
PCI-E Link Width switches from 16x to 1x after a restart.

Solution :
- Unplug PSU, wait a few seconds, replug PSU.
- Unplug PSU, reset CMOS, replug PSU, renter all BIOS settings, save, shutdown, power on.
- Combining slightly lowering or uping voltages of PCI-E, Vmch, VCore and FSB can usually fix it. Also try
disabling ASUS C.G.I and PEG Force 1x., and changing values to PEG Link Mode.

For me personally, changing the PCI-E frequency to 110 did the trick.

Quote :

Every post that I've read in which the problem was solved, it was due to upping the PCie (105-110) until the link returned to 16x.

Thanks to everyone in this thread for trying to help me out. Also a big thanks to Elliot AKA cP Ice for the help and to make me understand overclocking better

Further info:
http://forumz.tomshardware.com ... uck-x1-ftopict228577.html
http://forumz.tomshardware.com ... figure-ftopict219755.html


----------

Specs:
Intel C2D E6400
P5B Deluxe
Corsair Dominator DDR2 PC-8500
Inno3D 8800GTS 640MB

Windows XP, SP2

Latest BIOS and gfx driver (v158.22)

AquaMark3:
Stock : 83341
@2800: 151789
@3000: 162004
@3200: 166917

3D rendering:
@2800: 2:41.6 mins
@3000: 2:35.8 mins
@3200: 2:28.7 mins

@3200: 2:16.3 mins (when "OC" was unsuccessful)

LFS, Vixen 01:
Stock : Avg: 48.383 - Min: 32 - Max: 69
@2800: Avg: 55.617 - Min: 39 - Max: 78
@3000: N/A
@3200: Avg: 61.633 - Min: 42 - Max: 89

@3200: Avg: 36.767 - Min: 25 - Max: 55 (when "OC" was unsuccessful)

LFS, Aston 6h:
Stock : Avg: 63.683 - Min: 40 - Max: 76
@2800: Avg: 72.667 - Min: 51 - Max: 93
@3000: Avg: 78.200 - Min: 54 - Max: 97
@3200: Avg: 83.267 - Min: 60 - Max: 102

@3200: Avg: 48.533 - Min: 31 - Max: 65 (when "OC" was unsuccessful)

3D rendering is 9% behind from my previous 3,2GHz OC attempt - But thats nothing compared to the main problem I had - on the good side, my idle temperature is 1 degree lower than what it used to be @ 2.8Ghz: ~47 degrees... and I don't know why, lol.

... yikes. "JakG!" (runs off to find Jak)

Its not entirely impossible that the heat produced by the graphics card just heats the processor up beyond its 'down throttle' point.. Use a utility like 'coretemp' (google) to see what sort of temperatures you have somewhat exactly..

Quote from srdsprinter :

... yikes. "JakG!" (runs off to find Jak)

Quote from Niels Heusinkveld :

Its not entirely impossible that the heat produced by the graphics card just heats the processor up beyond its 'down throttle' point.. Use a utility like 'coretemp' (google) to see what sort of temperatures you have somewhat exactly..

I hear what you are saying, but somehow I don't think this is the case here. I tried to simulate this - first I did some 3D rendering and immediately jumped into AquaMark03 - final result wasn't bad at all, but then continued into STALKER as it has more heavy gfx. According o Everest monitor, no throttling ever occurred.

Then I thought the CPU isn't getting heated enough (ehem), so I fired up TAT and started a test with 100% load. Computer went crawling even at desktop but I managed to get into AquaMark03 - although quickly realized it was pointless, because the CPU was so loaded that to test anything was basically impossible this way.

Anyways - TAT topped 74/72, while Everest displayed 67 @ CPU temp - and still no throttling occurring.

Core Temp:
I use TAT (Intel Thermal Analysis Tool) and Everest instead - I'm told Everest isn't accurate, but I'm not sure about that myself - seems to display similar temperatures as TAT and Core Temp. However I'm suspecting if I really did reach 67 degrees (CPU temp - not core... thats allot).

But the main problem isn't to have something to display temperatures precisely. Main problem is to observe them before CPU cools down, and it does so very fast.

And oh.. no crashes even..

I've updated my first post with some results @ 3GHz - it's running fine at this frequency. And before anyone suggest I hit the limit, I might add that I experienced the exact same thing when I went from 2.66 to 2.8GHz - however somehow I managed to fix it.

are you sure your pci-e clock remains at 100 and the gpu doesnt clock itself down ?

Quote from Shotglass :

are you sure your pci-e clock remains at 100 and the gpu doesnt clock itself down ?

How can I make sure? PCI-E is set manually to 100 in BIOS, but I have no idea about GPU clocking itself down or not.

Quote from srdsprinter :

... yikes. "JakG!" (runs off to find Jak)

OK, i'm here, but Intel overclocking ain't my strong point.

First point of the wierd timing in 3DMark - i've had this on the Mars test in 3DMark when it goes >1 FPS (usually on the CPU test) - not sure how that helps though!

What temps are your GFX cards running at?Have you locked the PCI-e, PCI Bussess etc?

What vCore did you use to get to 3.2 GHz? Whats your cooling?

What Power Supply do you have?

Quote from r4ptor :

Anyways - TAT topped 74/72, while Everest displayed 67 @ CPU temp - and still no throttling occurring.

Core Temp:
I use TAT (Intel Thermal Analysis Tool) and Everest instead - I'm told Everest isn't accurate, but I'm not sure about that myself - seems to display similar temperatures as TAT and Core Temp. However I'm suspecting if I really did reach 67 degrees (CPU temp - not core... thats allot).

But the main problem isn't to have something to display temperatures precisely. Main problem is to observe them before CPU cools down, and it does so very fast.

Yer I'd use TAT over Everest, I've heard of far too many dodgy readings from Everest.

74/72 is pretty hot, as Jak says, what's your V Core? I feel with better cooling you could get that a lot cooler - I can't quite remember but I think C2D throttling comes in somewhere around there :S it may be 85 actually, not sure.

Agreed with the cooling down quickly thing, take them off load and within 5 seconds they can be back to normal temps.

even 70 is too high even for C2D so it might throtling.Some of the C2D start throtling even under 70 degree!! Do you have aftermarket cooling on the CPU or the standard box cooler bundled with the CPU?Also when you get BSOD it usually mean you pushed the CPU or memory too far.

LFS shows at 3.2Hgz that you have much lower FPS and LFs is very CPU bottlenecked game.It clearly shows your CPU is throtling at 3.2Ghz.

Whats your vcore for CPU?

Also on many motherboards when you set PCI-E to 100Mhz it will not lock.You need to put it rather to 101Mhz and it will lock down on all motherboards.Especially on MSI motherboard the PCI-E doesnmt get locked if manually moved to 100Mhz.

Is your graphic card overclocked?

Thanks for the replies, guys.

Ran LFS for 17 minutes in windowed mode and then took some readings:

GPU is running @ stock frequencies - can't (shouldn't clock it).

Temperature, cores:

Core 0: 59/59/59/59 - TAT/SpeedFan/Core Temp/Everest
Core 1: 56/56/56/57 - TAT/SpeedFan/Core Temp/Everest

Everest is generally +1 degree off @ Core 0.

Temperature, CPU:
51/51 - PC Probe II/Everest

Temperature, GPU:
66/65 - nTune/Everest

Throttling, CPU:
+/- 0.4 or so MHz in CPU-Z
Core Temp/TAT: no changes - if they even show any.

Throttling, GPU:
nTune: no changes - if it even show any.

PCI-E frequency changes:
Don't know how to see this

Voltage:
vCore: 1.35v
RAM: 2.25v

I'm uncertain about what these should be, but they are currently at:

FSB Termination Voltage: 1.40v
NB VCore: Auto
SB VCore: 1.60v
ICH Chipset Voltage: Auto

BIOS settings are same as before, except RAM voltage which was set to 2.15v - I forgot to change my timings this time. Couldn't boot (BSOD) and then went into BIOS and upped RAM voltage - still didn't occur to me that I should change timings.

PSU: Chill Innovation CP-510A - 510W - Couldn't find a better link - I've had it for 3-4 years.

Heatsink:
Scythe Infinity

Screenshots:
LFS running for 17 minutes (the readings mentioned above)
30 minutes after having paused ... inimized LFS (not closed)
TAT 100% Load for 1 minute Auch!

My temperature have always been too high - about 10-20 degrees C higher than what I should be getting according to what I'm reading around the web. Doesn't matter if the CPU is clocked or running as stock.

Computer have been idling (expect for me typing here) for 5+ minutes now:

Temperatures are 57/55 (Core @ TAT) and 50 (SpeedFan).

I changed to 3GHz last night, and been running with it for 12+ hours. Did some benchmarks and even watched 2 movies while doing a 7+ render :P Anyways.. as mentioned, my idle temps are generally too high. @ 3GHz I'm idling @ 48 - with 2.66GHz I'm around 44 degrees.

Lowest ever was 37-39 with stock speed @ 2.13GHz.

Have you tried re-seating the cpu cooler?

Also, i've read lots of people having problems with concave tops of their cpu, causing chronic overheating, hope that's not ur case.

Tried re-applying the cooler?

Tried a different (better) PSU? it could be a under-adequate PSU

Quote from srdsprinter :

Have you tried re-seating the cpu cooler?

Also, i've read lots of people having problems with concave tops of their cpu, causing chronic overheating, hope that's not ur case.

Quote from Jakg :

Tried a different (better) PSU? it could be a under-adequate PSU

Yes, I refitted the cooler few days ago - CPU temp fell by ~2 degrees - far from what I was hoping for.

srdsprinter: Yeh, I've read about that after a teammate pointed it out to me. I might do some tests to see how flat the CPU and the cooler are, but I won't be "lapping" for now.

Jakg: No, haven't tried with a different PSU - don't have any near me.

Just changed RAM timings to 5-5-5-15 and lowered vCore to 1.3375. Performance seems to be same as before (it's obvious enough for me not to bother doing any benchmarks) - CPU temp is lower by 1 degree.. "wow" :P

Edit: I wouldn't be surprised at all if my CPU surface is like Grand Canyon. When I first dismounted the cooler (RAM went dead 2 weeks ago - defect, not caused by OC.. didn't do any OC until this weekend for the first time ever), I noticed the 4 corners had left heat marks (not severe) on the cooler. I'm now wondering if its because the middle area is barely touching the cooler at all :|

i was afraid of that. heard it is a fixable, but not fun thing to have.

As jakg mentioned, you have a decently stable power supply?

Hmm those are really strange temps for a 1.28v core :S I'm really suprised by that, especially as your GPU is at normal temps, so I ruled out excessively hot weather and/or high case temp.

Don't wont to insult your intelligence, but you are using thermal paste yes? It's just I've seen people forget/not bother putting it on before and wonder why their temps are so high.

Regardless, I'd go with arctic silver or ceramique paste - not that that is the problem here, but will help.

Quick edit: HOW the hell is it running 3.2ghz at 1.28v?! Meh scrap that, its possible.

Also you should be seeing temps of around 50 - 55 with that overclock @ load

Are you using the fan on the scythe? That should be a very effective cooler if you are.

Quote from srdsprinter :

As jakg mentioned, you have a decently stable power supply?

I think it's stable - first time I wondered about this was when I bought new mobo, cpu etc, but wondered if I should keep my old PSU. I consulted with a friend, who didn't seem to be worried about my PSU - and he know more about hardware than I do.

But it might still be a good idea to try with a different one though.

Quote from pb32000 :

Hmm those are really strange temps for a 1.28v core :S I'm really suprised by that, especially as your GPU is at normal temps, so I ruled out excessively hot weather and/or high case temp.

I'm told that nothing can monitor C2D vCore yet - dunno, but it's always lower than what I set it to in the BIOS - I'll check in my next reboot what BIOS monitor tells me about vCore - because I think it shows a lower voltage there as well.

Quote from pb32000 :

Don't wont to insult your intelligence, but you are using thermal paste yes? It's just I've seen people forget/not bother putting it on before and wonder why their temps are so high.

Regardless, I'd go with arctic silver or ceramique paste - not that that is the problem here, but will help.

:P

I'm using Zalman Super Thermal Grease ZM-STG1. The shop didn't have anything else. When I asked why, he replied that this simple is the best...

Quote from pb32000 :

Also you should be seeing temps of around 50 - 55 with that overclock @ load

Yeah I should - right now I changed back to 3GHz, and doing an Orthos small fft test in the BG. By temp is @ 63 :| with a TAT 100% load test, it would be higher.

Fortunately the toughest stuff I do as normal use, is 3D rendering. Topped 55 degress after 2 mins in a test I just did (stil on 3GHz).

Ambient temperatur today is actually a bit hot. I would say that means +1 on CPU temp.

Quote from pb32000 :

Are you using the fan on the scythe? That should be a very effective cooler if you are

I did untill I started to overclock and ran into this heat probem. I'm now using Enermax Warp UC-12AEBS because I can alter the RPM. But turning it all the way up to ~2400RPM only lowers temperature by 1 degree... after long time.

Which also leads me to suspect that my CPU has a Grand Canyon.

Edit: This guy is using vCore @ 1.6v, but CPU-Z is showing 1.2v..

From here...

Ahh yer I see what you mean about unreliable Vcore readings, I'm guessing that guy didnt get a droop of 0.4v lol.

Maybe if its not too hard it'd be worth getting it replaced. It's faulty really if it's been manufactured with a concave IHS

i doubt its your psu, if you're not getting random crashes or fried memory. Hopefully its 500+ Watts though.

You don't have the Intel equivilant of the AMD Cool N' Quiet, which can down clock your CPU as it see's fit to, on?

Everything above we've covered is purely temperature based, but it Appears you're not in the zone that causes precautionary cpu slowing... You're just gettting weird slowdowns with your overclock...

Quote from srdsprinter :

You don't have the Intel equivilant of the AMD Cool N' Quiet, which can down clock your CPU as it see's fit to, on?

Everything above we've covered is purely temperature based, but it Appears you're not in the zone that causes precautionary cpu slowing... You're just gettting weird slowdowns with your overclock...

Au contraire - the 6400 throttle speed is roughly 75-80 degrees AFAIK. I wouldnt be suprised if this slowdown is due to temps.

Also the C2Ds automatically reduce their multiplier (on mine to 6x) when idle. Don't know if you can turn that off.

Quote from srdsprinter :

i doubt its your psu, if you're not getting random crashes or fried memory. Hopefully its 500+ Watts though.

Yeah, it is - 510W

Quote from srdsprinter :

You don't have the Intel equivilant of the AMD Cool N' Quiet, which can down clock your CPU as it see's fit to, on?

No, I turned everything off that I didn't need inthe BIOS, according to a C2D OC guide.

When I had my old system (AMD), I could see it throttling, but as mentioned earlier, I can't see anywhere that the current CPU is throttling, unless I force it to via TAT - thats still not close to auto throttling.

Quote from srdsprinter :

Everything above we've covered is purely temperature based, but it Appears you're not in the zone that causes precautionary cpu slowing... You're just gettting weird slowdowns with your overclock...

Thats what I'm thinking myself - I'm not an OC guru, and it's been years since I did it last time. but something is just not right here.

I didn't mention this, but I experienced the exact same thing when I went from 2.66GHz up to 2.8GHz - The tests in Everest would show that my MEM and CPU is clearly faster, but any tests in games or game related benchmark app were slower than 2.66GHz :|

Shortly after I got some guidance which basically only involved reducing vCore and RAM voltages - don't think I touched anything else, and when I tried again it was faster overall.

Now 3.0GHz and still going faster but experiencing the same thing @ 3.2GHz.

Quote from pb32000 :

Also the C2Ds automatically reduce their multiplier (on mine to 6x) when idle. Don't know if you can turn that off.

Wouldn't it be fixed if set to - for instance 8 like I have?

Quote from r4ptor :

Yeah, it is - 510W

theres a world of difference between a Seasonic 510w and a QTEC unit, imo i think it's your dodgy PSU - go for a nice 600 OCZ eXtreme

r4ptor,
reading it up and down I dont get whats your real issue.
Could you just sum it in simple way please.

With your 1st post I understood you started to loosing performance when overclocked more then 3Ghz.I noticed from the LFS bechmarks that your CPU is throtling because of the way lowered min FPS in LFS.

I said also the temp is too high and I wouldnt be suprised its badly throtling.You expected to cores go up to 3.4Ghz???Be aware that some of the latest bunch of E6400 cores has trouble of going over to 3Ghz so I wouldnt be suprised you got one.Also you need after market cooling for CPU if you want to go so high.the bundled cooler is really not for something like that.

If the issue is not what I have mentioned above please let me know.I really have hard times to understand what is your real issue.

Quote :

Originally Posted by pb32000
Also the C2Ds automatically reduce their multiplier (on mine to 6x) when idle. Don't know if you can turn that off.

Quote from r4ptor :

Wouldn't it be fixed if set to - for instance 8 like I have?

Its called EIST in bios.When its enabled the cpu multiplier get lowered when the CPU performance is not needed too much.

I know that similar function on AMD A64 procesors might cause lowered FPS in LFS because sometimes the LFS cpu usuage get really low for a sec,the CPU downclock a lot and in a second the LFS cpu usage raise it takes little time the CPU clock to default but it cause a lot of trouble.

Quote from Jakg :

theres a world of difference between a Seasonic 510w and a QTEC unit, imo i think it's your dodgy PSU - go for a nice 600 OCZ eXtreme

I'd hate to recommend him a new $150 PSU and find out that it isn't the problem...

@ Devil, he has a 6400, which has been discontinued since the 6420's release. I would recommend he looks up the stepping he has to see if others have issues. He's got a good cpu cooler.