Q6600 Temp differences of 12c

[cian1500ww]

Hi guys,
Just finished building my new rig. I'm having a little bother with the cpu. Its being cooled by a D-Tek FuZion V2 on its own loop with a 360mm rad. The cpu idles at:

As you can see there is almost a 12c difference between cores 0,1 and cores 2,3. Its the same on full load:

The block is sitting properly, would more thermal paste help ??

Also, the mobo is an Asus Rampage Formula. The software that came with the mobo seems to overclock the CPU slightly as you can see in the full load picture, would it be the software that could be causing it ??

[RoadconeTuning]

thats odd, i thought mine as bad w/ core 0,1 being 8c hotter than core 2,3 on water w/ my GTZ its now 4c diff... those temps seem pretty warm considering once my room gets to about 80*F my cpu sits at 47*C at stock speeds and 1.27 vcore...

try reseating, when i did mine i put a bead of AS5 across both chip dies and let the waterblock spread it out evenly as i've always done... low 30's idle and mid-upper 40's under load once my room temp goes up about 8*F from the heat...

[Leeghoofd]

Try the Realtemp sensor test first... could be some borked sensors... weird that the idle temps on them 2 cores are far lower than the other two. Also the temps for just 2.5Ghz ops are pretty high for watercooling. You can do better with a normal heatsink...

[Spacehead]

Lap the IHS. It is warped and that's the reason for temp difference.

My post in the Intel Info And Support subforum:
http://www.xtremesystems.org/forums/...9&postcount=29

My cpu had temp difference between dies 1 and 2
Let's say cores 1-2 are on die 1 and 3-4 on die 2
So they are 10 degrees apart from each other.
If you decide to lap the ihs, do really careful work , for making IHS really flat. Temp difference will even out and you get your temps under better control.

[unclewebb]

Spacehead and cian1500ww: Can you post a screen shot of the RealTemp CPU Cool Down Test?

I believe that part of the problem with Quads is that Intel uses a slightly higher TJMax value for cores 2 and 3. This is not documented by Intel. Intel says that TJMax can vary but were completely fuzzy about any details. On a Q6600 this might account for 5C difference at full load. That's normal. When it's over 10C difference, I'd definitely be having a close look at the IHS for flatness.

Edit: Use Prime95 Small FFTs when testing. This is the best program for putting an even load on all 4 cores.

[cian1500ww]

I removed the D-Tek Fuzion V2 and found that one of the tabs on the quick mount wasn't catching right so that could have been part of the problem, I've re-seated it and put on a nice bit more AS5. Temps seem a lot better now, this is on 100% load for 30mins:


I actually lapped the proc a few days ago, I'm not sure if I did it right though, as the copper only really came through at the corners, but it was very shiny. I went as far as 2500 grit with it.

[cian1500ww]

Spacehead and cian1500ww: Can you post a screen shot of the RealTemp CPU Cool Down Test?

I believe that part of the problem with Quads is that Intel uses a slightly higher TJMax value for cores 2 and 3. This is not documented by Intel. Intel says that TJMax can vary but were completely fuzzy about any details. On a Q6600 this might account for 5C difference at full load. That's normal. When it's over 10C difference, I'd definitely be having a close look at the IHS for flatness.

Edit: Use Prime95 Small FFTs when testing. This is the best program for putting an even load on all 4 cores.

Heres the cool down test:

I took this after re-seating the block and applying more AS5.

[unclewebb]

My best guess is that your TJMax is somewhere around 104C on core 2 and core 3. The Q6600 G0 I had seemed to be about 100C for cores 0 and 1 and 105C for cores 2 and 3. You'd need to run Prime 95 Small FFTs for about 10 minutes at a higher core voltage or overclock before this 4C or 5C difference will become very obvious.

Hopefully Spacehead comes back here and posts that test so I can see what a square CPU is supposed to look like.

[Spacehead]

I have not checked how good the contact is this time. On my cpu the edges are also a bit rounded.

[Spacehead]

When I added some voltage and clocks I started to notice that 5 degree temp difference, but it is so damn hot here that I got reboot before it finished.

[cian1500ww]

I've done some overclocking today and part of the reason for my high temps at idle was down to the CPU voltage being left on AUTO. I was able to get it down as lows as 1.16V for stock. I've upped the clock on the CPU to 3.2Ghz with the voltage at 1.4V:

Anyone know why CPUZ says my voltage is lower than what I set it to in the bios ???

[Gator!]

thats wierd im getting the same temps on my Q6600 running at 3.4ghz with a tuniq tower...something isnt right...

[cian1500ww]

thats wierd im getting the same temps on my Q6600 running at 3.4ghz with a tuniq tower...something isnt right...

I think thats the best I'll get, I might be able to get the voltage down a bit but I'm quit happy with those temps. I'll keep oc'ing anyways and I'll report back soon. I don't think the temps are that bad. Whats your voltage at ?

[cian1500ww]

I've remounted the block using the original D-Tek Fuzion V2 mounting kit. I was using the quick mount which had the same clips as is on the Intel stock coolers. The temps have dropped a few degrees but now I'm getting a constant 5-6 degree difference between cores0,1 and cores 2,3. I've done all that I can do apart from changing the block itself. The only thing I can think of is that the out barb is on the top of the block where cores2,3 would be located. Cores0,1 are at the bottom of the block, near the logo afaik. I'm using the standard nozzle as well so I'd say if I changed it to the quad nozzle the temps might level out.

[unclewebb]

The best guess at why Intel seems to set TJMax higher for core 2 and core 3 is that this helps prevent all 4 cores from reaching the thermal throttling point at the exact same time. Core 0 and core 1 will throttle first when they near 100C and if this can't control the heat and get the CPU back in line then core 2 and core 3 will throttle next at 105C. With two independent Dual Core dies within a Quad, there's no problem with the first one running with a 6.0 multiplier to control heat while the second die continues to run with its full 9.0 multiplier. During thermal throttling, the multiplier is cycling hundreds or thousands of times a second to control heat. The hotter the CPU is, the more time it will use the 6.0 multiplier.

This allows a much more gradual transition so the average user is a lot less likely to notice a big drop in performance during some extreme task. It's just a theory that Intel has never commented on but it's the only explanation that I've been able to come up with that explains all of the data that I've seen. You can square your IHS up until it is beyond perfect and stand on your head but this 5C difference remains all the way up to 90C and beyond when all 4 cores are equally loaded with a program like Prime95 Small FFTs.

Spacehead: When testing on a hot day I just reduce my overclock with SetFSB. By dropping 300 MHz, you get a lot more temperature head room and less chance of a reboot.

Here's how my Quad looks when using a split TJMax.

Anyone know why CPUZ says my voltage is lower than what I set it to in the bios ???

It's normal. The actual voltage you get is less than what you set in the bios and it is usually less at full load than at idle. That's the way Intel designed it.

Edit: This same difference of 5C is evident on Core i7 as well. Here it is core 0 set to 100C and core 3 set to 105C. Have a look at enough screen shots that are running Prime95 Small FFTs and this exact 5C difference goes well beyond just random chance or everyone not knowing how to apply thermal paste correctly.

[Spacehead]

unclewebb that explanation seems completely plausible. I still have a little sensation that what if my contact between apogee and ihs isn't the absolutely best yet. This heat wave is killing me. I am starting to get headaches and confusion, not good.
I used only one little drop of Arctic Coolings MX-2 directly on the center of the IHS, and as the IHS is now flat, and Apogee certainly is flat, the TIM will spread evenly between under strong contact pressure, which is also evenly applied.

So what kind of correction should I apply to Realtemp? Will it work right if I put TJMax to 105 for cores 2 and 3?

Damn you tortured your cpu unclewebb What is the voltage? 1.8V?
My previous CPU X3210 couldn't take more than 1.41V on Prime95 load, it caused computer to reboot. If that could be fixed (sudden reboot) it could have gone over 3.8GHz easily.
Current CPU loves volts, but with current summer temps inside it can't operate effiently.

[NjH20]

having the same issue with my Q9550.

i re-seated and re-applied my block and past serveral times.

idle.
cores 1&4 = 40C
cores 2&3 = 30C

load
cores 1&4 = 60C
cores 2&3 = 58C

i'm just assuming its a bad temp sensor as when the cpu is at load, the temps on all 4 cores are only 0-2C diffrent.

[setter]

My q9550 has a bit of difference between cores as well.


Was the same on my previous q6600 that was lapped along with my TRUE black.


Havent lapped the q9550 but ive remounted it a few times and its still the same.

[unclewebb]

So what kind of correction should I apply to Realtemp? Will it work right if I put TJMax to 105 for cores 2 and 3?

I know for my Q6600, TJMax = 100, 100, 105, 105 made the most sense. The CPU Cool Down Test helps show this or running Prime95 Small FFTs for about 10 minutes until things stabilize also works to show the difference. Usually from 65C to 97C, the 5C difference is very clear.

Sudden reboots are a sign that you have reached the edge of the thermal envelope. You can run a Core based CPU very hot or very fast but not both at the same time. My E8400 is 100% stable for hours at 98C while bouncing off the thermal throttle at 3600 MHz. The same CPU can run Prime stable at 4000 MHz but to do that I have to keep the maximum core temperature down around 60C to 65C or I'll lose stability and either error out of Prime95 or get one of those random reboots.

Damn you tortured your cpu unclewebb What is the voltage? 1.8V?

Too much voltage can kill a CPU. I wouldn't want to do that!
Turning off the CPU fan is my way to create some heat while running Prime 95 Small FFTs of course. If you think the above test is nuts then check out a 3 hour torture session at 98C.



setter: The 5C difference I'm talking about is at full load while running Prime95 Small FFTs which is the best program to load the 4 cores equally. Programs like LinX create more heat but the load on each core is not equal so I avoid it when testing.

On 45nm Quads, instead of a 5C difference, this difference can approach as much as 10C between sets of cores within a Quad. Each CPU is unique though and should be checked.

NjH20: Try posting a screen shot of RealTemp while running Prime. Most programs don't bother ordering the temperature data coming from each core into their correct physical order based on APIC ID information. RealTemp does and makes these patterns easier to see.

[setter]

setter: The 5C difference I'm talking about is at full load while running Prime95 Small FFTs which is the best program to load the 4 cores equally. Programs like LinX create more heat but the load on each core is not equal so I avoid it when testing..

Yep i prefer to use p95 myself, ive used ibt in the past and noticed that loading wasnt level.

q9550.


q6600.

[unclewebb]

68, 68, 62, 62 at full load looks like a lot of Q6600 CPUs I've seen.

Your Q9550 will start to show a similar pattern when it gets about 15C hotter.
Give it some extra voltage to create some heat if you're curious.

[Spacehead]

Just showing some high voltage love:


Temp difference of solid 4 degrees. Maybe I can accept that. I wish to get my cpu to throttle , but then I have push lots of volts to it and could it die?

[PatRaceTin]

i think it normal 12C diff. between core

[unclewebb]

i think it normal 12C diff. between core

On a 65nm Quad, it's not normal at all.

Spacehead: Thanks for proving the point I was trying to make. A 4C or 5C difference in reported core temperature has nothing to do with an actual temperature difference. I'm beyond convinced that Intel sets TJMax higher on core 2 and core 3 to better control thermal throttling. The average difference I've seen is about 5C higher.

It's usually by about 65C or 70C and higher that these sensors will no longer show any slope error and will change 1:1 with changes in the core temperature. Below 65C these sensors tend to wander a little and will either move faster or slower than changes in the core temperature. Too bad Intel's presentation last year at IDF didn't include some more factual information about these sensors and how they set them at the factory so software developers could write more accurate temperature monitoring utilities.

Edit: Small FFTs will give more accurate results but no need to torture your CPU any more.

[setter]

68, 68, 62, 62 at full load looks like a lot of Q6600 CPUs I've seen.

Your Q9550 will start to show a similar pattern when it gets about 15C hotter.
Give it some extra voltage to create some heat if you're curious.

Definitely gonna give it some more voltage, if i can get it prime blend stable im gonna push the clockspeed a bit higher, 8 hrs small fft is fine but im failing blend after 3 hours.

vcore 1.21250 in bios, (1.192 idle, 1.208 load) LLC enabled
pll, 1.54
nb, 1.14
vtt, 1.14
v dimm, 2.02