Real Temp - New temp program for Intel Core processors

[unclewebb]

There are two different Load Meters in the recent versions of RealTemp. If you go into the RealTemp Settings window and select the TM Load option then you should get a load number very similar to what the Task Manager shows. It uses the same Windows function as far as I know.

The original load meter that RealTemp has been using is not a true load meter. It measures the percentage of time a CPU is in the C0 state. The method I'm using is documented to be accurate for the Core i7 CPUs only. I've always had excellent success with this on my E8400 as well as all of the other 65nm CPUs I've tested on my motherboard.

I know on Core 2 mobile CPUs, this number actually starts to increase as the CPU idles down, similar to what you are seeing. When Super Low Frequency Mode (SLFM) kicks in and various other power saving features, it can go up to 60%. I originally thought that this was a sign that as the MHz went down, the CPU would have to spend more time in the C0 state to take care of the background tasks.

Do you have anything enabled in your bios like C-state tech or similar?

It's possible for this meter to read very low on an E8400. I live in North America and I'm not sure if your motherboard might have some built in energy saving feature or something else that I don't know about.



It looks like the new TM Load meter is a perfect option for you.
The Clock Modulation option can also send the basic RealTemp load meter higher at idle.

Monstru: I'm working towards the next official version. I'm hoping that if I use version numbers up quickly then I will be motivated to finally release 3.40. Mostly minor bugs at the moment and areas where there is always room for a little improvement.

[stasio]

TM Load option works perfect.
Thanks.

[KURTZ]

there is an error for this E6400 B2 ...

[unclewebb]

The first Core 2 I ever owned was an E6400 B2 stepping. I definitely did some hands on testing with my temperature gun on that one.
Happy memories. That one could run Prime stable at 3640 MHz for enough hours to get me into the [H]ard OCP database at a very respectable level for an air cooled processor. Of course, I had to drag my computer out to the garage in the middle of winter for that run but that's one of the perks of living in Canada.

Here's the history of TJMax. When Core 2 CPUs were first introduced, there was a bit in the CPU that software could read to determine what TJMax the CPU was, supposedly. If this magic bit was a 1 then the CPU was TJMax = 100C and if it was a 0 then it had to be TJMax = 85C or maybe that was reversed, I can't remember. It had to be one or the other though, based on this bit. This information was never in any Core 2 documentation. It instead came from some old Pentium 4 documentation that had absolutely nothing to do with the new Core 2 CPUs. One programmer made an assumption and then another one followed and another and etc.

Intel's original IDF presentation brought up the fact that this bit was not defined for Core 2 Desktop CPUs but I guess the other programmers decided to ignore this and they still continue to use TJMax = 85C to this day for the original Core 2 CPUs.



The question becomes, why didn't Intel come forward and set the record straight about TJMax when these CPUs were first released. If TJMax is closest to 90C, like I claim, and software assumes TJMax = 85C then all software will report that these CPUs are running 5C cooler than they actually are. That's a convenient way to create a cool running CPU. If the truth is that the average CPU has a TJMax closer to 90C and not 85C then if software was updated, everyone's CPU would be running 5C hotter than originally thought. Customers wouldn't be happy with that and RMA returns would likely go up.

I can certainly understand the motivation to remain as quiet as possible about this issue. Intel has nothing to gain so why bother.

I will continue to use TJMax = 90C on these early CPUs and it makes me happy to know that no one else has the guts to tell it like it is.

RealTemp originally followed the herd and used TJMax = 85C just like everyone else. By using this number, when the surface temperature of the IHS was 80C RealTemp would also report 80C. Everything looked good.

rge and his testing and papers about heat transfer in CPUs showed me the light. If the surface temperature of the IHS is 80C then the hottest spot on the core is not the same, it has to be higher. Heat dissipates very rapidly over a small distance in CPUs. The docs and rge's testing with a calibrated sensor showed me that the hottest spot on the core is approximately 5C hotter than the IHS surface temperature that I was measuring.

I hated to admit that I was wrong but I was wrong. This proved that TJMax couldn't be 85C. It had to be higher and for this E6400 B2, it was likely very close to 90C. That's why I use that number and why I have decided to go against Intel's PR news release about TJMax where they introduced the term TJ Target which has only confused users even more.

I've tested more CPUs since then and nothing has changed my mind. Intel says they raised TJMax 10C when they introduced the new G0 CPUs to save some money on heatsink costs. They also said that TJMax = 100C for the 45nm E8000 series of CPUs and everyone is happy with that number. When I use TJMax = 100C for a 45nm CPU, the core temperature reports 5C higher than the IHS surface temperature like it should. To get that same relationship with an early B2 CPU, I need to use TJMax = 90C. Using the 80C TJ Target number they released as TJMax would mean that the IHS surface temperature is hotter than the hottest spot on the core which goes against the laws of physics. The source of the heat, the core, can't possibly be cooler than the IHS surface temperature. When my testing was brought to Intel's attention, they chose to quietly ignore it. No surprise.

Kind of a long winded answer to your question. Thanks for bringing this up. Lots of users assume that TJ Target and actual TJMax are the same but they are not.

KURTZ: I just noticed in your screen shot that SpeedFan reports 50C for the CPU temperature. It's impossible for the core temperature to be less than this number. Only RealTemp reports the cores higher than the approximated Tcase temperature which is how it should be.

[loonym]

Thanks for the history lesson unc and also for another great beta. Working flawlessly with my mobile T9550.



alternative d/l link for RealTemp beta 3.37: http://lakesidepc.com/RealTempBeta337.zip

[unclewebb]

Good to see you Loonym. The last few recent RealTemp updates have been for better support for the Core 2 mobile CPUs like you're using. Still a couple of minor issues to look at before the next official release. I think I've been saying that for a few months now but this time I mean it.

I've been meaning to contact you about the XS Bench. rge did some testing with his i7-920 and his i7-950. He was able to find out that this bench runs faster clock for clock on the newer D0 stepping Core i7 compared to the original C0 stepping Core i7 CPUs. This benchmark is fairly small and fits in the on chip cache so it looks like Intel must have done something to speed up the cache performance like reduced latency timing or something like that. If you do any XS Bench mark testing then send me your scores and tell me what stepping CPU you are using. Just curious.

[KURTZ]

thanks for the feed-back Uncle

[loonym]

Good to see you Loonym. The last few recent RealTemp updates have been for better support for the Core 2 mobile CPUs like you're using. Still a couple of minor issues to look at before the next official release. I think I've been saying that for a few months now but this time I mean it.

I've been meaning to contact you about the XS Bench. rge did some testing with his i7-920 and his i7-950. He was able to find out that this bench runs faster clock for clock on the newer D0 stepping Core i7 compared to the original C0 stepping Core i7 CPUs. This benchmark is fairly small and fits in the on chip cache so it looks like Intel must have done something to speed up the cache performance like reduced latency timing or something like that. If you do any XS Bench mark testing then send me your scores and tell me what stepping CPU you are using. Just curious.

This is pretty much in line with what I've seen in regards to stepping and scaling. I'll try to get a little more data and report back.

[randomizer]

My cousin has a Q9550 E0 and upon loading up Real Temp it was using a Tj Max of 90C (it is supposed to be 100C by default isn't it?) and it didn't show the CPU model at the top left as it should. This was using Real Temp 3.00. I thought it was kind of odd

[unclewebb]

If you ask your friend to send me a screen shot of the latest version of RealTemp and CPU-Z on the same screen then I should be able to fix this up for him.

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

I'm always fixing little things like this as I go along so if I've already fixed it in the latest version, then I won't have to try and fix it again.

Sometimes the string of information within the CPU leaves out a few details. Many ES processors are like that. When that's the case, I can usually add an extra line of code to get it properly recognized. If your CPU is top secret then just send me a PM and I'll keep 'er quiet.

This string of information is displayed in the Specification box of CPU-Z. RealTemp takes that info and tries to figure out what CPU you have and what TJMax it should be using. Obviously, it doesn't get them all right but there's usually a reason. I've seen some CPUs where that string of information was not written to each core of a quad core by Intel. I think I changed RealTemp a while ago so that it tries to read this info from core 0 when that happens.

Edit: As for the GPU temperature not showing up properly, that might be an Nvidia issue with their latest drivers. Scroll back a page and you should see a fix for that as long as the temperature sensor is not damaged.

[jcool]

How's the Gulftown-ready beta coming along?

[nigelke]

Is having fun with Realtemp ^^

[unclewebb]

How's the Gulftown-ready beta coming along?

Probably a few more days. I take it easy on the weekend. Too many games to watch on TV.

Don't worry jcool. When I get it done, you'll be my tester.

Kind of interesting that this new 6 core CPU seems to be organized internally as 2 separate 3 core CPUs sort of like how the original Core 2 Quads were really 2 Dual Cores stuck under the same IHS. The 32nm Dual Core also seems like it is really a triple core CPU with the center core disabled.

http://www.xtremesystems.org/forums/...postcount=3727

In 2010 when Intel starts selling triple core CPUs, remember, you heard it here first.

A triple core with hyper threading would be a great processor for enthusiasts and should overclock like crazy. It would leave AMD's triple core CPUs in the dust.

[randomizer]

It shows up fine in 3.37

[nigelke]

hmm I am working on a Dell Desktop [optiplex gx620]
and when i run Realtemp I get an error that this CPU is not supported
Its a Intel Pentium D

I will try to make a screenshot later today

[jcool]

Probably the internal suckage sensor alert ("Warning - Netburst detected. Shutting down immediately.")

[randomizer]

hmm I am working on a Dell Desktop [optiplex gx620]
and when i run Realtemp I get an error that this CPU is not supported
Its a Intel Pentium D

I will try to make a screenshot later today

The Pentium D doesn't have a Digital Thermal Sensor, which is required for Real Temp.

[nigelke]

ah i feel stupid

[unclewebb]

It shows up fine in 3.37

Thanks for posting that randomizer. One less thing to fix today.

If your friend is curious, he can try running this program:

http://www.fileden.com/files/2008/3/...7/CPU_Name.zip



It just reads the name string from each core of a CPU so his Quad should show the same string of info for each core. RealTemp 3.00 was probably just having a bad day. I think it's very rare for this info to be missing from any of the cores.

Tell him to read page 150 to look for ways to get his GPU temps working correctly.

[RejZoR]

What is "TM Load" used for exactly?

[unclewebb]

What is "TM Load" used for exactly?

Good question. The original RealTemp load meter was based on the percentage of time the CPU is running in the C0 state. This is documented for the Core i7 and I found it gave great results on my E8400 as well so that's why I originally decided to use it.

On a Core 2 mobile CPU that supports deeper C sleep states, the C0% number actually goes up significantly as the CPU idles down and no longer represents CPU load. My best guess is that in this situation, as the CPU idles down and the cores start to go to sleep, a CPU has to start spending a bigger percentage of its awake time in the C0 state to get the background tasks taken care of. I've seen this meter report as high as 70% on a mobile chip during this situation. Kind of odd but this high number is actually a good sign. It shows at idle that your CPU is mostly asleep, which is a good thing, and it's running very efficiently.

No one ever complained but most laptop owners probably don't like seeing a 70% load number when their CPU seems to be idle. If you have that problem and don't like it then you can use the TM Load option which calculates the CPU load using the same Windows function that the Task Manager uses to calculate load.

I prefer the original RealTemp method because it is more efficient and more accurate, especially in XP, as long as you're not using the deeper sleep states. On a Core i7 or i5, I'm not sure how the two meters compare. It might depend on whether you have C3/C6 and C-States enabled in the bios.

[punkrockpolak]

Informative post, uncle. I'm not sure I'd be too happy seeing my laptop at 70% when it's idle. It's one of those things that irk you when you know they shouldn't heh.

[Ajfina]

i ran core temp and this one and both were displaying same temps...

[randomizer]

@unclewebb: I'll tell him to take a look at that program, thanks.

I've heard plenty about C3/C6 states, but what about C7? It doesn't seem to get much of a mention anywhere. My BIOS allows me to limit how deep a C-state the CPU will go, and I can limit it to C1 through C7. I can also demote C3/C6/C7 requests to C1, or C6/C7 to C3 "based on uncore auto-demotion information" if I want to, although I'm unsure why I would.

[unclewebb]

punkrockpolak: As long as you understand what the C0% meter is telling you, it can be useful on a laptop. For comparison, I have a T7200 that shows a C0% number of 10% when the CPU is idle. It idles at (6.0x166MHz) 1000 MHz.

A more modern P8400 supports Super Low Frequency Mode (SLFM) which means the multi can drop down to 3X at idle so that's equivalent to (3.0x266MHz) 800 MHz. It also supports a feature where internally, it can ignore half the clock pulses so it's really only running at 400 MHz. It can use C3/C6 which the T7200 I have doesn't support so when a P8400 idles down, it has to work like hell when it is awake to process the background tasks because it is effectively running so slow and is going into sleep mode half the time. The Task Manager load meter will show 0% or close to it when this is going on but that's only telling part of the story. The C0% might be up in the 60% range on a P8400 which at least gives you some idea of what's really going on inside the CPU. I think both load meters tell you something so I included both of them.

randomizer: I was helping a user the other day and suggested that he should try enabling C3/C6 so his multiplier could reach its highest value. He wasn't planning to overclock the BCLK so I thought this might give him a little boost in performance. He didn't like it.

He thought his system felt less responsive. On his i7-920, enabling C3/C6 allowed his average multiplier to go from 21.0 to 21.3 while running a single threaded bench like Super PI. 133MHz BCLK x 0.3 is only equivalent to a 40 MHz boost. Not only did he not see an increase in performance, he was losing performance in Super PI. He seemed to have a lot of Vista baggage interfering with his testing but even best case, you're not going to see a huge difference in performance if your multi is only changing by an average of 0.3.

I haven't played with C7 so I can't say what the results would be. You might save a tiny amount of power and maybe your cores will idle a degree cooler.

It's up to each user to test these things out and see if they provide any benefit. When overclocking, most motherboards don't drop the core voltage at idle so a lot of these power saving features are not that useful for the typical XS reader.