5820k Prime95 CPU throttling?

GotNoRice

GotNoRice

[H]F Junkie
Joined
Jul 11, 2001
Messages
12,003
Currently I'm running my 5820k @ 4.5Ghz using 1.35v with an H100i GTX and 4 fans in push-pull. That is 125 BCLK and a multiplier of 36. I'm using the 125 BCLK mainly so my memory can run at it's XMP speed of 3000Mhz.

I was running Prime95 and had CPU-Z open on a 2nd monitor. I noticed that it was running at 4250Mhz, at a multiplier of 34 (though still showing a range of 12-36). Temps never broke ~65C on even a single core while this was happening,

I quickly tried several games, and during all of them the CPU clocks up to 4500Mhz as it should. I tried using several versions of 3dmark, running the CPU tests, all of which are able to peg all 12 logical cores to 100%. The CPU bumped up to 4500Mhz just fine with those tests also.

I went back to Prime95 to do some testing, and this is what I found:

Using 1 Prime95 thread, the CPU clocks up to 36 (4500Mhz)
Using 2 Prime95 threads, the CPU clocks up to 35 or 36 (4375-4500Mhz)
Using 3 Prime95 threads, the CPU clocks up to 35 (4375Mhz)
Using 4 Prime95 threads (or more), the CPU clocks up to 34 (4250Mhz)

This seems to only be happening in Prime95. No core temps ever went above 65C during this testing and they were mostly much lower. Any ideas why it seems to be throttling in Prime95, since it doesn't seem to be temperature based? I recall something about Xeons throttling when AVX was used, and I think this version of Prime95 uses AVX. Is that relevant to what is going on?
 
Intel Turbo Boost uses a short term and a long term power limit. If you are running a heavy duty app like Prime95, it is likely that your CPU is hitting the long term turbo power limit. The CPU will automatically reduce the multiplier just enough which forces the CPU to stay just under this power limit. The multiplier is being finely tuned like this, hundreds of times a second.

Have a look in the bios for the turbo power limits and start increasing them.

Try running RealTemp GT. It accurately calculates the average multiplier when turbo throttling is going on and it will report the average multiplier to one decimal place so it is easier to see when you are throttling and by how much.
 
GotNoRice said:
Currently I'm running my 5820k @ 4.5Ghz using 1.35v with an H100i GTX and 4 fans in push-pull. That is 125 BCLK and a multiplier of 36. I'm using the 125 BCLK mainly so my memory can run at it's XMP speed of 3000Mhz.

I was running Prime95 and had CPU-Z open on a 2nd monitor. I noticed that it was running at 4250Mhz, at a multiplier of 34 (though still showing a range of 12-36). Temps never broke ~65C on even a single core while this was happening,

I quickly tried several games, and during all of them the CPU clocks up to 4500Mhz as it should. I tried using several versions of 3dmark, running the CPU tests, all of which are able to peg all 12 logical cores to 100%. The CPU bumped up to 4500Mhz just fine with those tests also.

I went back to Prime95 to do some testing, and this is what I found:

Using 1 Prime95 thread, the CPU clocks up to 36 (4500Mhz)
Using 2 Prime95 threads, the CPU clocks up to 35 or 36 (4375-4500Mhz)
Using 3 Prime95 threads, the CPU clocks up to 35 (4375Mhz)
Using 4 Prime95 threads (or more), the CPU clocks up to 34 (4250Mhz)

This seems to only be happening in Prime95. No core temps ever went above 65C during this testing and they were mostly much lower. Any ideas why it seems to be throttling in Prime95, since it doesn't seem to be temperature based? I recall something about Xeons throttling when AVX was used, and I think this version of Prime95 uses AVX. Is that relevant to what is going on?
Click to expand...

65c @ 1.35c with a modest aio is a wrong temperature reading.
Nothing more.
 
o51Y2gO.png


There is a program called Limit Reasons that explains in real time why a 4th Gen Haswell CPU is throttling. I am not sure if the 6 core CPUs are supported but it might tell you what the real reason is.

PL1 and PL2 are the power limits that I was talking about. You can see in the pic that my laptop CPU is throttling because it has exceeded the long term (PL1) power limit. This program was written by Dufus on NBR and is included in the ThrottleStop download.

ThrottleStop 8.00 beta 3
https://www.sendspace.com/file/tmc96z

Intel individually calibrates their core temperature sensors on each core of each CPU that they ship. It would be highly unlikely that they are all wrong. Limit Reasons also shows when the thermal throttling temperature has been reached but that is obviously not your problem.

Edit - EDP (Electrical Design Point) also known as the current limit.
 
sblantipodi said:
Your readings are surely wrong
Modest aio + 1.35v + prime95 can't be 65c.
Click to expand...

Interesting that you feel you can make that determination, absent of any context, given that nowhere in this thread did I mention anything about my ambient temperature, how long I was running Prime95, which Prime95 test I was running, etc. Probably because this isn't a thread about temperatures. The CPU begins clocking down in Prime95 immediately upon starting the test, and the only reason I posted my temperature was to show that it obviously had not hit Tj Max.

Oddly enough, after rebooting, it no longer seems to be throttling CPU clocks in Prime95. Nothing changed other than the reboot, so I'm not sure why that would be the case.
 
GotNoRice said:
Oddly enough, after rebooting, it no longer seems to be throttling CPU clocks in Prime95. Nothing changed other than the reboot, so I'm not sure why that would be the case.
Click to expand...

If you're hitting the power maximum of the VRMs, you could see behavior like this. If you hit the power maximum duty cycle, then it will tell your CPU to throttle even if you're well below max temperature.

I had a similar problem on my Core i5 2500k and a cheap P67 motherboard - after 15 minutes of Prime 95 the crappy VRM would report over duty cycle, and throttle the processor. Running ThrottleStop prevented this from happening.
 
ThrottleStop lets you see what your turbo power limits are set to.

http://i.imgur.com/KCVfKo5.png

By default, Intel recommends that the long term limit be set equal to the rated TDP of the CPU and the short term limit should be set 20% to 25% higher than the long term limit. My 4th Gen 4700MQ has a TDP of 47 Watts so the long term limit is set to that and the short term limit is set to 58.75 Watts (47 Watts + 25%). If there is an issue with your bios, this might be set to one value when you boot up and these limits might change after you do a stand by / resume cycle.

Try running CPU-Z, go to the About tab, click on the Save Report (.TXT) button and upload that info somewhere convenient like www.pastebin.com and post a link here. I will show you how to interpret the info in that file so you can find out what your turbo limits are and you can check to see if they are changing after standing by.
 
You must log in or register to reply here.
Back
Top