Real Temp - New temp program for Intel Core processors

[randomizer]

After running SuperPi with Affinity set to CPU 0 and all major background tasks either killed or set to CPU 0 only I still only managed to reach 21.8x with my i7 920. The other threads were usually managing a C0% of <3%, some consistently <1%. With unaltered affinity (except for SuperPi to prevent load balancing) it typically doesn't break 21.4x. Turbo might be useful for a single speed bin increase on LGA1366 but it certainly doesn't do much better than that.

[unclewebb]

randomizer: Could you try another comparison? I'm interested to see what's better without changing the Affinity of any of the background tasks and without killing any of them either. More like how the average person has their computer set up.

In this condition, is it better to use Set Affinity.. and manually lock Super PI to a single thread or core or is it better to leave Set Affinity... alone and allow the CPU to manage this and let it move Super PI to whatever thread it wants? What's your estimate of the average multiplier in each case? On a Core i7 with hyper threading enabled you could allow Super PI to run on the first two threads since they usually both belong to Core 0 as long as RealTemp is showing that your APIC ID as 01234567.

The new socket 1156 CPUs that have 4 or 5 bins of turbo boost available get more out of leaving C3/C6 enabled so the CPU can use the highest multiplier. The Core i7-9xx series processors that can only average an extra multiplier boost of 0.3 or 0.4 beyond the +1 turbo boost isn't really worth while. I assume that's why most performance motherboards disabled this extra boost as soon as you start overclocking. Better to have it locked at a steady 21.0 than jumping between 21.0 and 22.0 which is likely to cause instability when overclocking.

[Monstru]

Hey Kevin,

now everything looks normal

[unclewebb]

A 32nm hyper threaded dual core looks like fun. I hate heat and fan noise so this finally looks like a CPU I'd like to have; once all the Foxconn sockets are off the market.

RealTemp still needs to call your new CPU a Core i5 instead of a Core i7 but at 4.7 GHz, who cares?
I think I fixed that minor bug in version 3.37 or 3.38



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

The significant difference in core temperature at full load while running Prime 95 Small FFTs looks like more evidence that Intel "accidentally" sets TJMax slightly higher on core 1 compared to core 0 to better control thermal throttling.

I hope they admit to that someday. Thanks for sharing.

[Monstru]

Bwhahahah! Now I will try 3.38, thx Kevin.

Edit - Uf, it says 3.37

Anyway, the i7 bug is solved. Here ya go:

[unclewebb]

Edit - Uf, it says 3.37

Anyway, the i7 bug is solved. Here ya go:

Thanks for showing me that. I thought I fixed that but I lose track sometimes.

There's no major difference between 3.38 and 3.37 so you're not missing anything important. It was some minor adjustment for a mobile CPU. I think I uploaded it to SendSpace instead of my regular FileDen.

[randomizer]

randomizer: Could you try another comparison? I'm interested to see what's better without changing the Affinity of any of the background tasks and without killing any of them either. More like how the average person has their computer set up.

In this condition, is it better to use Set Affinity.. and manually lock Super PI to a single thread or core or is it better to leave Set Affinity... alone and allow the CPU to manage this and let it move Super PI to whatever thread it wants? What's your estimate of the average multiplier in each case? On a Core i7 with hyper threading enabled you could allow Super PI to run on the first two threads since they usually both belong to Core 0 as long as RealTemp is showing that your APIC ID as 01234567.

What should I do with i7 turbo (ie. a program that most people won't be using all the time)? Leave it to use all threads as well?

[unclewebb]

To accurately gather data from your computer, RealTemp and i7 Turbo need to be able to access all of your cores. Adjusting them with Set Affinity... isn't a good thing to do. The log from RealTemp might be your best bet when gathering some data. I need to update the logging abilities of i7 Turbo so it gathers and saves data for each thread. I'm not looking for a super scientific test. Just some numbers. A multiplier difference of 0.1 or 0.2 isn't something a person would ever notice in normal use. If we need a more scientific tool, I'll tune up i7 Turbo with some better logging abilities.

[Monstru]

Thanks for showing me that. I thought I fixed that but I lose track sometimes.

There's no major difference between 3.38 and 3.37 so you're not missing anything important. It was some minor adjustment for a mobile CPU. I think I uploaded it to SendSpace instead of my regular FileDen.

Link please

[stasio]

Link please

RealTemp 3.38
http://www.fileden.com/files/2008/7/...eta%203.38.zip
or
http://www.xtremesystems.org/forums/...&postcount=212

[Monstru]

Thx stasio!

[Wishmaker]

RealTemp 3.38
http://www.fileden.com/files/2008/7/...eta%203.38.zip
or
http://www.xtremesystems.org/forums/...&postcount=212

I don't find something normal :p. I get 20 degrees less than before. I wish my Asus would have that temperature. Unfortunately, I get idle temps the way 3.37 and 3.36 show . So, I guess ... HOUSTON!!! WE HAVE A PROBLEM

[burebista]

Yep, but you have one constant value: distance to TJMax. That's why I keep telling that temperature related with TJMax is irrelevant as long as the only value that's matter for your CPU thermal behavior is distance to TJMax.

[unclewebb]

Wishmaker: Thanks for posting that. Maybe that's the reason I didn't want to share version 3.38.

I was moving some code around so it looks like it's still a work in progress. If you want you can manually set TJMax to 105C.

At least it looks like it is showing the Intel Dynamic Acceleration feature of your CPU. Run a single thread of SuperPI and during the middle of it take a screen shot so I can see what your multiplier is reporting. This CPU can use a 13X multi when only a single core is active. The average is usually less than that due to background processes.

[unclewebb]

By default, version 3.38 seems to be using the correct TJMax=105 for a T8300

Can you go into your Settings window and click on the Defaults button?

Do the numbers at the top beside the Set TJ Max button show 105? It's possible that it found 85 in your INI file and decided to use that but the Default button should set this to the correct 105. If it does then click on OK so it can save it and overwrite the 85 that's in your RealTemp.ini file.

[jcool]

New RT Beta for Gulftown

[unclewebb]

New RT Beta for Gulftown

Shhh, it's a secret. if stasio or Monstru find out, they'll hunt you down.

I just sent you the beta 2 version. Beta 1 wasn't reading TJMax from all 6 cores so hopefully that's fixed now. When I get the Cool Down Test and Settings window fixed up, I'll post a download link.

[stasio]

First 3 core TJ Max 102.
But nice,waiting for link.
Edit:
Core Temp show VID!

[jcool]

Due to popular demand... Q3QR on water, realtemp GT Beta 2:

[unclewebb]

Core Temp show VID!

I think Core Temp is assuming that it is some sort of Core 2 processor so it tries to do a VID calculation on whatever data it finds. There's still no known way to read VID from Core i5/i7/i9 CPUs.

Thanks jcool. That looks better. Now it's reading TJMax = 102 from each core correctly.

[nigelke]

Jcool: thats sweet

[stasio]

Thanks jcool. That looks better. Now it's reading TJMax = 102 from each core correctly.

Uncle,as per your post:http://www.xtremesystems.org/forums/...&postcount=240
TJMax = 105

[unclewebb]

I believe that jcool switched to a different CPU today when the mail man showed up. My MSR Tool isn't set up to read all 6 cores yet but hopefully it is adequate to read MSR 0x1A2. If anyone has a 6 core CPU, can you click on the link in stasio's post and run the MSR Tool program and post your results. If RealTemp is using 102C for these CPUs then we should see 66 hiding in that hexadecimal number.

[jcool]

Yeah, that is a Q3QR 3,07Ghz Xeon. The other ones were Q3QS (2,53Ghz Xeon) and Q2WV (2,4Ghz Xeon). All are Westmere-EP though, not Gulftown (meaning these are all Dual QPI DP chips, as opposed to Gulftown = i9 XE w/o 2nd QPI link)

Here's your MSR, it is indeed a 66 (evil CPU? )

[stasio]

66 hex = TJMax 102