Suspicious Readings from Task Manager While Running Prime95 on an Intel Core i7-870

the_ultra_code

Commendable
Apr 6, 2017
52
0
1,630
Hello Tom's Hardware members.

I am having quite a time overclocking my retro/modern gaming PC's CPU (which is also my main PC, 'cause it is the most powerful PC I own), and I was wondering if you could help me out.

As courtesy, here's a quick list of it's specs:

  • ■ Intel Core i7-870, cooled by an Artic Freezer 33 eSports edition cooler (the one with two fans on it)
    ■ Asus P7H55-M Micro ATX motherboard
    ■ 8 GB G.Skills Sniper 2133MHz DDR3 RAM
    ■ EVGA Nvidia GeForce GTX 780 Ti
    ■ Sound Blaster X-Fi PCIe soundcard
    ■ Corsair TX650M 650W 80+ Gold PSU
    ■ Windows 10 Pro

Okay, so recently I upgraded the i5-750 in the system to an i7-870, and so, wanting to squeeze a bit more juice out of the system, I decided to overclock it to 3.6GHz, the "max" turbo frequency of the CPU, as stated on Intel's website. However, if I remember correctly, I noticed some strange things happening in Prime95. I decided then to bump down to 3.2GHz, the highest speed that I have ever seen this CPU turbo up to on default BIOS settings. Again, the same result. And today, just now, I decided to turn off load-line calibration to see what would happen, and again, although better this time, the same thing happens.

What I am talking about are "random" drops in CPU usage while running Prime95 using the "Small FFTs" torture test. The torture test chugs along just fine, when all of a sudden you can see in Task Manager CPU utilization drop from 100% to ~40% for like 3-5 seconds, then going back up to 100%. The higher the overclock, and (it seems) the more voltage that is being feed to the CPU, the sooner this event happens and the quicker it happens again in the torture test.

Now, let me preface by saying I am a completely new novice at CPU overclocking (the only other overclocking experience I have is in overclocking a Voodoo3 3000 in my Pentium 3 machine, and that's it). My current settings in the BIOS are to enable "XMP" in "High Frequency" mode, so that the custom timings for the RAM are given to the motherboard, then with the multiplier set to 24, BCLK to 133, QPI the lowest frequency available, all voltages set to auto (except RAM, with the XMP profile setting it to 1.65V), and all power settings (such as C-state stuff) disabled. It seems that any overclock up to 3.6GHz is stable, from what I can tell, but that Prime95 somehow brings the system down in such a way as to reduce CPU usage during the torture test to half of what it should be.

Now, from what I have read, these events point to poor VRM cooling, bad VRMs (at least quality-wise), or (and this is me guessing) just not "strong" and "robust" enough of a VRM solution. I did notice that, after replacing the dusty old thermal pad with new cheap Chinese ones, and then replacing those with Arctic's thermal pads, which are 4x more thermally conductive at 6.0W/mK, that the time between me starting Prime95's torture test and the first occurrence of CPU usage dropping increased, so this has me to believe that the motherboard's VRMs are somewhat responsible for this "instability".

So, what I want to know is this: Is it indeed a VRM problem, some other problem or combination of problems entirely, or something that really means nothing? I doubt it means nothing, because I can't ignore the feeling that something's not right.

Any help would be appreciated. Thanks!
 
Solution


A fan pushing air out through the top grill would seem to be what is needed; does it have to go on the cpu header? Dont you have a spare...
It's likely that you are seeing thermal throttling. Can you provide the temperature readings of the CPU when you stress it?

The reason why I think its Thermal throttling is because CPU's can slow down when it hits a certain temperature, then speed up again once it is cool enough. If you can provide the temperature readings of the CPU while you stress it, this could confirm if this is the case.
 
I seriously doubt that is the case. See for yourself.

TvnLJcy.png


4hLahJs.png
 
I agree with the above.

Also, the temps reported by the motherboard sensors are an average, individual cores can be running much hotter (download and run Speedfan or CPUID HWMonitor to check core temps).

Boost speeds are exactly that, speeds the cpu can run for SHORT periods of time; trying to run them 24/7 will cause overheating unless you improve the heatsinking capabilities.
 
Thank you CaptainCretin for the suggestion of HWMonitor. I'll check it out.

However, I doubt my heatsink capabilities are lacking. The heatsink is recommended up to 210W, and supports a max TDP of 320W. Would a small boost clock of say around 200-300 Mhz from the CPU's base clock of 2.93GHz throw the CPU's TDP so high above the i7-870's listed TDP of 95W as to throttle the CPU in any way? I mean, the CPU doesn't even thermal throttle in Prime95, and its Tj-Max is 40 degrees C! I hardly think that my CPU is melting. :)
 


OK, it could be a "not spot" in the thermal pad/paste causing a single core to overheat, hopefully HWM will show you what is happening temp-wise.
 
Okay, an update.

I decided to look around in the BIOS, and I tried disabling the "CPU TM Function." It turns out, by doing this, CPU usage stays a rock-solid 100% @ 3.2GHz in Prime95's Small FFTs torture test during the 32 minutes I ran the test. Okay, that's something.

To see if it was a VRM issue, I decided to point my infrared thermometer gun at both the large passive VRM heatsink right "behind" the CPU, next to the rear case fan and motherboard I/O, and at the top section of the motherboard, "above" the CPU, between the RAM slots and the CPU PSU plug. This is where things get interesting.

I decided to "overclock" the CPU to three different frequencies: @ 2.93 GHz ("stock"; 24 mult, 123 BCLK); @ 3.2GHz (24 mult; 133 BCLK); and @ 3.6GHz (24 mult; 150 BCLK).

Now, across all three overclocks, the VRM heatsink stayed around 40 degrees C. Not bad.

However, the temperature for the top section of the motherboard greatly varied between clocks. At "stock", temps increased (aiming from the RAM slots to the CPU PSU plug) from around 30 C to nearly 70 C; at 3.2GHz, temps increased from 30 C to 80 C; and at 3.6GHz, temps increased from 40 C to 90 C! The worst part is is that at 3.6GHz, after a few minutes of Prime95, the PC shut itself down probably due to the temperature of motherboard components. Ouch.

So, in conclusion, the reason why the CPU throttles is not because of the CPU or the VRMs (at least the ones covered by the large passive heatsink), but because of the rest of the motherboard components' temperatures, if I am correct. I would like someone to confirm this theory of mine.

However, this information raises for me a few questions: Where does this motherboard get these temperatures from, and why doesn't the motherboard show such sensors to the OS? If it is not getting it from sensors, then where is it getting the idea or what data is it calculating to determine to throttle the CPU? And, if normally, with "CPU TM Function" enabled, the CPU would be throttled at a given clock, would it not be wise to run the CPU at higher clocks (say, in this case, at 3.2GHz) for a long period of time, without risking motherboard or other component damage?

If someone can both confirm my theory and answer my questions, that would be much appreciated. Thanks!
 


OK, I heard there were some stability issues with HWinfo; certainly I had issues with it crashing last year.

As for the temp readings, is there an airflow issue in that area? Sometimes two fans cancel each other out and create a dead spot where the air just gets hotter and hotter.

The cooler you have points left/right, so the space above it might not be getting any airflow at all.
 
Well, I'm using a Fractal Design Focus G Mini as my PC case, so the top of the top-most front fan is where the CPU cooler's fans' centers are, so in short, no, no "direct" airflow is reaching the top portion of the motherboard. Maybe if I were to install another fan right above the motherboard would help (it kind of looks like I could), although that would mean I would have to add that fan to the CPU cooler's fans' splitter, resulting in three fans being powered off one fan header. Probably not a good idea.

As for the questions I just asked, do you have any answers you could provide, perchance?
 


A fan pushing air out through the top grill would seem to be what is needed; does it have to go on the cpu header? Dont you have a spare chassis header or can you not run a power adapter cable.

just about any 120mm fan should do it, as long as it isnt too noisy.
 
Solution