Something fishy about TAT....

[rdhood]

I don't trust the TAT, and here is why:


First, I have a Gigabyte DS3 v3.3, and an E6300 overclocked to 3Ghz.

Let me preface this by saying that I start running two prime95's in the background so that both CPU's are at 100% load before ever running TAT. So even when I am at "idle" in TAT, the CPU's are 100% loaded.

When I run the Intel's TAT 100% load, Temp spikes to 56C. When I stop, temp drops to 50C. At all times, CPUs are at a constant 100% utilization. Now, at first I thought "Okay". But what I noticed is that if I click between 100% and idle, the temp IMMEDIATELY jumps/falls between 50C and 56C. Its like a step function. I believe that the temp could spike to an immediate 56, but also cool off to an immediate 50C? And do this 100 times a minute... as fast as I can click a mouse?

NO WAY. If it took even 5 seconds to cool back down, that would be believable. But instantaneous heat up and cool down by 5C as fast as I can click the mouse? NO WAY.

There is something fishy about TAT. Heat-up and cool-down generally do not exhibit behaviour like step functions.

The second anomaly is that the 100% thermal load produced by TAT is somehow hotter than the 100% load produced by Prime95 by 5C. I simply don't believe it.

Anyone else notice this?

 

[pinoyako]

i experience that too..no way the temps can drop that quick right?

 

[yquo]

I have noted differences using different power schemes like desktop or mobile/notebook computer (sorry don't know how is translated but you get the point)

 

[CompuTronix]

rdhood, nothing "fishy" here. TAT is intended to be run individually, which precludes program conflicts such as this. From the Core 2 Duo Temperature Guide:

Background

Intel has 2 distinct C2D thermal specifications, and provides a test program, Thermal Analysis Tool (TAT), to simulate 100% loads. Some users may not be aware that Prime95, Orthos, Everest and assorted others, may simulate loads which are intermittent, or less than 100%. These are great for stress testing CPU, memory and system stability over time, but aren't ideal for testing the limits of CPU cooling efficiency.

Testing

C1E / EIST = Disabled
CPU Fan / Computer Case Fans = Manual 100%
Load 1 = TAT @ 100% 10 Minutes
Load 2 = Orthos @ P9 Small FFT’s ~ 88% 10 Minutes

TAT will expose insufficient CPU cooling and computer case cooling, or excessive Vcore and overclock. At no other time will a CPU be so heavily loaded, or display higher temperatures, even when highly OC'd during worst-case / real-world loads. After CPU thermal behavior has been benchmarked by TAT, then Orthos or assorted other programs can be used with Core Temp and SpeedFan 4.32, to observe less extreme CPU temps, while stress testing for system stability. Orthos 88% Load will indicate ~ 5c lower than TAT 100% Load. Although Windows Task Manager, under the Performance tab, may show Orthos CPU usage at 100%, this indication is inaccurate.

The same holds true for Prime95. If you run TAT individually as intended, then results will be consistent, and free of code and execution conflicts.

Hope this helps,

Comp 8)

 

[rdhood]

I can accept that. But I still don't accept the instantaneous addition/removal of heat. My HSF just isn't that good!

 

[CompuTronix]

Again, this is typical of program execution conflict, so I have complete confidence in TAT's ability to apply a true 100% Load to the cores, however, the temps may not always be accurate. From the Core 2 Duo Temperature Guide:

Variables

...the dual Tjunction temps can be erroneous due to calibration inaccuracies in chipsets, BIOS releases, driver versions, and motherboard utilities. Intel’s spec for thermal sensor accuracy is +/-1c, so temperatures can be quite accurate on hardware / firmware platforms free of manufacturer's deficiencies. Even false temperatures have offsets which can be analyzed, and when given correction factors, may still be reasonably accurate.

Troubleshooting

(D) Any hardware and / or software may misreport Tcase and / or Tjunction temps.

(G) TAT may not run on Vista. Orthos Priority 9 Small FFT’s simulates 88% load ~ 5c lower.

Offsets

If temperatures don’t meet the Parameters, then SpeedFan 4.32 can be configured to correct for Tcase and Tjunction offsets. From the “Readings” tab, click on the “Configure” button, then click on the “Advanced” tab. Next, click on the “Chip” field directly under the tabs, then use SpeedFan’s “Help and HOW-TO” icon included in the installation Program Group. Under “Contents” click on “How to configure” then click on “How to set Advanced Options”. Read this entire section including “Other interesting options” and “Temperature x offset. When configured, SpeedFan will provide a means to display all 3 Tcase and Tjunction temperatures correctly. SpeedFan is also extremely useful for observing temperatures and voltages using the “Charts” tab, while thermal benchmarking with TAT.

Hope this helps,

Comp 8)

 

[mekender]

well i know with my cooler, my temp will fall from 55 down to 40 in less than 2 seconds after stopping prime 95/tat to me that seemed really fast, but i dunno

 

[CompuTronix]

That's correct. If you use the "Charts" tab in SpeedFan 4.32, you will see how rapidly the thermal plots rise and fall as load rises from Idle to 100% then falls back to Idle. The initial rise and fall is nearly instantaneous, then curves out, which is quite normal.

Hope this helps,

Comp 8)

 

[rdhood]

The initial rise and fall is nearly instantaneous, then curves out, which is quite normal.

Hope this helps,

Comp 8)

....quite normal for TAT, evidently. It's not normal *at all* for my previous real world experience with heatsinks and fans.... and I've watched a helluva lot of CPU temps in my time. Generally, heatup and cool down look more like a sawtooth wave than a step function.

 

[CompuTronix]

As I said, if you observe the "Charts" in SpeedFan, then you can use any program you like, TAT, Orthos, Prime95, etc., and the graphical characteristics and thermal behaviors will be the same. At a sample rate of 3000 miliseconds, it will always appear most similar to a "square wave" representation, with variations in amplitude proportional to load.

Hope this helps,

Comp