Q9550 Yorkfield Temp Issues on E0 Stepping

[Fike]

I got an E0 stepping Yorkfield Q9550 from New Egg last week. Everything is assembled and running. Vista 64 installed.

Temps are high. Idle is upper 30s while the BIOS lowers the multiplier to 6X instead of the 8.5X that is run under load. Under a load of 4X Prime95, it shows temps getting into mid 70s in under 10 minutes with an open case. I don't consider this acceptable to proceed with my install and build of everything so I need to get temps down. The Intel data sheet says 71.4 degrees C is the max operating temp at TJunction.

I reinstalled the heatsink with new Arctic Silver 2 thermal paste. Temps remain the same. I can't imagine that a new generation of arctic silver (what is it now, version 5?) would make anything more than a minor incremental difference. I put a dollop a little smaller than the size of a BB and then before installing heatsink, I spread it to a thin layer with a razor blade. Between applications I cleaned the heatsink and processor with alcohol.

OPTIONS:

• Return CPU and heatsink to NewEgg for replacement because it runs out of spec.
  Use more thermal paste.
  Get better thermal paste and reinstall heatsink.
  Locally, buy a decent heatsink and fan to see if that makes a difference. which quiet one should I get?

I have heard of some issues with stock intel coolers having heatsink surfaces that are not completely flat--so they don't have good contact with chip. I guess that is a possibility.

It seems unlikely with a processor that has been out so long as the Q9550 that they would be having such poor temp performance issues--particularly considering the E0 stepping is supposed to be the more overclockable stepping. These kinds of issues should not be making it out of Intel QA, thus my presumption that there is something wrong with the heatsink.

Suggestions anyone????

 

[Kenmitch]

The stock cooler kinda sucks....Aftermarket would be better.

I don't think your temp is unreasonable for stock cooler....Whats temp with case closed?

One option would be to try and lower vcore in bios....As long as you don't overclock it alot.

My Q9550 vid is 1.150 and I'm in the process of testing it out at 1.100 so far so good. I'm
not gonna OC it anyways as it's overkill for a HTPC anyways.

I'm gonna play around with it today and see how low I can get it and compare the temps using
Intel Burn Test as it heats up alot faster than prime does.

Thanks,
Ken

 

[Fike]

hadn't thought of undervolting. I may see if it runs stable at lower vcore. I haven't closed-up the case to test performance yet, so I don't know what case temps will be yet.

I too am not overly concerned with overclocking this processor. Limiting factors on a system like this are really in the IO--hard drives or network--so the processors will rarely top-out for very long, particularly all four at once. I am mostly concerned with reliability of the system, so temps are an important factor for the longevity and reliability of hard drives, processor and memory. I will probably buy a new heatsink...maybe arctic cooling freezer 7 pro.

 

[Kenmitch]

Try stuffing a Q9550 in a Silverstone SGO2-F not too many cooler options available

I'll update this post for you in a couple of hours with the results.

By the way I'm am running with the stock cooler at the moment also

Thanks,
Ken

 

[Fike]

what temps are you seeing?

I know that 70 degrees C isn't outrageous for an overclocked system, but intel's data sheet says 71.4 degrees C is max operating TJunction temp, so I figured I should achieve that at stock speeds before I consider myself stabilized.

 

[Pneumothorax]

Quite frankly the stock cooler on the core2quads is barely adequate. You basically need a well-ventilated case with good airflow to get decent temps at stock speeds. Intel is so cheap they throw in the same heatsink/fan on the q9550 as the E5200 for pete's sake!

 

[Idontcare]

Originally posted by: Fike
intel's data sheet says 71.4 degrees C is max operating TJunction temp

Where are you getting this info?

 

[Fike]

Originally posted by: Idontcare

Originally posted by: Fike
intel's data sheet says 71.4 degrees C is max operating TJunction temp

Where are you getting this info?

From the Intel Datasheet.

On page 76 it identifies the 95 watt Q9550 and refers you to table 5-4 and Figure 5-3. When you get to table 5-4, it refers to 71.4 degrees C for TCase at 95 watts. I was mistaken; it was not Tjunction it was Tcase.

 

[Idontcare]

I was wondering if that was the source of the confusion here, its common to make this mistake between thermal specs, tcase, tjunction, tjmax, etc.

TCase is for system builders and is NOT a CPU temperature maximum, it is the max temp that the center portion of the outer-surface of the IHS should reach when the chip is operating at its TDP limit.

Tcase is an engineering constraint that system builders need to manage so their thermal design can properly handle the cpu's TDP while maintaining enough engineering margin to avoid hitting TJmax under most usage scenarios.

See slide 14 of this official Intel document regarding DTS and TJmax.

For Q9550 the TJmax is 100C.

Unless you are doing what you need to in order to properly measure Tcase for your system then you should ignore the thermal spec and go with the TJmax reading given by the DTS.

 

[Fike]

Yikes!! 100C. That is hot. I guess I don't really have a problem. With that said, I have had one crash while burning-in somewhere in the mid70s. I may try to cool it down a bit, but I guess I shouldn't worry so much.

thanks

 

[Idontcare]

Yeah its not to say you don't have a temp issue or that you would want to run your cpu near its 100C tjmax, just saying first things first is to remove any confusion from the topic that you might have had in regards to your cpu's max temp.

Now regarding your cpu's temps...are you absolutely positively convinced the HSF's pushpins are fully and properly latched on all four corners?

Your temps seem high to me for a stock 45nm E0 yorkfield and we've seen countless threads here where the culprit turns out to be a pushpin that managed to not fully push thru the mobo hole to properly latch (not the same thing as the pushpin being fully pushed down, a fully pushed down pushpin from the top view does not guarantee the pushpin itself has pushed thru the mobo and latched from the underside).

I experienced this myself and the only way I could convince myself I had properly latched the pushpins was to install the HSF with the mobo outside the case so I could visually inspect the mobo underside and confirm all four pushpins had indeed finally latched. Then and only then were my temps were finally inline with everyone elses.

 

[Kenmitch]

Originally posted by: Fike
what temps are you seeing?

I know that 70 degrees C isn't outrageous for an overclocked system, but intel's data sheet says 71.4 degrees C is max operating TJunction temp, so I figured I should achieve that at stock speeds before I consider myself stabilized.

Silverstone SGO2-F with Silverston cross-flow fan and Silverstone fan controller the one with the 2 1" grills in it....Only reason I bought it was for airflow

The test was conducted as follows. Booted into windows with a house fan aiming at front of case until all HD activity was done. This was done mainly because the stock cooler doesn't do a good job as far as dropping the temp down from loaded to unloaded....Proves it's inferiority! Also I wanted to save some time and find the lowest vcore I could run at sooner than later. Turned fan off and waited for idle temps to stabilize for a minute or 2. Noted idle temps(Real Temp) Fired up Intel Burn Test and used the max temp shown for each core for my data. Vcore loaded readings are from CPU-Z.

Q9550 vid in real temp is 1.100 min and 1.150 max (max confirmed in bios) Whats your vid?

The results with room temp at about 77*F

Vcore set on auto = 1.170 (WTF) in bios 1.088 loaded temps at idle 43,36,39,42 loaded 73,65,68,69

Vcore set to 1.100 = 1.122 in bios 1.040 loaded temps at idle 41,35,39,40 loaded 70,62,66,67

Vcore set to 1.075 = 1.090 in bios 1.024 loaded temps at idle 41,35,39,40 loaded 68,60,64,65

Vcore set to 1.068 = 1.074 in bios 1.008 loaded temps at idle 40,35,39,39 loaded 67,60,63,64

Looks like my MB tries to compensate for vdrop ???

Anything lower will not pass Intel Burn Test.

Currently testing the system with prime with the lowest vcore that would pass Intel Burn Test. I'm at just over 3hrs. into prime torture test(blend) temps are comparable to what Intel Burn Test was.

This data is a your mileage may very tho....I'd imagine that your case would have alot better ventilation than mine....Guess maybe mine is better for the test tho

Thanks Microcenter for the great deal on the CPU....Guess maybe she gets down into the S class for free as a bonus also!

Thanks,
Ken

 

[Lemon law]

If it were me, I would be doing three things.

1. Using something like real temps, I would be looking at all 4 core temps under sustained equal load, at no point should the lowest temp differ radically from the highest. I can't exactly remember off the top of my head the point at which Intel calls the processor defective.

2. I would take load temperatures with the side panel off and also on. If the side panel on temps are significantly higher, you are partly barking up the wrong tree when you clearly have case ventilation problems that also need addressed.

3. You hit an issue on processor flatness, but its useless to speculate when its fairly easy to test. Find a short known flat straight edge, if nothing else a small square of windows glass cut the same size as the processor top, place it on top of the processor, if you can make it rock side to side, you know the processor top is convex, and if you feel it rocking, its really bad. If instead the processor top is concave, it gets a little harder, but that is where something
like a small tube of oil paint comes in, and if rubbing on a known flat rubs the oil paint off the edges and not the center, its a good clue. And then instead of lapping the top edges off of a processor, thus voiding any warranty, you can instead butcher the mating surface of your cheap heat sink until both surfaces perfectly match with the test being equal paint rub off. In machining parlance its called hand scraping, not an easy process on two large mating surfaces of fairly hard steel, but far easier on the soft copper of a small heat sink. But as someone who has worked as a machinist, I would simply track the surface with a dial indicator and would almost instantly know how far off flat both mating surfaces are. And the fact that your idles temps are low
and then the heat sink fails to wick away heat does suggest you have a heat transfer problem where the processor meets the heat sink.

But you also have a 95 watt processor and that is a small furnace on a 45nm fab. The other thing you might do is see what temps you have in routine gaming use, but anything above 60C would scare me. With 70 C being unacceptable.

And of course the Kenmitch solution of lower vcore also makes sense no matter what your temps are.

 

[BTRY B 529th FA BN]

100*C???

I thought 74*c was max

 

[Idontcare]

Originally posted by: BTRY B 529th FA BN
100*C???

I thought 74*c was max

max for what? max temp allowed for the cpu is TJmax.

 

[BTRY B 529th FA BN]

Originally posted by: Idontcare

Originally posted by: BTRY B 529th FA BN
100*C???

I thought 74*c was max

max for what? max temp allowed for the cpu is TJmax.

So, if my core temps are reading 74*c i still have 26 more degrees until a self shut-off?

 

[Idontcare]

Yep.

Make sure your coretemps are read with the proper TJmax offset though, some programs assume tjmax is 95C and other assume 105C. I use coretemp and adjust the offset until the Tjmax it uses for temps is correct per Intel.

 

[BTRY B 529th FA BN]

WOW, and I was scared at temps of 78 - 80*c - Gratzi

 

[Idontcare]

Originally posted by: BTRY B 529th FA BN
WOW, and I was scared at temps of 78 - 80*c - Gratzi

That's not to say you should feel OK about running your cpu that hot. Any temp increase reduces its operating lifespan. And if you are overclocking, or undervolting, then temps matter in terms of the voltage necessary for stable operation at that temp.

I won't run my cpu >75C, I've no desire to push the Vcore high enough to run stable at that temp and with that power consumption.

 

[Fike]

Well, I RMAed the chip. It was consistently into the mid 70s and then it crashed once. After that the BIOS went into boot thrashing (don't know what to call it when it reboots repeatedly trying to find an acceptable speed) and finally settled at a multiplier of 6X. That is 2GHz. That sucks. The new one should arrive today. I will be starting with Arctic Cooler 7 instead of screwing with the factory heat sink. I also took the mboard out so that I can be absolutely sure it is installed correctly the first time.

My target is to never exceed 70C under load. As long as it gets to its stock 2.8GHz speeds, any extra speed I can achieve with the arctic cooler 7 while staying under 70C will be okay.

 

[Kenmitch]

Gigabyte motherboard? Just wondering as it is a nice feature when one has bad
bios setting(s)

My e5200 is in a gigabyte board and this feature came in handy a couple of times when
I was trying to see what she would do.

 

[Fike]

yeah, it is a Gigabyte EP45-UD3R. I like the board. It seems good. It was doing what is was supposed to be doing. It was the chip that was failing.

 

[BlahBlahYouToo]

yeah, something is definitely wrong.
i have the same CPU from MC, E0 stepping running at 3.4ghz @ 1.2v with an OCZ vendetta2.
idle temps are high 30s, low 40s. load temps hit low 60's.
now obviously i'm running a better cooler, but i'm also oc'd whereas you're not.

as for the AC freezer pro, it was a good budget cooler back in the day but has been surpassed by the newer heatpipe HSF's.

 

[Kenmitch]

Just wondering how CPU #2 is working out? Any diff?

 

[Fike]

She is working perfectly. the new processor with the arctic cooler 7 idles around 38-40C and with the Intel burn in tool, it just gets into the low 60s.

I had some memory stability problems with the cheap OCZ 1066 memory I bought. Fixed timings and throttled back to 800 MHz and it seems to be fine now. I am still running burn-in tests.

Weird thing. If I run Prime 95 standalone overnight with the Blend Test (lots of ram tested), it will run at 95% overnight and not crash. But, if I use Realtemp 3.0 and have it launch the Prime95 Blend Test, it fails in about ten minutes. I have run 5 passes with the IntelBurn Test and it was fine. I'm gonna run some more tests today without RealTemp, and if it remains stable through to tomorrow, I will start migrating my data and apps over.