Throttling in Core i5 2500

[ashetos]

Hi folks, this is my first thread. This is more about curiosity than a real problem that I'm facing.

I have an intel DH67CL board (no overclocking possible) and an intel Core i5 2500 processor paired with a ARCTIC COOLING FREEZER 7 PRO REV.2.

The thing is, when I run intel burn test (runs linpack iirc), with 4 threads, be it pinned or unpinned, the CPU clock drops to 1600MHz (i see it via CPU-Z) and the reported GFLOPs drop accordingly, after a couple of minutes.

The CPU temperature is the highest in the system and it doesn't exceed 63C. PCH and other component temperatures never exceed 60C. CPU-Z initially reports 3400MHz, which is probably the first turbo bin, after a while it reports 3300MHz and it reports the throttled 1600MHz.

So what is the explanation for this? Could it be that the board is not good enough for this workload somehow? Is it that Core i5 2500 cannot stay below 95W on average when running such a workload?

Discuss!

 

[sm625]

I had a similar problem with a gigabyte mobo, cant find the thread but I did make one. Disable intel motherboard-related software. Look in task manager and find anything motherboard related and kill it.

 

[ashetos]

I just ran linpack in linux and the exact same thing happened. I'm pretty much convinced it's not a software problem. That leaves a firmware or hardware reason for this kind of behaviour, probably...

 

[ShintaiDK]

It shouldnt throttle due to temps. So try reset the BIOS, maybe do a BIOS update if there is any.

 

[ashetos]

Latest BIOS is on. The thing is, BIOS has an average TDP setting of 95, and I cannot change it due to it being H67 or something. This is what I'm most suspicious of...

 

[ShintaiDK]

Latest BIOS is on. The thing is, BIOS has an average TDP setting of 95, and I cannot change it due to it being H67 or something. This is what I'm most suspicious of...

95 is fine. It will not pull that.

The wierd thing is you start out fine for a while, then slowly drops. I wonder if some other components is overheating. Maybe the VRM. Is your case very hot?

Also how much memory do you use in the test?

 

[ashetos]

VRM temp is reported as 50C, both in Windows and in Linux. Case is Coolermaster HAF 922, and it's very cool. I've also just cleaned the dust off. I use 4GB data set, tried maximum also, no difference.

The interesting thing is that for every iteration, the CPU is at 3300MHz for the first half and 1600MHz for the second half. Also, any standard application that causes 100% CPU utilization does not cause throttling, it's only linpack. The floating point units seem to be more taxing to the CPU.

 

[ShintaiDK]

Can you try something else? like Prime95, small FFTs?

I have a feeling nothing is wrong CPU wise.

 

[ashetos]

Just tried prime95 with small FFTs. Behaviour is the same, starts at 3400MHz, quickly drops to 3300MHz, and then alternates between 1600MHz and 3300MHz.

I would expect the drop to 1600MHz to be permanent if this was a thermal issue, but it seems to be periodic. Note that CPU utilization is obviously always 100% during all those runs. Windows power profile is High performance.

 

[ShintaiDK]

What tool do you use to measure the temps?

 

[ashetos]

Intel desktop utilities for Windows and lm_sensors for linux.

Note that with the stock cooler intel burn test hits 100C eventually.

 

[ShintaiDK]

If it throttles due to heat, it actually jumps relatively wildly around in frequency. But when your top temperature is 63C, then you are very far from tjmax.

1600Mhz is the default idle clock as well. So sounds like the CPU changes between active and idle for whatever reason.

Did you try run it under linux? Does it behave the same way there?

You can download it here if needed:
http://software.intel.com/en-us/articles/intel-linpack-benchmark-download-license-agreement

 

[ashetos]

Yes, I posted above that I also ran linpack for linux. Exactly the same behaviour in linux.

It drops to 1600MHz but I wouldn't call it idle because the processors are fully utilized for both Windows and linux (task manager, htop utility).

When the processor throttled due to tjmax (stock cooler, 100C) the clock would also drop to 1600MHz.

In this case though, the cooler is more than enough for this processor.

Just noticed, after all this stress testing, Memory temperature and VR Temperature are stuck (bugged probably) to 128C. All the previous time their temp was normal. Definitely something fishy is going on.

 

[ShintaiDK]

Maybe time to simply RMA the board.

 

[ashetos]

Unfortunately I am absolutely sure the retail shop here in Greece is not going to sympathize with such a description of a defect, too technical for them and not convincing (no applications fail etc.). I am a little disappointed in the board in general. It had the features I liked back in 2011, but the build quality, capacitors etc seem sucky, and now this.

 

[fixbsod]

And yet I recall all the oh noes (not you per se) from the crowd when Intel announced they were pulling out of the mobo business. Pfft. Intel makes CPUs first and mobos wayyyyy in the back and now not at all.

Unfortunately I am absolutely sure the retail shop here in Greece is not going to sympathize with such a description of a defect, too technical for them and not convincing (no applications fail etc.). I am a little disappointed in the board in general. It had the features I liked back in 2011, but the build quality, capacitors etc seem sucky, and now this.

 

[Arkaign]

Your temps are fine. Update to latest BIOS + jumper reset to defaults if you haven't already, and try disabling C1/EIST/whatever power advanced settings are in there that you can find.

 

[ashetos]

Ok, I just disabled turbo, speedstep, C states and the CPU idles at 1600Mhz... Seems like I cannot disable speedstep after all. There are no power options left to tweak in the BIOS. I'm too tired to jumper reset to defaults right now, and I don't hope it corrects this situation. Maybe the board revision being old (does not support ivy bridge) or the CPU being old (early adopter of sandy bridge) has something to do with this.

 

[Magic Carpet]

Change Windows Power Profile to "High Perfomance"? Double check, the minimum processor state is set to 100%.

 

[ashetos]

Posted above that power profile in Windows is indeed high performance. All these also happen in linux (Fedora 18). This doesn't seem to be a software thing.

 

[Piroko]

Honest question, why do people keep trusting those thermal diodes?
Just open your case and feel if anything (especially VRMs) on your Mobo is too hot to touch. Also check for popped capacitors and if the fan in your PSU is running.

 

[ashetos]

Honest question, why do people keep trusting those thermal diodes?
Just open your case and feel if anything (especially VRMs) on your Mobo is too hot to touch. Also check for popped capacitors and if the fan in your PSU is running.

I guess I don't know any better! I'm not really a hardware or hands on guy, as far as equipment is concerned. I'll just leave it at that, I don't feel comfortable doing electrical engineering experiments as I'm no expert. Thank you all for your input.

 

[FalseChristian]

I used to have my i5 2500K at 4.5GHz throttle down to 1.6GHz, too. But since I've upgraded the BIOS on my Asus Z68-V/GEN3 to 3402 my CPU stays at 4.5GHz at all times.