Real Temp - New temp program for Intel Core processors

[unclewebb]

randomizer: How about a screen shot of RealTemp, Core Temp and HWMonitor while running Prime 95 Small FFTs on all 8 threads. Once your core temperatures have stabilized, are all three programs reporting the same now? I've seen quite a few screen shots where HWMonitor is reporting a few degrees different than RealTemp and I'm not sure why. It might have been the older version.

HW Monitor used to interfere with the system timers that RealTemp and i7 Turbo use so RealTemp 3.00 would show some sky high multipliers. Can you see if this bug has been fixed?

[rge]

Someone was just commenting about HWmonitor reading different temps on OC.net a while ago...so was checking mine. HW monitor does report 4 core temps, but temps are still out to lunch. It reads 2-3C too high at idle and 3-4C too low at load compared to real temp and coretemp. Cant make any sense out of it...thought about some long time rolling average, but at idle, just does not make sense.

RT, coretemp, HWM at idle


all 3 at load

[unclewebb]

Thanks for the numbers rge. I also saw that post on OC.net about HWMonitor reporting different core temperatures and it just doesn't make any sense. TJMax is written into each Core i7/i5 CPU which RealTemp and Core Temp are reading correctly and RT and CT are both reading the temperature sensors correctly for each core so I'm not sure what HWMonitor is up to.

The formula can't get much simpler than this:

Reported Temperature = TJMax - Digital Thermal Sensor reading

You can't argue about those two numbers on the right so all software should be reporting the exact same thing when the core temperatures have been allowed to stabilize. During transitions between idle and full load, it's possible for two different programs to be reading the sensors at a slightly different moment and for two programs to report slightly different temperatures but at full load or at idle, all 4 cores should be almost identical. Core Temp and RealTemp show that.

The only difference between RealTemp and Core Temp would be if the APIC ID reported by RealTemp isn't in numerical order. When that happens, RealTemp sorts the temperature data into the correct physical order while Core Temp does not. This used to be an occasional issue with Core 2 Quads on some motherboards but most of the newer socket 1366 or socket 1156 CPUs and motherboards are usually OK and the bios correctly assigns threads in the correct order.

[unclewebb]

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

This is a bug fix version.

In SLI mode the GPUs should be labeled as GPU 1, GPU 2, GPU 3 etc.

The second bug is more serious and could result in a RealTemp crash when not using an Nvidia card by clicking on the Reset button in RealTemp. Both bugs are now fixed.

Edit: I see a new RealTemp feature on the horizon.

[randomizer]

As you requested, below is a comparison of the 3 programs (load being 8 threads of P95 Small FFTs). The results, like rge's are odd, but not as dramatic. Load is consistently 1C below Real Temp on all cores, and would be for Core Temp save for the fraction of a second difference in update times between Real Temp and Core Temp. Idle is almost identical to the other two programs. How is the power consumption for the CPU calculated, just using a simple formula based on TDP, voltage and clock speed? Or is there some way of reading the power consumption? I noticed that rge's is 130W flat consistently.

Also don't look at my chipset temp (TMPIN1), it could shock you. It was 10C higher yesterday since it was 29C ambient

Idle


Load

[rge]

HWM definitely read Randomizers better. Since programs sample at different times, best way is to compare is max temps under load, they should all be nearly same.
Coretemp and real temp max load are both 68, 68, 68, 66. HWM is 67, 66, 67, 65. Mine is 3-4C off, guy on OC was 2C off, randomizer is 1-2C off. Interestingly, I put my pc back to stock, then HWM was only 2C off at load on 3 cores, 1C off on 4th. Then OCed to 4.55 all were off at least 4C with HWM at load.

Randomizer, can you try OCing your cpu and see if error increases while overclocking like it did on mine.

[kpablo]

I remember Unclewebb saying about implementing a G15 plugin in to the Real Temp, one independent to Riva Tuner... this have been already implemented?

[unclewebb]

I remember Unclewebb saying about implementing a G15 plugin in to the Real Temp, one independent to Riva Tuner... this have been already implemented?

Sorry, that's still on the things to do list.

[randomizer]

Randomizer, can you try OCing your cpu and see if error increases while overclocking like it did on mine.

Great, now I have work to do Are you theorising that the overclock itself is causing the mismatch in temperatures, or just the resulting higher temperatures? If the latter, I could just turn off my CPU fan

[rge]

Great, now I have work to do Are you theorising that the overclock itself is causing the mismatch in temperatures, or just the resulting higher temperatures? If the latter, I could just turn off my CPU fan

I was wondering if increase power cause more temp fluctuations, and HW monitor is doing some type of rolling avg of more than just the hottest core. TAT (intel thermal analysis tool) interfaced with PECI which avg ?more than just the hottest core, so you got some weird avg, which unfortunately did not give you the correct hottest temp, which in protecting hardware is really all that matters. One of HWM updates mentions HECI, so dont know if that is what they are doing, ie something like TAT.

Just trying to see if it is increased power causing increased loading fluctuations or if its just different cpus/different mobo, etc. It is not temp itself, already tried that by turning fans off, increased error is related to increase power dissipated.

Basically just curious wtf HW is doing....clearly not averaging just hottest core, I tried testing for that.

[randomizer]

If power consumption is what you want to test, I'll just play with the vcore. I have mine undervolted by 0.16V so by restoring it to stock and even overvolting it should show if power consumption is the reason, or at least a contributing factor. It will be interesting to see how the power consumption reading in HWmonitor changes as well (if it does). Mine doesn't seem to be locked at the TDP like yours.

[unclewebb]

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



I've added ATI GPU temperature monitoring to this version. I am using the ADL (AMD Display Library) SDK 2.0 to accomplish this which isn't the most efficient hardware monitoring library.

Update: This seems to work on the 2000, 3000, 4000 and new 5000 series of ATI cards. The above post of my X1950XT was just for an example to show the general layout of the new ATI Information window.

On some GPUs in some situations, the performance of this library is terrible. On slower cards, I wouldn't recommend sampling the ATI GPU temperature sensor more than once every 5 seconds. RealTemp also includes the ability to easily temporarily disable this feature if you want to run any A / B comparisons or benchmarks.

If this new feature works for you without any issues then that's great. If not, you can always turn it off completely by using the INI option:

NoGPU=1

I haven't heard of any problems yet on systems with single GPUs. If you are using a x2 or any unique CrossFire version then some feedback would be great. RealTemp is designed to report the highest temperature of the hottest GPU in CrossFire mode in the button on the main GUI. This button opens up another window where you are supposed to be able to monitor each GPU individually. With multi monitors it might label things funny like, GPU 1, GPU 2, GPU 3, GPU 4 but if you have two GPUs in your system then GPU 1 GPU 2 should belong to the first GPU and GPU 3 GPU 4 should belong to the second GPU. I wish I had some more time and some actual ATI hardware to test on but I've had to wing it because my new 5750 crapped out half way through development. I decided to get a refund rather than try my luck again. I might wait for a 5850 next time I'm feeling lucky.

On Nvidia systems, RealTemp can report the GPU temperatures for each GPU in SLI mode but if you are not in SLI mode, then it can't report both GPUs. That's a work in progress.

One user had an ATI card as his main card and an Nvidia card as his second card. RealTemp can't correctly handle situations like that either. It's one or the other at the moment.

I added another INI option that might help a little in the above situation:

NoATIGPU=1

That might allow only the Nvidia GPU to be detected and reported in RealTemp so you can choose what one you want to see.
If I had some more hardware to recreate the infinite number of combinations I could probably do better but I don't.

If this new feature is of any use to you then post a screen shot.

[burebista]

If this new feature is of any use to you then post a screen shot.

Of course. 2600XT.

[unclewebb]

Thanks burebista. I thought it might work on the 3000 series but I wasn't expecting to see it work on the 2000 series.

It does not work on my 1950XT.

[WaterFlex]

Thats a great idea, Bob I`m about GPU monitoring.

[Monstru]

Works like a charm for me

[unclewebb]

Thanks for the feedback. I might have to do a minor adjustment to better support CrossFire. At the moment when a GPU goes to sleep on a x2 system, RealTemp might not be able to read the temperature history of both cores. I'll see if I can find a work around for that issue without having to wake up that core.

GPU-Z will always be my first choice for GPU monitoring but RealTemp looks OK as a second choice when you just need to see the temps.

[Raptor-X]

Unclewebb:
Have you published a guide regarding idle calibration & TJ Max settings? I suspect that I need to tweak my settings...

cooldown.jpg

THANK YOU for the great tools.

[unclewebb]

Raptor-X: Your welcome. Don't let the beta word scare you. RealTemp 3.46 is looking like a good one.
The download also includes some of the other tools I've been working on like i7 Turbo.
http://www.fileden.com/files/2008/3/...alTempBeta.zip

There is an old guide about calibrating RealTemp that was written with the 45nm Core 2 in mind but it hasn't been updated. CompuTronix over on Tom's Hardware has an extensive guide that you might want to check out.

I find that the sensors on the i7-900 series are excellent compared to the 45nm Core 2 Duo and Quad sensors so calibrating is much easier. I don't find getting too technical about calibrating to be worth it anymore. The sensors on your i7-960 seem to be working just like all of the other Core i7-920 CPUs I've seen. When installed correctly, almost all of them follow a very familiar pattern.

Core 0 tends to be the most accurate so I always recommend to just leave it as is. At full load running Prime 95 Small FFTs, core 3 always reports that it is running 5C cooler. If you haven't read the last 100 pages of this forum in your spare time then somewhere I came to the conclusion that Intel very likely sets TJMax slightly higher on this core to better control thermal throttling so all 4 cores don't reach the throttling point at the exact same time which might noticeably kill performance. The two center cores are usually somewhere in the middle and often times very similar to core 0.

Your numbers show slight differences in the slope of the temperature curves of the first 3 but they're all so similar that it's not worth trying to make them perfect with calibration factors. The quick calibration I've come up with is to simply adjust TJMax. I'd set TJMax to 100, 99, 100, 104. This should get core 0 and core 1 to line up a little better from idle to full load. Core 2 shows less Distance to TJMax at idle compared to core 0 but it shows slightly more at full load. Actual TJMax might be closer to 101 or 102 on this core but the slope error cancels some of that out at idle so I think you should just set it to 100 like core 0 and keep things simple. Core 3 shows a 4C difference at idle compared to core 0 and that difference grows to 5C at full load. Once again there is a slight amount of slope error so if you want your idle temps to look nice set Core 3 to TJMax = 104C and if you want your full load temps to be accurate then set it to 105C.

A degree one way or the other isn't worth worrying about. Slight differences in how you applied your thermal paste might be responsible for some of these slight differences but I wouldn't waste time re-doing it unless you think that you really botched it. The truth is that you can apply your paste 101 times and lap your CPU and heatsink until your hand hurts but the 5C difference between core 3 and core 0 is not going to go away. Do a quick calibration adjustment or just ignore it like everyone else seems to do.

[Raptor-X]

Unclewebb:
Thank you! As always, your efforts are greatly appreciated. I will give 3.46 a spin tonight.
Regards,
Raptor

[stasio]

Unclewebb:
Why RealTemp GT show only 3 cores ?

[Monstru]

I think GT was for Gullftowns, no?

[unclewebb]

That's right. The GT version is designed for the 6 core Gulftown CPUs. Internally, these new CPUs are organized as 2 x 3 core CPUs. When you run RealTemp GT on a Quad, it will only show you the first 3 cores. I could probably fix this but the regular version of RealTemp should be used if you have a Dual or Quad core so I left RealTemp GT as is to encourage users to use the regular version.

RealTemp GT will also sort of work on the dual Quad Gainestown systems but at the moment it will only report the first 3 cores of each CPU.

[stasio]

That's right. The GT version is designed for the 6 core Gulftown CPUs. Internally, these new CPUs are organized as 2 x 3 core CPUs. When you run RealTemp GT on a Quad, it will only show you the first 3 cores. I could probably fix this but the regular version of RealTemp should be used if you have a Dual or Quad core so I left RealTemp GT as is to encourage users to use the regular version.

RealTemp GT will also sort of work on the dual Quad Gainestown systems but at the moment it will only report the first 3 cores of each CPU.

Yeah,I suspect must be this.Maybe will be easier for you to have one common RealTemp.

[unclewebb]

Maybe will be easier for you to have one common RealTemp.

It probably would have been easier for me but I was thinking about users. 99.999% of users won't be spending $1000 to have a 6 core CPU for a long time so I thought, "Why should I make the RealTemp code more bloated just for a tiny percentage of users?"

I hate inefficient software so that's why I created a separate version for the Gulftown guys. The GT version was easy to get up and running quickly without worrying about screwing up the regular version of RealTemp that most people use.

More work for me but a more efficient program for both users.