Real Temp - New temp program for Intel Core processors

[Xello]

Hi guys,

I read a post somewhere, i think it was on this thread, about varying temps sometimes being the result of uneven contact on the cpu. Hypothetically speaking, if someone suspected that they had this situation, where perhaps one corner wasn't as tight as the others, which corner correlates to which core on the die?

For example, if my core 3 had 10c higher temps than the other cores, could i attempt to fix that by tightening the top right corner of my waterblock? Or maybe bottom left... you get the idea

Anyone know?

[bowman]

[fornowagain]

[unclewebb]

No wonder it's hard as hell to keep a Core i7 cool when cranking up the MHz and core voltage. There's no surface area to dissipate the heat.
Maybe Intel should come up with a perpendicular socket so a heatsink could be applied to both sides of that little CPU to keep it cool.
The upcoming 32nm version should put out less heat but with the reduced surface area, you might not be any further ahead.

Xello: I played around with heatsink tension once and the results to core temperature were pretty minimal when air cooled until the heatsink was about ready to fall off.

The slope error from one core to the next is usually not that much but TJMax seems to vary from one core to the next and this is especially noticeable on 45nm Quad core processors. Intel hasn't publicly documented what the typical TJMax variance is between cores on the same Quad. Most Core i7 screen shots I've seen seem to vary by about 5C with core0 reporting the hottest temperature and core3 the coolest. 65nm Quads can vary by 5C with core2 usually reporting the coolest temperature and I think 45nm Quads can vary by as much as 10C from one core to the next on the same Quad. Run a CPU Cool Down Test and post your results.

steelsky: If you read all 120+ pages then you deserve a gold star.

rge and I usually do our testing with the case open. It sounds like you have good airflow but it's easier to make direct comparisons by taking the case and airflow out of the equation. Your idle temps at low MHz and low voltage might be a couple of degrees too high but nothing too out to lunch. Your method of applying heatsink paste may not be the AS5 way but it's probably not that bad either. My Q6600 is very similar. Core0 and core1 are typically equal, core3 is the same or 1C less and core 2 droops down about 5C less then the rest of them. This is more sensor error than user error.

How much core voltage does your Q6600 need to run Prime stable at 3000 MHz? My Q6600 has a Max VID of 1.3250 volts and not surprisingly, it's a crappy chip. I also have an early P965 motherboard that loves Dual Core chips but isn't that happy with the Quads. I'm a quiet freak so 3 GHz is pretty much my practical 24/7 limit with quiet air cooling for this CPU. There are better Quads and better motherboards that can probably run 3 GHz Prime stable with default voltage but I need about 1.39 volts which equals more heat. Post a Cool Down Test and hopefully by the end of the weekend I'll have a chance to get caught up and have a look at some data. Post a screen shot of CPU-Z as well so I can see your core voltage.

[speckled10]

Uncle - first of all congrats again on a brilliant program.

I have 2 questions:

1. Have you thought of adding CPU Case temperature reading to your program ?

If you read my post here

http://www.xtremesystems.org/forums/...15#post3548415

I am having huge problems trying to get an accurate Tcase reading from various software - as a way of testing the 'Tcase + 5C = Tjunction' theory.

2. Is there an accurate estimate yet of i7 Tjmax ?

Realtemp defaults to 100C, which I have heard is the i7 figure - but If I believe PC Probes Tcase readings on my system, and the formula above - it would indicate Tjmax is actually 105C on my CPU.

That is to say - under prime load I find PC Probe currently reading Tcase about the same as Realtemp core temps w/ Tjmax = 100C.

Alternatively, if I believe Everest Tcase reading - it is 5C+ below core temps w/ Tjmax = 100C

You can see my dilemma

[SFaznSpEEdSTeR]

Could anyone tell me why if I have a weird chip?

It's a E0 Q9550 OC'ed to 3.825GHz on 1.3125V on a 780i FTW and my temps don't look right

Basically at idle temps, it's at 31/31/39/37 while at full load on Prime95 the highest they go sometimes during high ambient temperature is like 49/49/47/48



Sorry for the uncropping of the image

I took the screenshot before the test was fully done because when I took it, I thought it was already done

Besides that, the last two throttling TJMax distance has all the same numbers as the last throttling you see there so it wouldn't really matter anyways even if I run the test again. Unless I put my chip on a pretty good load, the last two core barely moves. For instance, they would be at 31/31/39/37 at idle and right when blend Prime95 starts, it only gets to 38/38/39/37.

My default TJMax is 100C because I have a craptastic 1.25V CPU VID and I have yet to adjust anytime.

One problem that I could think of is that right now on my S1283, there are dents on bottom of the HSF on my heatpipes showing the 4 corners of my CPU so it could probably be the lack of contact between my CPU and HSF and I don't think there is enough MX2 to fill that hole but I'm not sure

[Xello]

Ah ok, so the cores are all pretty much right next to each other in a line down the center of the IHS, i guess it's not that big of an issue then. Sometimes you forget how small the little blighters are

[kemo]

kemo: It's possible that both your sensors are stuck at a Distance to TJMax of 55. Core 1 definitely is and core 0 is a maybe. You could try another Cool Down Test at a lower core voltage and MHz to try to get your CPU to run cooler to test this. In Cairo you probably don't have as many options like I have with my backyard sitting at -30C today. Core 0 looks usable.

too bad i dont have -30C but all i can do is this
2.4GHZ Vcore <1.1V Fan on Full , After Test 6 both cores will stuck on 55 and Core 1 is stuck from the beginning till the end , that CPU used to run for 3 months @ 80+C so i am not surprised if both sensors are damaged

[unclewebb]

kemo: That test proves it. Both sensors are stuck at a Distance to TJMax of 55 just like you thought.

I don't think that running your CPU at a hot temperature for months damaged anything. These sensors weren't designed to report accurate temperatures from idle to TJMax and it's likely that your sensors perform exactly the same as the day you bought your CPU. At least they're sort of accurate when the Distance to TJMax is 55 or less as long as you correct for the difference in TJMax for each core.

speckled10: Accurate TCase temperatures don't exist. That's the reason why Intel gave up on them and doesn't use TCase temperature information to control their CPUs anymore. Thermal throttling and thermal shut down is all based on the data coming from the core sensors. If your core sensors aren't sticking like kemo's and you check the calibration of the core sensors, most of them are quite usable. Don't make comparisons between TCase and core temperatures. Out of the box and as is, you're comparing two inaccurate sensors which is meaningless.

2) Starting with Core i7, Intel stores TJ Target information within each core of the CPU that RealTemp, Core Temp and Everest are able to read. The problem is that this number is still only the TJ Target and the actual TJMax for each core may not be exactly the same as the TJ Target. There is still slope error where these sensors move at a different rate than the temperature changes at as well as TJMax not being an exact number. Core i7 is a big improvement over the 45nm Core 2 sensors but there's no such thing as perfect sensors. All sensors need to have the calibration checked.

[speckled10]

2) Starting with Core i7, Intel stores TJ Target information within each core of the CPU that RealTemp, Core Temp and Everest are able to read.

OK - I see from your post:

http://www.xtremesystems.org/forums/...postcount=1136

that Realtemp is reading Tj target - but is it being displayed in the program anywhere ? Can you tell us what it is ?

Or is it the Tjmax value that we see under settings (100C for all cores on my copy of Realtemp) ?

[unclewebb]

Or is it the Tjmax value that we see under settings (100C for all cores on my copy of Realtemp) ?

That's right. RealTemp reads the value from Core i7 processors and uses that value as TJMax.

Intel did the play on words with TJ Target not necessarily being the same as TJMax when they released this information for the previous 65nm processors.

By having this information stored within each Core i7 processor it would be possible for them to have a different TJ Max value for each core and software could read that and adjust accordingly. So far, all I've seen used is 100 for this value for all Core i7 processors but maybe in the future Intel will use this capability.

As far as I know this value is only a TJ Target number and actual TJMax may be slightly higher.

[ptelles]

Apparently my front intake fan has decided to fail which explains why Everest didn't list it in the screen cap. After manually encouraging it to crank over, it doesn't appear to have any impact on the closed case temps.

ptelles: It doesn't look like CPU-Z is reading the correct core voltage from your motherboard. Everest showing CPU1 at 1.21 volts is likely correct. Your testing shows that TJMax is not likely 100C. It also looks like Intel's TJ Target of 85C for your CPU would be too low.

Based on your testing, my best guess is that TJMax is likely closest to 90C. Intel says that TJMax is not an exact number but I think using 90C will get you accurate temperatures from idle to TJMax. That would put your reported temperatures above your room temperature at a similar amount to what rge found during his testing. rge tested a Dual Core, but based on power consumption at idle, a Quad would only be about 1C higher during this test. Your additional core voltage might also contribute another degree compared to rge's testing.

9C above your room temperature of 25C equals 34C. If you use TJMax = 90C, your reported idle temperatures are going to be right in the 34C range so I'd go with that.

Thanks Uncle. CPU-Z 1.49 apparently mistakes the empty second processor socket's vcore for that of the first processor.

Per your suggestion, I changed Tj Max to 90°C. I also determined that in fact my intake fan did not fail as I had originally thought. With the exception of the CPU cooler, all other fans were under BIOS control. Apparently the BIOS would randomly decide not to start the NB & intake fans at power-on.

I decided to disable BIOS fan control, add 3 exhaust fans to the left side of the case and increase my overclock from 2.83GHz to 2.9GHz.

I ran an eight hour Prime95 blend torture test. Real Temp reported a high of 54°C and a low of 37°C. Ambient room temperature for the duration of the torture test varied between 25.33°C and 27.11°C. Not bad eh? Your thoughts?

[Brama]

Hello, I don't know if this was already reported but with last beta in my system (Xeon E3110), when the temperature goes over Tjmax, the "Tjmax distance" values are numbers starting from 128.

I explain better: I have Tjmax at 100 C°; when I reach 100 ° the "Tjmax distance" correctly displays 0; when the temperature goes over Tjmax, i.e. 103°, "Tjmax distance" displays strange numbers, from 128 and upper.

I suggest to display "over" or something like this, when the temperature go over Tjmax and not these strange numbers.

[burebista]

Readme. Nothing is strange in RealTemp.

[randomizer]

Yea, I thought it was funny too, until I discovered it was meant to be like that

[Naja002]

Just wondering if there is any chance for a feature/program that will monitor/display temps of more than one PC over a network?

[KURTZ]

guys, and what about the mobos with only ONE GTL setting like the P5K? they can set only auto, 0.63, 0.61, 0.59 etc.

[unclewebb]

I suggest to display "over" or something like this, when the temperature go over Tjmax and not these strange numbers.

The Distance to TJMax box always displays the raw data coming from the on chip digital thermal sensors. Intel's point of view is that once you go beyond TJMax that this data is meaningless but if you change the interpretation of it, I think it's still very meaningful. One user sent me some screen shots of his Q6600 at 122C trying to determine the thermal shut down temperature. Intel only documents the mobile chips which have a shut down temperature of 125C. The temperature data coming from these sensors when you are beyond TJMax is probably more accurate than at idle.

Once you go beyond TJMax, HOT should be displayed in the Thermal Status area so that is your warning that the Distance to TJMax data might look a little screwy.

ptelles: You still have lots of temperature head room so don't be afraid to push that chip a little harder.
I think XS bans users if they find out that your CPU isn't running 24/7 on the ragged edge of self destruction.

Naja002: Just wondering if there is any chance for a feature/program that will monitor/display temps of more than one PC over a network?

That sounds like a useful app but I have no idea where to start.

[Draxx]

unclewebb, any chance you looked over my numbers yet?

[unclewebb]

unclewebb, any chance you looked over my numbers yet?

I was hoping all you guys on the weekend went away so I'd have some time to work on RealTemp today.

Here's a calibration you can try based on the data you've provided.

TJMax = 100, 100, 108, 108
Idle Calibration = 0.0, +3.0, 0.0, -3.0

If you don't have a thermal paste issue, then these numbers will get your 4 cores to line up a lot better and your idle temperatures should be a similar amount above room temperature as what rge found during his testing.

Run Prime95 Small FFTs for 30 seconds or so on and then 30 seconds off for a handful of times and check out how the 4 cores track each other. They should be similar with these settings and probably more accurate in the temperature range you run at than what these sensors provide out of the box. Perfect temperatures was my original goal but I've learned to settle for reasonably accurate.

[Draxx]

Thanks mate

[steelsky]

Cheers Unc, Ill wear the gold star with pride hehe - excuse the delay i had a software problem and the comp wouldnt boot... some time with the original vista disc seems to have cured it eeek.... but anyway haha.

Opened up the case and ran the low mhz test again...Vcore down to 1.100 hope this imageshack stuff works lol..


By steelsky at 2009-01-05

a cool down test....


By steelsky at 2009-01-05

and finally a snap of in the box hehe


By steelsky, shot with Samsung S85 at 2009-01-05

Just to clarify some of my reading... VID means the advised voltage from intel for that chip to run stable at stock speeds?

My chip also has VID of 1.3250 ... DOH... it took about 1.343 or the next notch up to get it stable at 3ghz...

Ive noticed i can get a bit more stability by raising the MCH voltage - What does the MCH voltage actually add voltage too? ie will it be the CPU still getting the voltage somewhere or is it something on the motherboard?...

Yeah Im the same with regards to 24-7 noise... and altho the comp isnt that noisy id rather have it under its max with in reason and 3Ghz seemed reasnoble, but i was hoping to do it at stock volts... but maybe i need to play more.

Erm ill have to have a play with the overclock as ive been runing stock for a few days now... will have a play and get back to you... Overall anything in the pics suggest any major issue?

Appreciate you taking a look

[shazza]

Count me in as another one with a headache! I've read this entire thread over the past week ... thank you so much unclewebb for all your efforts. I've learned a lot, and now am a little less confused about temps. Haven't seen too many i7 examples ... here's my Cool Down test with an air-cooled i7 920 (ASUS P6T Deluxe ... all stock settings). It seems okay, but just wanting to doublecheck:



Thanks again for all the helpful info!!!

[Lsdmeasap]

Logfile or CSV's not being created in 2.84

Is this a known issue?

[day187]

Hi

Is there any chance of putting a Logitech G15 plugin in the Real temp app so it shows on the Logitech LCD

Thanks

Day