Real Temp - New temp program for Intel Core processors

[randomizer]

My E6600 looks the same in XP with C1E enabled and EIST disabled. It floats around for a while but finally settles at 0.667

[rge]

Both turbo programs worked well for core i7 with EIST/C1E disabled (2nd pic). But actually the first turbo program works better for reading with speedstep enabled (2nd pic), it is like the second keeps taking a reading and kicking it out of speedstep

But for calculating mhz, turbo on or off, both work, just the first one worked better for speedstep readings. Of course I have nothing to compare with for sure, since yours is only program reading multi correctly with speedstep on.

[bowman]

Like this?

[unclewebb]

bowman: The way Core i7 works is it has a base multiplier and then if the Turbo option is being used, it will bump the multiplier up by +1 or +2. In your picture the tester shows 20 as your base and it has been bumped up from 20 to 21 which is 5% higher or (1.05 times) as much.

The problem that rge discovered with RealTemp 2.89 and Everest is that when he was manually setting a lower multiplier in the bios, when Turbo mode kicked in, it would boost his multi right up to the max in one step. This is how Intel designed this feature.

This would cause the actual multi to jump from say 16 to 23 but RealTemp and Everest were only reporting a +1 boost from 16 to 17. CPUZ is handling this problem OK but it's reporting half multis (0.5) for the Core i7 which I don't believe exist. That's more of an averaged multi.

Edit: The other problem is if you use C1E, then you need to set your Power Options appropriately for it to work properly with Core i7 or any Core processor for that matter.

Here's what happens in XP:





Unless you specifically set your Power Options / Power Schemes to Portable / Laptop, you may not get the full benefit of C1E if you are using a Quad or Core i7. When set to Home/Office Desk, the multipliers on my Q6600 are going back and forth from 6.0 to 9.0 on various cores continuously. Most software may report it as 6.0 on all cores or 9.0 but this testing program would disagree. Watching what's gong on in MSR 0x198 of a Core 2 CPU would also cause most users to disagree.

The same thing happens in Vista. With Vista you need to set your Minimum processor state to about 50% or lower to make sure it keeps all of your cores at the lowest possible multiplier when C1E kicks in.

Now that I fully understand what's going on, a solution for RealTemp will be easy.

[randomizer]

Is this a problem only with quads?

[Vatos_locos]

is this normal ? almost 10 *C difference on the core 0 and core 3
tj @ 100 on all cores

[bowman]

Oh right. The ASUS doesn't let me do that. Either Turbo on and multiplier at Auto, or new multiplier and no Turbo. I guess that's the reason they did that.

[unclewebb]

Is this a problem only with quads?

This seems to be an issue for Quads or Core i7. To me it looks like a bug in the operating system.

Edit: For me I think it might also be a problem with my old motherboard not getting along with a Quad. When I turn SpeedStep off in the bios, the MSR Tool shows that it is only getting turned off for 2 of the 4 cores.

If your C1E / SpeedStep bios settings don't agree with your power profile within your operating system then the multiplier that you end up with might not be consistent and your actual multiplier will float around. If C1E / SpeedStep is enabled in your operating system and your bios, then a Core i7 at idle will be nice and steady with the correct 12 multiplier.

Vatos_locos: Your results look very normal. At full load where these sensors are a little more accurate, your screen shot only shows a difference of 2C. I'd say that's better than average.

Post a screen shot of the CPU Cool Down Test if you want me to have a better look. Some sensors can get stuck at lower temperatures and it's common for them to move at different slopes. Your 45nm Quad is far better than some of the ones I've seen.

[Hardc0r3]

here are my readings from an intel e8400 e0 on abit ab9 quadgt official bios 17(cpu under test) idle with c1e and EIST, everest read and bios read 1.03v idle vcore and min settings in bios is 1.215v . Also@3600 mhz 1.215 vcore without c1e and EIST and room temperature 23.9 c. I manually set tjmax to 95 from 100 both on realtemp and on everest , cause with tjmax @ 100 i get stuck @ 42 ,39 , and it think its wrong, its 2 much . Scythe infinity fan.

[unclewebb]

Hardc0r3: If you have sticking sensors, you shouldn't adjust TJMax to make your temperatures look nice at idle. Try running the CPU Cool Down Test and post your results. It might show you exactly where your sensors stick. Irfanview is a good free program that will let you crop your screen shots down. With the left mouse button draw a box and then go into the menu, Edit, Crop Selection

[Vatos_locos]

thanks unclewebb here is the cooldown test you asked but at difrent voltages/infrequences than before
the cpu and the chipset are under water with a triple rad if that helps

[unclewebb]

Vatos_locos: The slopes of core1, 2 and 3 are pretty similar. Core0 is a little different.

Intel uses slightly different TJMax values from one core to the next. I would recommend using TJMax=100, 100, 102, 102 for your 4 cores. I would also use -2.8 for a correction factor for core0.

If you want to test this calibration, at low MHz and low core voltage you should get something like rge outlined here:

http://www.xtremesystems.org/forums/...postcount=2429

For a water cooled Quad at 2000 MHz and 1.10 volts, about 7C over your water temperature near your CPU is typical.

None of your sensors are sticking in your screen shot.

[Vatos_locos]

ok thanks uncle :p i'll do those settings

the only problem is that the lower vcore i can set on my board is 1.2 :p but i'll try it

[Hardc0r3]

uncleweb but i did run the cooldown test , look @ the 2 screenshot, and i changed tjmax after reading in here that idle temp should be 8 to 9c more than room temperature. Senzor are missreporting below 25%

[unclewebb]

If you enable SpeedStep and C1E and set your voltage to Auto, it might drop lower than that. Make sure your Power Options are also set correctly to get your board to drop down to the lowest voltate and multi.

Hardc0r3: Now I see your Cool Down Test. Your screen shot was so big it was hard to see it.

It shows that core0 is getting stuck at 58 and core1 might be sticking at 61. Try running your computer at low volts and low MHz. 2000 MHz and 1.10 volts if possible. You need to see if your core1 sensor can go to a bigger number than 61 ( if core1 can report a cooler temperature).

Rule 1 is you can't calibrate a stuck sensor and you shouldn't adjust TJMax lower to try and compensate for a stuck sensor either.

[Hardc0r3]

If you enable SpeedStep and C1E and set your voltage to Auto, it might drop lower than that. Make sure your Power Options are also set correctly to get your board to drop down to the lowest voltate and multi.

i did, look @ the last screenshot is idle temp , in bios as i said minimum which can be set is 1.215, but even with c1e and eist, cpu default fsb 333 and multi 6x, and vid going down to 1.03 (everest is the only software that can read corectly my cpu vcore together with abit uguru app on my MB Abit ab9 quadgt ), the temperature won't go below 37 with 34(the stupid core 1 senzor is stuck @ 58 distance to tjmax), which is the exact same idle temperature when i am @ 3600 1.215v without c1e and eist.
Later edit:

should i put it back to 100? cause then, instead of having 37 with 34 which is 10c more than my room temp when idle, i will get 42 with 39 minimum and that is 15c more than room temp?
SOrry everyone for my shots i am running on 24" monitor i will adjust those from now on.

[unclewebb]

If your sensor(s) are stuck then your idle temperatures will not be correct. These sensors aren't designed for accurate idle temperatures anyhow. Set TJMax back to 100 and ignore your idle temperatures. At least your load temperatures will be fairly accurate if you do this.

[kasio]

My i7 920 using stock settings/cooler. Temps are way too high and I'm not sure what the problem is.

Anyone got any advice?

[rge]

Well good job unclewebb, looks like real temp 2.89.5 gets to be the first to accurately read mhz and multi correct on i7
turbo on 182x22 bios. So should read 182x23 with light load on 1-2cores and idle. And 182x22 at full load.

At idle, turbo on, both cpuz and realtemp read accurately at 182x23, coretemp is bugged/not reading mhz/bclk correctly.

At full load all 4 cores loaded, realtemp and cpuz read correctly at 182x22, coretemp bugged.

But at only partial load with turbo enabled only realtemp is reading correctly, see pic. cpuz is reading half multis/quarter multis...not really sure what. Coretemp again is bugged.

[unclewebb]

http://www.fileden.com/files/2008/3/...alTempBeta.zip

Thank you rge. I couldn't have done it without the documentation you found for me and all of the testing you've contributed to getting this working right. Core i7 is a little different than Core 2 so it took a while to find out what's changed. If you notice any other issues just let me know.





kasio: My opinion is that the Intel OEM cooler is barely adequate for Core i7. If you are interested in getting more out of your CPU then I would highly recommend an aftermarket cooler. There was a reason that Intel was handing out Thermalright Ultra eXtremes to testers when they introduced Core i7.

You can try remounting your cooler and maybe pull your board if you're not sure if the push pins have seated properly. For what it costs for an i7 and a motherboard and some DDR3, you might as well blow the budget a little more and get a decent cooler. You're already within a couple of degrees of thermal throttling starting to kick in.

[kasio]

Thanks unclewebb, I probably will get a new cooler anyway but I was worried that these temperatures were not normal. I did try remounting the cooler with new thermal compound but there was no improvement. I need to clarify that these temps are normal for a stock cooler and that my CPU is not faulty.

[randomizer]

Oops

[unclewebb]

http://www.fileden.com/files/2008/3/...alTempBeta.zip

I always liked the CPU Load feature in Core Temp so I added this today.



It compares the time your CPU spends in C0 vs C1 and seems to give a good approximation of CPU Load without needing much overhead to calculate this. My Quad with two threads of Prime looks about right.

Now the question. Is this important information?
Should I update RealTemp and make it look like this.



or should I try to add some progress bars like SpeedFan uses?



What do users want to see? It might also look better balanced with CPU Load put in the middle between Distance to TJMax and Minimum

[burebista]

Maybe above Temperature and make it progress bars.

[unclewebb]

I just tried Core Temp with 1, 2 or 3 threads of Prime. The load gets shared amongst all 4 cores so the percentages are constantly jumping up and down and it looks like useless information to me. I kind of like the single load number which was nice and constant when testing. I might keep it as is. You can never go too wrong by keeping things clean and simple.

Edit: I've noticed one interesting thing so far. The CPU Load percentages in the CPU Cool Down Test are approximations based on the Intel documentation. If you have the main RealTemp window set to CPU Load while you do a cool down test, you'll notice a big difference for some of the lower steps between theoretical CPU Load and actual CPU load. I think I will update the Cool Down test to display the actual percentages. A graph of the temperature curve might be more interesting now.