Undervolting 4790k and "uncore" keeping all power saving stuff on

[EliteRetard]

Staying stock 4.4GHz, I've never even seen a UEFI BIOS 'till now, or played with a CPU past 2009...so not entirely sure what's what. I can probably figure out how to undervolt the main core, but there's a million other settings here too. The objective is to reduce heat as much as possible and still have the automatic clock and voltage increase/decrease. I've tried 3 types of coolers and still getting horrid temps (90c+).

 

[Cerb]

What 3 coolers?

Using the latest motherboard firmware?

Does it do in applications you use regularly, or just stress-testing?

What are your under-load voltages?

 

[BSim500]

I can probably figure out how to undervolt the main core, but there's a million other settings here too. The objective is to reduce heat as much as possible and still have the automatic clock and voltage increase/decrease. I've tried 3 types of coolers and still getting horrid temps (90c+).

Negative vcore offset, ie, offset of -0.05v, etc? Do you get those temps in regular usage (gaming, x264, etc) or just under AVX "power virus" stress testing like Prime / IBT / Linpack, etc? The latter tend to be artificially 20c higher than normal due to Haswell's +0.1v auto-overvolt (but only on AVX benchmarks). Try loading all cores with X264 video encode (or stress test with non-AVX Prime v26.6) for 10mins and measure temps then.

 

[EliteRetard]

The stock cooler, which I knew was a joke before I even put it on...it ran ~60c idle and running a game demo it easily hit 100c and I stopped it.

Tried a Scythe Shuriken, idle at 40c but still hit 90c+ running game demo...bought some bolts and such and tried that instead of the crappy plastic clips, no change as far as I can tell.

Tried a big orb cooler and bolted it down, get ~38c idle and just hitting 91c on hottest core running demo. Slightly better than the Scythe, but this is 2x the size and crazy loud. It wont actually fit in my case without modding it, but I wanted to try it anyway.

Stress test with Prime95 blend quickly hits 100c on any cooler even with case open.

Under load I'm getting 1.23 volts. Using AS ceramique in a thin layer, which worked for my other systems (have a 125w PHII staying under 55c stress test)

One thing I'm concerned about is the socket clamp system...it and the MOBO bow significantly once I clamp down the CPU. There's no brace on the top of the MOBO, which seems weird to me.

This is what I used to see:

This MOBO I'm using looks kinda like this:

There's no brace or anything to support the board on top. I don't have a large straight edge and I just cant tell if the socket mechanism is interfering with the coolers, the mechanism seems really close to the height of the IHS...the paste seems to be contacting though.

I really have no idea why I'm having so many issues here. I knew the Intel CPUs were hard to cool but this is insane. I have a much higher power CPU being cooled with a really crappy old tower heatsink (freezer 64). Towers cant really be so vastly superior can they? Is it possible I have a faulty CPU?

I found some weird voltage offset setting or something in the BIOS and I was able to reduce the core voltage, but I think it reduced it across the board and not just at load, I got crashes so I set it back to auto. I was able to see load voltage of ~1.8 though it didn't seem to change temps much.

 

[EliteRetard]

Just as a test I tried it with the fan unplugged...I get similar temps. Mid to upper 30's idle and 90c+ load. I could feel the heatsink getting fairly warm, and when I stopped the test the core temps dropped pretty quick.

I have no idea what this means.

Edit:
Actually only thing I can think is the CPU is producing more heat than the heatsink can handle with or without a fan and the automatic software controlling the CPU is keeping the temps similar.

 

[BSim500]

Just as a test I tried it with the fan unplugged...I get similar temps. Mid to upper 30's idle and 90c+ load. I could feel the heatsink getting fairly warm, and when I stopped the test the core temps dropped pretty quick.

I have no idea what this means.

As mentioned above, try a non AVX version of Prime and post temps. Try also running your heaviest real-world app. If temps are significantly lower the 'problem' is Haswell's internal +0.1v auto-overvolt design "feature" which only kicks in under AVX stress tests (whose only use is to tell you how hot your PC runs under AVX stress testing). If the real-world temps are nearer 70c under most of your heavy games / apps, you have little to worry about from general day to day use.

 

[EliteRetard]

Negative vcore offset, ie, offset of -0.05v, etc? Do you get those temps in regular usage (gaming, x264, etc) or just under AVX "power virus" stress testing like Prime / IBT / Linpack, etc? The latter tend to be artificially 20c higher than normal due to Haswell's +0.1v auto-overvolt (but only on AVX benchmarks). Try loading all cores with X264 video encode (or stress test with non-AVX Prime v26.6) for 10mins and measure temps then.

With that version of prime I get similar temps to the game demo @ ~90C
I didn't know there was a difference between prime versions...
Voltage is running lower too, 1.197 vs 2.3

 

[EliteRetard]

System was sitting idle and just started a virus scan, using ~10% CPU @ ~55c
And the fan is plugged back in.

Wondering if I should try cranking down the bolts more...but there's already a lot of bowing from the CPU clamp. Hmm...

Edit: OK I'm going to sleep...I'll play more in the morning.

 

[james1701]

What case and mobo are you using?

 

[Cerb]

I was able to see load voltage of ~1.8 though it didn't seem to change temps much.

1.8V is what's going into the CPU, not what the cores are running at. The Vcore you're looking for should vary between about 0.9V and 1.2V, depending on load.

Seeing what the voltages are is why I didn't go ahead and mention doing that. Setting the Vcore, with a stiff LLC, might do better, if the problem is high voltage. My CPU runs fine at -0.1V, but that also means -0.1V at idle, and some CPUs may have trouble with that. Knowing what it's actually running at before starting gives a point of comparison, a baseline to start from, and a way to tell if the mobo is screwing things up, along the way (which I think is likely, but might be something that can be worked around). Unfortunately, each CPU sample may exhibit slightly different results.

IMO, the Shuriken is likely the best to keep on there, and it should have no problems keeping it cool. I think you have the clamp on fine (otherwise, the heatsinks would show much different results), but that the CPU is a hot one, and/or the motherboard is telling it to use voltages that are too high (I've seen reports of that with some Asus and Gigabyte Z97 boards).

Speaking of which, what is the mobo?

Something is definitely not right...but what?

 

[escrow4]

A 4.4GHz 4790K isn't exactly stock. Sure it should run at that rated all turbo speed but you are hitting the limit of what you can push Haswell to. This reminds me why I haven't clocked up my 5930K. The volts and heat just go through the roof.

 

[Majic 7]

Something is weird. I just switched from my Corsair H70 to the stock cooler. Idles around 30, new Prime blend runs around 80, topped at 93 on second test of five, small ffts- immediate 100 and throttle. Games at around 40 to 60. This is with everything at stock, no XMP. With XMP Prime blend hovers around 90 to 100 with frequency going from 3700Mhz to 4200Mhz. Still idles 30 to 35 even after shutting down Prime.

 

[PPB]

That is too big a difference to be a normal behaviour. What game was it?

 

[EliteRetard]

*snip*

Speaking of which, what is the mobo?

Something is definitely not right...but what?

Asrock z97e-itx/ac
Intel core i7 4790k
Crucial low profile 2x8GB DDR3 1600 CAS 9
Silverstone Sugo SG05 w/ 450w PSU
EVGA GTX970 (ACX2.1?)

I've also replaced the front fan with a high performance unit and have the filter and grill removed for more airflow. PSU is "upside-down" so it pulls air from the top vent and not the CPU area. Been testing with various game demos like the Metro LL & Mafia II benchmarks at 1920x1080 maxed settings. I've got Unigen and a few others, as well as 2 versions of Prime now that I've been playing around with.

Here's a video showing my BIOS (not sure if it's the same revision):
https://www.youtube.com/watch?v=s41RnQQc6wI

I've disassembled the system again, and I'm going to try re installing the Scythe cooler in a different orientation and using different bolts/washers mounting etc. The stock mounting mechanism is quite poor, very flimsy.

 

[EliteRetard]

OK I just pulled off the orb cooler thing, and I can see it did not mount correctly this time. it only touched a little bit on the sides and maybe a tad in the center. The pattern is very weird, but it's clear 70% of the cooler was not touching the CPU. Since I apply the paste in a thin smooth layer across both the cooler and CPU I can see where the paste is still untouched.

Yet this was cooling better than the Scythe was. I'm still thinking there is something wrong with this CPU clamp and/or mounting system. I did try the stock plastic Intel clips on two of the coolers and had issues with them as well. I gotta see if I can figure this out.

 

[flexy]

ER,

Not sure what you're trying to do but it's relatively pointless. Here is why:

Taking off Vcore FOR UNDER LOAD will likely backfire since you're just asking for instability, assuming that your CPU is at stock and stock voltage. It's not that Intel deliberately over-volts them.

Taking off voltage FOR IDLE is pointless since power-states are active anyway. My Haswell is OC to what I can (about 4.5Ghz and 1.22ish Vcore) - HOWEVER at idle that voltage is reduced and so is the multiplier!

Get HW info and run it and look at the *actual* Vcore which is dynamic, since so called core c-states are active (C6 and C7), look at the ACTUAL Vcore right now on my desktop:

This thing basically doesn't use ANY power There is no point in undervolting.

 

[flexy]

And just to repeat it...those CPUs *under load* at 4.4G whatever DO get hot, it's not a secret. 90C, while not pleasant and cool to look at...is COMMON.

90C while certainly high is still not "dangerous", I think it throttles from 95C and shuts off at 105C. Just accept those things get HOT..and if it blows get a RMA from Intel for their crappy CPUs Alternatively delid it and risk destroying your CPU.

Note: If you use OFFSET or Adaptive...or in general use BIOS settings without really exactly knowing what you do.... it adds more voltage to your CPU under load. So if you want to make sure that a certain V is set and keeps DO NOT USE ADAPTIVE AND DON'T USE OFFSET...you need to use a manual CPU voltage in bios. Start with 1.1 or something and then see how it goes, go gradually up til..maybe 1.25 max on air. (V1.22 for me already goes up to 92C under heavy stress testing like OCCT etc.), so be warned.

 

[EliteRetard]

I was hoping to only reduce power under load, not mess with the idle power or auto clocking etc. But I finally found a straight edge to check this socket with and I was right, the damn clamp and MOBO is warping so bad it was indeed interfering with the CPU cooler. I just removed the clamp and am going to try to gently bend it and hope this will keep it out of the way.

 

[flexy]

Could you let me know what you mean with "brace on top"? Do you have a picture to show a similar cooler config like yours? (board bending due to cooler I have a hard time to imagine.....what backplate is this..how exactly does your mounting mechanism look?)

+++

Edit: While some heavier coolers MIGHT come with a top brace, MANY (including the 212 EVO) don't. Their mounting is actually pretty dinky at first glance, the four round standoffs you screw in and a small backplate.

I actually used the relatively dinky mount of the EVO 212 for my old Thermaltake Ultra 120 Extreme which is WAY heavier than the 212 EVO. However I don't see problems with this and I also checked and the board certainly doesn't bend either. (This would be very weird if not impossible given that there is a backplate which SHOULD make bending impossible....UNLESS you did something essentially wrong...even then I can't see the board bending???)

 

[NTMBK]

For the love of god man, why did you not just buy a low voltage chip? The i7 4790T is a thing that exists, you know.

 

[Cerb]

Setting a max Vcore, and then fiddling with the rest, usually works well enough. In that case, even with a slower fan, case cooling should be no problem. If mounted well, the Scythe should handle it fine, at fully stock settings. All 4.4GHz might be problematic, but typically, a stiff LLC with slightly reduced max voltage can do the job. A quick spike to 90C+ is normal, but then it should even out, as the TDP limit kicks in.

It's not going to be 55C under load, but something definitely is off.

If you have a backplate to mount with, use that, instead of the PCB directly.

 

[EliteRetard]

Could you let me know what you mean with "brace on top"? Do you have a picture to show a similar cooler config like yours? (board bending due to cooler I have a hard time to imagine.....what backplate is this..how exactly does your mounting mechanism look?)

+++

Edit: While some heavier coolers MIGHT come with a top brace, MANY (including the 212 EVO) don't. Their mounting is actually pretty dinky at first glance, the four round standoffs you screw in and a small backplate.

I actually used the relatively dinky mount of the EVO 212 for my old Thermaltake Ultra 120 Extreme which is WAY heavier than the 212 EVO. However I don't see problems with this and I also checked and the board certainly doesn't bend either. (This would be very weird if not impossible given that there is a backplate which SHOULD make bending impossible....UNLESS you did something essentially wrong...even then I can't see the board bending???)

I posted pictures earlier, look at the socket clamp, the old intel ones used to use a full metal brace around the socket to strengthen everything and prevent warping. This MOBO has no such brace and the clamp warps the board when closed on the CPU. In this case I was pretty sure the clamp was interfering with the coolers...but I couldn't quite tell. Well as I said, I found a straight edge and I was able to confirm that indeed the crappy CPU clamp was warping along with the MOBO and was sticking out just a hair above the IHS on one side.

Now that I've mangled the crap out of the stupid clamp it not only warps everything less, but it now sits just a micron or two lower than the IHS. I'm just about to mount up a cooler and see how it compares to before.

None of the coolers I have included any braces or backplates themselves either, so I went and bought my own bolts and washers and such and have been using them to bolt the coolers down. I did try the stupid plastic clips first, but the bolts and washers actually spread the load out a bit more and wont break off or do anything stupid.

I don't see how any backplate would fit though, there is so much junk on the back all over the place in and around the bolt holes and CPU etc.

 

[EliteRetard]

I found an image thats prtty close to what I'm doing, they also have a Scythe cooler but its a bit different than mine. First, the cooler that I'm trying to use (since it fits the case etc):

And then I swapped those stupid clips for bolts kinda like this (you can see how flimsy the clip arms are on mine):

I believe I'm using 8-32 machine bolts, .75" long with stacked neoprene and standard washers.

 

[flexy]

So basically this is a custom/alternative mounting method? (Similar what I did with mine where I use the EVO212 clamp for the Thermalright Ultra Extreme).

The problem here (I THINK) is that it's possible that with this custom mounting method you can easily introduce stress, bending etc. eg. by over-tightening the screws etc...and the lack of a backplate doesn't help either, it would make the problem worse. The problem won't be apparent with the original "plastic clip" thingies because with them you are never able to introduce that big a force that it would bend anything as you can with the bolts.

So...as a first test you can try use the plastic clips again (because less pressure)..to see whether it also stresses the board and see whether it gives better temps.

Or..re-mount as you did before, but do NOT overtighten the screws at all, in fact try a very careful mount where the HSF sits "just right" but not more.

Other alternative...look for an option to mount (or an entire new cooler for that matter) with a backplate as well. EVO 212 is rather cheap. It's not "high end" but at least it could solve the problems with bending.

I would also recommend another method of applying TIM (not sure about Ceramique, anyone and their mama uses MX-4)..and then not spreading but a tiny blob or a tiny strip maybe 3/4 of an inch down the die like this:

***

This MOBO has no such brace and the clamp warps the board when closed on the CPU.

If it's really that the CLAMP from the socket warps the board, something is off with the board. A backplate may also be the solution here, but I wouldn't say it's normal that closing the socket would warp the board, not at all.

 

[escrow4]

The fact this is in a tiny m-ITX case with a tiny m-ITX mobo with way less airflow over the VRMs isn't helping any either. And where is the backplate for the cooler? That fat chunk of metal at the back of the mobo that supports the weight?