Are large core temperature differeances possible?

[Lemon law]

Ok, here is the story and its somewhat a warranty issue. My wife bought a computer with a intel Pentium D 915 dual core chip with a pressler 65 nM core. All reported temperatures are well within the nothing to worry about range with the lone exception of core 0 temps, which are basically way way too high. At zero cpu load, core zero runs 50-55C, put any load on the cpu, core zero temps routinely go past the 63.4 C red line, and under sustained load it can go as high as 70C. With the stock intel heat sink and fan getting annoying loud as it goes from a quiet 2500 RPM to a sewing machine sound as it throttles to 4000 RPM.

At the same time, my reported core 1 temperatures stay rock steady at about 33C. Maybe understandable if core 1 is doing nothing, but I have a number of applications that will use both cores of the processor. And since some of my temperature monitoring programs will report each core temperature and each core load load in real time, I can watch them run. And I get the same basic thing, even though both cores are under similar heavy load, core zero reported temperatures rise quickly under load
and core 1 temperatures rise hardly at all. At idle there is a 17 C difference, under equal heavy load it can soar to a 35 C difference.

Ya, I know Silicon is a basic insulator that conducts heat slowly, I do not know how far apart each temperature sensor is, but still its basically two separate chips stuck together and all the common heat generated must flow into the same heat sink. But a 35 C temperature difference seems hard to believe unless the core 0 sensor lies. It seems to me that if the heat sink and fan cannot handle the heat generated fast enough, sooner or later the heat from core 0 should reach core 1 even if core 1 is idle.

The dealer has been willing to remount the heat sink and fan which some what eliminates a bad initial install or something that vibrated loose during shipping to no effect.

And now we are looking at a different and better aftermarket heat sink to get core zero temps down.

And now I get to the mental masturbation question I cannot answer.

But all stock and after market heat sinks and fans share the same contact base with the processor, heat generated in the processor flows into the heat sink base,
and better aftermarket coolers simply have what amounts to a bigger radiator in a car.

And to use the same automotive analogy, a car will overheat if the radiator gets clogged. At the same time, an infinitely large car radiator will do no good, if the cause is a thermostat stuck in the closed position.

So I have to also look at the possibility that somehow I have a warped chip top
which would be the generic equivalent of a stuck thermostat. And somehow my hear sink base makes a good contact with the parts of the processor involved with core 1 and a very poor contact with the parts of the processor involving core 0. Can anyone with better knowledge of actual intel chip architecture comment on that hypothesis? Because if that is true, a better aftermarket cooler will not help.

And I, as a warranty issue also have to work with the seller, because I am holding his feet to the fire on the heat issue. He may not go with a lap the processor solution, and it voids the processor warranty. I also wonder if I got a defective intel chip, and an in warranty exchange from intel would be best.

Any comments welcome.

And ps, no overclocking is involved, vcore volts are as low as possible and all bios setting are stock.

 

[RallyMaster]

http://forums.anandtech.com/me...=2220755&enterthread=y Boom.

 

[Lemon law]

While I thank you for the post and the link, RallyMaster, after pursuing the link and the side links, it tells me exactly nothing about how to resolve any of my questions. And for added information, for some reason, real temps refuses to install on my wife's computer. But the rest of my temperature monitoring programs all seem to be in fairly close agreement.

Meanwhile I will try some more testing with prime95.

 

[Lemon law]

After doing some primes 95 checks, getting the data from the chip sensors, and then writing intel, receiving a series of a inane intels replies showing that no one there knew how to read, and keeping up with the process of holding intels feet to the fire of truth, I finally have a flat intel statement.

Any differences of core temperatures greater than 15 C constitutes in an intel chip constitutes a defective intel chip and is an intel warranty issue.

 

[RallyMaster]

Or it could just mean that the sensors are wrong like they always are and you're being paranoid.

 

[Lemon law]

Ok RallyMaster,

Put your self in my shoes.

Should I surrender to paranoia denial, put up with an stock heat sink and fan that screams in my ears half the time, and go lalala its not happening?

Or would you hold the seller's feet to the fire and asked it be fixed under warranty because reliable heat sensors are part of entire intel chip package?

And does your computer regularly register over 70C under routine use with zero overclocking?

 

[RallyMaster]

Um...you could just replace the heatsink. It's really not that big of a deal. Pentium Ds do run hot. IIRC they're just two Prescott cores glued together.

 

[Lemon law]

Please RallyMaster, I don't want to get into a snit with you, but are you really considering the problem? I have been there done that considered the better heat sink and fan option. But when the same hint sink is involved in cooling core 0, which shows a giant reported temperature problem, and at the same time cools core 1 that shows nothing to be alarmed about, and thus its kinda of dubious hypothesis.

But that is all I have, nothing but illogical hypothesis, and no damn method to tests anything on anything but a sheer gamble basis.

 

[BonzaiDuck]

The Pentium D was a toaster to begin with. It is possible that it has a defective sensor. For what those of us here use TJunction sensor readings, it seems that a lot of Penryn duo Wolfdale and quad Yorkfield cores have "stuck" sensors, while Intel says that "the sensors weren't meant to read idle temperature values." It sounds like a warranty disclaimer.

Here's a thought -- to see if it isn't just a bad sensor.

Some monitoring software available -- such as Everest Ultimate -- reads the two core temperatures and the TCASE sensor as well. TCASE should be lower than both the cores by some number of Celsius degrees charactistic of that particular processor model. The TCASE sensor is between the cores, and is supposed to measure the temperature at the heat-spreader.

So you load up the processor and discover as much as a 35C difference between the cores. Find out if the TCASE temperature is higher than the low core, because -- if it is -- then the problem is a bad sensor in that core. Both cores under load should read higher than TCASE.

The Smithfield and Presler cores were -- as I said -- 135W toasters. Later revisions brought down the thermal wattage and TDP -- closer to a 100W.

If it were me, I'd just replace it with a Conroe C2D if the motherboard is compatible with the Conroe. If not, and if everything else (fully loaded threads in Task Manager when running Prime95, for example) is fine but for the temperature reading, I wouldn't worry about it. Intel wasn't all that eager to replace it for you, were they?

The tipoff in all this -- if Task Manager shows that a stress-testing program is loading both cores 100% -- is a core that just remains at 33C.

Now -- I don't know what the thermal specification for that processor is. The thermal spec is stated in terms of TCASE, and the cores should be 10 to 20C higher than TCASE -- I'm guessing there, too, because of the thermal reputations of the Smithfield and Presler chips. If one core sensor shows 70C, that may still be normal for a Presler.

I've come to a conclusion that faulty temperature sensors aren't enough reason alone to get rid of a processor for practical usage. They are just a nuisance. For over-clocking, I'd at least hope that the TCASE temperature is adequately reported, and that the spread between idle and load TCASE would correspond to the thermal efficiency of the cooler -- that is -- if you're familiar with the cooler's behavior even in benchtest reviews, TCASE spread temperatures would be in line with idle-to-load spreads, unless something were really wrong.

Presler cores ran pretty damn hot, though. That's why I skipped both Smithfield and Presler as I started planning my computer-building projects these last couple years.

If you want to cut the noise, dump the stock heatsink and fan, and get a ThermalRight or other high-efficiency model that fits a 120mm fan.

EDIT:: BUT HEY! IF Intel says it's a warranty issue, then RMA the damn thing.

 

[RallyMaster]

Originally posted by: Lemon law
Please RallyMaster, I don't want to get into a snit with you, but are you really considering the problem? I have been there done that considered the better heat sink and fan option. But when the same hint sink is involved in cooling core 0, which shows a giant reported temperature problem, and at the same time cools core 1 that shows nothing to be alarmed about, and thus its kinda of dubious hypothesis.

But that is all I have, nothing but illogical hypothesis, and no damn method to tests anything on anything but a sheer gamble basis.

Honestly, there's nothing there that you should really get your panties in a bunch about. It's a FUBARed thermal sensor. If the heatsink doesn't feel hot to the touch and the chip isn't melting or blowing up your motherboard, there's isn't ANYTHING to worry about. Stop being paranoid. If you really feel that you've been jipped out of the peace of mind, RMA the thing. Otherwise, I'd just let it be.

 

[BonzaiDuck]

Originally posted by: RallyMaster

Originally posted by: Lemon law
Please RallyMaster, I don't want to get into a snit with you, but are you really considering the problem? I have been there done that considered the better heat sink and fan option. But when the same hint sink is involved in cooling core 0, which shows a giant reported temperature problem, and at the same time cools core 1 that shows nothing to be alarmed about, and thus its kinda of dubious hypothesis.

But that is all I have, nothing but illogical hypothesis, and no damn method to tests anything on anything but a sheer gamble basis.

Honestly, there's nothing there that you should really get your panties in a bunch about. It's a FUBARed thermal sensor. If the heatsink doesn't feel hot to the touch and the chip isn't melting or blowing up your motherboard, there's isn't ANYTHING to worry about. Stop being paranoid. If you really feel that you've been jipped out of the peace of mind, RMA the thing. Otherwise, I'd just let it be.

That may be true -- probably IS true. But even on those toasters, there wasn't a spate of reports about "bad sensors" on five different forum-web-sites.

This "high-incidence" of reports suggesting that Wolfdales and Yorkfields have a frequent problem on the low temperature-end, with Intel saying "the sensors weren't meant to measure idle temperatures," has me ticked off. I LOVE my E8600; but with software that only controls fans and other devices through the core temperature sensors and no way to do it through the TCASE sensor, it's a catch-22 for what otherwise be a totally thermally-controlled system.

So -- yeah -- I have a little "wedgie" of my own . . . .