Haswell i5-4670K Temps & Cooling

[Abakus]

Hello,

My i5-4670k read off the temps in BIOS and testing software to around 60c, going up to 71c. This was on stock cooler and stock compound. So I decided to use another after market heatsink/cooler (Rosewill RCX-Z90-AL 92mm which is apparently 3000 RPM) as I heard almost all after market heatsinks are better than stock. I used it with Arctic Silver 5 compound. The bios read 45-50c with software saying about 38-47, with a peak of 52c. This is all on idle. Here is an image of CoreTemp, for example. Temps keep rising and dropping really fast so it was hard to get a good picture.

Real Temp reports around 30-38c.

I tested the thermal sensors through software and they didn't give back any errors/problems, but I only did test it through one software.

I understand the Haswell line runs hotter than Ivy and can stand up to 100c before becoming a problem, but the temps I have now just doesn't seem right on idle. But I'm not very knowledgable on CPUs (coming off a '09 HP AMD build). I have googled this to no end, but most people seem to have better temps than I do on idle/stock so that's what finally led me to ask on a forum.

Is the Arctic Silver 5 a bad choice for compound and if not, what is the "best" compound and application method available (I do pea size method)? And should I look to invest something like water cooling for my computer since the Haswells get so hot? And if so, would this be a good choice?

Hydro Series H90 140mm

Or should I just go ahead and put in extra money and get the H100i? I really want to keep my CPU as "cold" as I can; it is my first computer build and would like to get as much life out of it as possible.

Any and all help would be greatly appreciated. Thank you!

PC Specs:

Mobo; MSI Z87-G55
RAM: Corsair 8GB
PSU: 600M Corsair
(No GPU)

 

[Hail The Brain Slug]

Was the 60C, up to 71C from the stock cooler at idle or under load? This seems extreme, but there's a few things I can think of. Is your processor lowering the multiplier and the voltage at idle to save power and reduce heat, or is it always at 3.8 GHz and at 1.16 volts? You probably won't get real-time information like that out of coretemp, you will need to get CPU-z or the Intel Extreme Tuning utility.

For reference, I just upgraded to a 4770k, and with the stock cooler I idle around 28-32C. My processor clocks back to 800 MHz at idle, and the voltage drops to around 0.75 volts or so. Under load, the cooler is trash, but the idle temperature is quite adequate.

I'm waiting on an H90 to show up. It's a little disappointing that I can't even run Prime95 on the stock cooler without hitting 100C+.


Regarding the cooler: I chose to get the H90 over the H100i or H110 because it's cheaper (by about $30), performs within 5 degrees of them in the benchmarks I've seen, and because I want to mount it to my rear exhaust and use the top as a cold air intake. In my opinion, I don't see any reason to get the H100i or the H110 unless you're going to overclock significantly.

 

[Abakus]

Hello! And thanks for replying.

Yes, that was with stock cooler and at idle—absolutely nothing done or run except the temp checking softwares. I have no idea if it is doing any of that. Is this something I can check via the BIOS or a quick look into some software? But as far as I'm aware, it is always at 3.8 GHz and I never saw the voltage go higher or lower than that via the software checks.

Would running CPU-z or the Intel utility be safe under my current temps? If it is, I'll run them and then give some screen shots.

As for the coolers; I'd like to overclock in the future (not sure by how much but probably only a little; I'm a simple demands person lol) as soon as I figure everything out. I didn't know there was only a 5 degree difference between the H90 and H100i, that makes a pretty big difference in picking one. However, I'd like to mount on the top of the case as I plan on getting the Corsair Carbide Series 400R or any other similar case. :hmm:

P.S

The only time I stress tested it was with aftermarket and it jumped to 94c really fast and then took forever cooling down. I decided not to risk anything and turned the computer off until I could sort the problem out.

 

[BonzaiDuck]

. . . . . . .
. . . Regarding the cooler: I chose to get the H90 over the H100i or H110 because it's cheaper (by about $30), performs within 5 degrees of them in the benchmarks I've seen, and because I want to mount it to my rear exhaust and use the top as a cold air intake. In my opinion, I don't see any reason to get the H100i or the H110 unless you're going to overclock significantly.

We have about three or four threads which touch on this topic in the "Cases & Cooling" forum. I posted a link to a Frosty-Tech's review that may have included the CoolerMaster Nepton 280L. There's something like 50 coolers with measured results.

Compare the H90 with the Noctua NH-U14S. Noctua is going to release a "double" U14S similar in design to the now-dated NH-D14.

Some of the other threads are useful exchanges about "custom-water" vs "AiO water cooling" vs heatpipe coolers.

Since the difference between top-end AiO's and top-end heatpipes is at most 5C to 7C degrees, the air-cooler can close the gap with lapping, nano-diamond paste or Coollaboratory Liquid Ultra or Pro. The CLU has about a 1C degree advantage over diamond paste. It also can get dropped on the motherboard, where it congeals into little metal balls like liquid mercury -- contaminating traces and components with conductive material. Diamond is non-conductive.

Finally, the ratings for some top-end heatpipes like the D14 are lower than they should be if you replace the two limp Noctua fans with a single 100+CFM 140x25mm PWM fan. Some such fans do not appreciably increase noise, and can be thermally controlled. This is worth another 5C degrees.

The pumps on the AiO cooler might not have the quality of custom pumps -- some more expensive than an AiO cooler kit. There have been announcements here and there that Anand members' pumps had failed. So there's that risk, as opposed to a more expensive custom setup.
Your choice.

 

[ShintaiDK]

Idle temps are normal. Also remember the temeprature accuracy is low at low temps. And perfect near tjmax max (100C).

 

[coffeejunkee]

Go into your bios and make sure EIST, C1E and C3/C6/C7 are enabled. Don't leave them on auto, set them to enabled.

Temps in bios are not really relevant, all power saving options are inactive while in bios.

 

[Abakus]

Okay, I rechecked my BIOS' readings for the temps; it starts at 27c and then rapidly builds up to 47c, with a jump to 57c.

@ShintaiDK

They are? Okay, should I worry though when it reaches 68 on idle (twenty minute wait) or 75 when browsing the internet?

@coffeejunkee

I will check out the options. I think I might have seen them set to disabled. :hmm:

EDIT:

I went into the BIOS to enable the above listed things. Which was: EIST, C1E. For the "C3/C6/C7" I'm assuming I have to pick one of them, so I decided on C3. I looked into the BIOS manual and it said it ranged from low to high, so I guessed low was the best option and picked C3.

Okay, CPU-Z...I couldn't find the temps so I saved the report and post it here at pastebin. I took a new screenshot of CoreTemp and RealTemp as well (on RT the third core would jump to about 40 at its highest peak) after I made the BIOS changes. As for the Intel Utility, this is what it reports;

http://s18.postimg.org/n88ct6irt/Capture.png
http://s18.postimg.org/4rdy2d2tl/Caputer_2.png

 

[tren001]

Idle temps are normal. Also remember the temeprature accuracy is low at low temps. And perfect near tjmax max (100C).

I don't think the temperatures he is getting is normal, especially with a Haswell CPU.

I just built a rig with the exact same i5-4670k Haswell CPU with a stock cooler, no overclocking, and my idle temps are between 15-25 degrees Celcius. I have an aluminum mid-tower Lian-Li case with low to medium amount of fan cooling, along with a relatively warm 7870 Gfx card.

I haven't messed with the fan controls, but it appears to operate at 25-30% of max speed.

Right now I just finished running a 15 minute burn-in with Furmark OpenGL, where the GPU is under full load and 1 out of 4 CPU cores is at 100% load.

Max temperature for that core hit 38 C over the 15 minute run
Max temp for the other 3 cores (under idle basically at 5-25% activity) was in the low 30s C

At the conlucsion of the test, it took exactly 60 seconds for the core temps to fall to their normal 20 degrees C.

I would think this is what you should expect a Haswell GPU core temp to be like?

 

[BonzaiDuck]

I don't think the temperatures he is getting is normal, especially with a Haswell CPU.

I just built a rig with the exact same i5-4670k Haswell CPU with a stock cooler, no overclocking, and my idle temps are between 15-25 degrees Celcius. I have an aluminum mid-tower Lian-Li case with low to medium amount of fan cooling, along with a relatively warm 7870 Gfx card.

I haven't messed with the fan controls, but it appears to operate at 25-30% of max speed.

Right now I just finished running a 15 minute burn-in with Furmark OpenGL, where the GPU is under full load and 1 out of 4 CPU cores is at 100% load.

Max temperature for that core hit 38 C over the 15 minute run
Max temp for the other 3 cores (under idle basically at 5-25% activity) was in the low 30s C

At the conlucsion of the test, it took exactly 60 seconds for the core temps to fall to their normal 20 degrees C.

I would think this is what you should expect a Haswell GPU core temp to be like?

Load testing means loading up all the cores to 100%. The last three processor generations all had a throttling limit around 100C in their specs -- including Haswell -- I think. The Ivy Bridge E cores were made to throttle at about 90C.

I recall here from last year that people were pushing their IB chips in some cases close to or above 90C. That's why the de-lidding obsession: you could see that drop to 80C or below.

 

[cytoSiN]

I would make sure that your heatsink is properly seated and tightened as per the instructions. 30s are fine, but 40-50c at idle is high for Haswell with your setup. Assuming everything else is at stock settings (voltages, etc.), then I would get Prime95 or Intel Burn Test (something that runs ALL cores at once) and run it for at least an hour with both RealTemp and CoreTemp running, along with CPU-Z. If temps get out of hand you'll blue screen long before you damage the CPU at stock settings. In my opinion/experience, 30-35c idle is perfectly fine, and 70 (even up to 80) would be fine under heavy load because you'll never push things that hard in real world settings. If temps get too high and you blue screen, then I suggest you remove the heatsink, clean off the thermal compound, and reapply/reinstall very carefully.

If your temps stay reasonable but you still blue screen, then it's likely a memory issue that we can address separately.

 

[coffeejunkee]

I went into the BIOS to enable the above listed things. Which was: EIST, C1E. For the "C3/C6/C7" I'm assuming I have to pick one of them, so I decided on C3. I looked into the BIOS manual and it said it ranged from low to high, so I guessed low was the best option and picked C3.

I meant you have to enable all of them. C-states are like sleep states for cpu, the higher the number the more parts of the cpu are clocked down or disabled.

There's no disadvantage in enabling them, when required the cpu wakes up in a fraction of a second. They're are also required for the cpu's turbo boost function to work properly.

 

[ratjacket]

Temps keep rising and dropping really fast so it was hard to get a good picture.

that is not idle look at the load % and its not at stock its running under turbo, 3.4GHz is stock with out turbo

"I meant you have to enable all of them. C-states are like sleep states for cpu, the higher the number the more parts of the cpu are clocked down or disabled."

what he said

it can go down to 800Mhz at idle with c states enabled

cpu-z will show you the multi range and details dynamicly

 

[Abakus]

I would make sure that your heatsink is properly seated and tightened as per the instructions. 30s are fine, but 40-50c at idle is high for Haswell with your setup. Assuming everything else is at stock settings (voltages, etc.), then I would get Prime95 or Intel Burn Test (something that runs ALL cores at once) and run it for at least an hour with both RealTemp and CoreTemp running, along with CPU-Z. If temps get out of hand you'll blue screen long before you damage the CPU at stock settings. In my opinion/experience, 30-35c idle is perfectly fine, and 70 (even up to 80) would be fine under heavy load because you'll never push things that hard in real world settings. If temps get too high and you blue screen, then I suggest you remove the heatsink, clean off the thermal compound, and reapply/reinstall very carefully.

If your temps stay reasonable but you still blue screen, then it's likely a memory issue that we can address separately.

I reseated twice, with the application method of the pea size compound at the center. I'm pretty sure the heatsink is seated properly; it doesn't wiggle or anything. I even did the "cross" pressure method when installing it.

I have run Prime95. It was the result that got really hot and then took forever to cool down. I didn't get any blue screens during the test. Temps just got to the 90s and then took forever to drop to 40c.

I'll try it again though and report back.

I meant you have to enable all of them. C-states are like sleep states for cpu, the higher the number the more parts of the cpu are clocked down or disabled.

There's no disadvantage in enabling them, when required the cpu wakes up in a fraction of a second. They're are also required for the cpu's turbo boost function to work properly.

Oh, I couldn't enable all of them. In "Package C State" I have a drop down box to select which one to enable. I'll take a better look at it.

that is not idle look at the load % and its not at stock its running under turbo, 3.4GHz is stock with out turbo

"I meant you have to enable all of them. C-states are like sleep states for cpu, the higher the number the more parts of the cpu are clocked down or disabled."

what he said

it can go down to 800Mhz at idle with c states enabled

cpu-z will show you the multi range and details dynamicly

I never activated anything called Turbo, so I'm not sure how it happened. Been like this since I hooked it all up straight out of the package. Should I disable it to lower my temperature?

Alright, I'll go back into the BIOS and check out the C states again.

edit

Okay, maybe I misunderstood, but I cannot enable all of the options for C State, only one. I turned it to C7 and the Mhz dropped to 799.99 and would randomly jump to 2000 for a while, with a rare peak of 3000. Intel's Ulitity reports 34 at its lowest, a steady 45 and a jump to 54C. Turbo has been disabled.

 

[ratjacket]

800Mhz sounds like its working

turbo will enable depending on load if all c states are set correctly

turbo normally will have two levels one for single core loaded and one for all cores which is slight less GHz

if the c states are not set this can malfunction

check you manual in the bios section, it may spell it out

prime 95 should generate high temps but 90 seems quite high (at stock)with an after market HSF

however the slow droping temps is a worry
it could be 1 of two things

your HSF installation isnt tight enough or mounted incorrectly

or the IHS has poor contact which doesnt allow heat to dissapate

After load is reduced your temps should drop rapidly

unless you have some kind of weird fan profile set it to 100% all of the time for fan speeds

The jumping up in down in MHz from 800 to 2000 indicates the computer is trying to do something wether it be background task or running programes etc

CPU-z is great for seeing this, kill all running programes etc, load it up and wait after awhile it should sit idle at 800 until you move the mouse or do something

with turbo boost when you run prime if it goes to 3800 MHz with all cores active its not fucntioning correctly it will boost past stock of 3400 but it shouldnt hit turbo max unless you are only using 1 core

 

[ratjacket]

found some setting for power states and turbo in your manual

Fanc control can be set in the harware montior section of your BIOS

Intel C-State [Auto]
C-state is a processor power management technology defined by ACPI.
[Auto] This setting will be configured automatically by BIOS.
[Enabled] Detects the idle state of system and reduce CPU power consumption
accordingly.
[Disabled] Disable this function.​

C1E Support [Disabled]
Enables or disables the C1E function for power-saving in halt state. This item
appears when "Intel C-State" is enabled.
[Enabled] Enables C1E function to reduce the CPU frequency and voltage for
power-saving in halt state.
[Disabled] Disables this function.​

Package C State limit [Auto]
This item allows you to select a CPU C-state mode for power-saving when system
is idle. This item appears when "Intel C-State" is enabled.
[Auto] This setting will be configured automatically by BIOS.
[C0~C7s] The power-saving level from high to low is C7s, C7, C6, C3, C2,
then C0.​

[No limit] No c-state limit for CPU.

EIST [Enabled]
Enables or disables the Enhanced Intel® SpeedStep Technology.
[Enabled] Enables the EIST to adjust CPU voltage and core frequency
dynamically. It can decrease average power consumption and average heat production.
[Disabled] Disables EIST.


Intel Turbo Boost [Enabled]
Enables or disables the Intel® Turbo Boost. This item appears when the installed CPU supports this function.
[Enabled] Enables this function to boost CPU performance automatically above rated specifications when system request the highest performance state.
[Disabled] Disables this function.

Enhanced Turbo [Auto]
Enables or disables Enhanced Turbo function for all CPU cores to boost CPU performance.
[Auto] This setting will be configured automatically by BIOS.
[Enabled] All CPU cores would be increased to maximum turbo ratio.
[Disabled] Disables this function.​

​

​

​

​

 

[Abakus]

I just checked my heatsink setting again. It doesn't wiggle at all. Is there a way I can check to see if it is set properly? I tried Prime95 again and let it test until it reached 94c and turned yellow, then I stopped it. It was about 40 minutes I do believe. The good news is that it cooled down to 29-38 in twenty five seconds! I took some screenshots of the process in case it was needed, but only when the temps got higher.

Turbo disabled; fresh start after cool down & Temps read by multiple softwares a minute or two after the computer started up. (killed all programs in the background, which was only two; Live Updater and Sound Blaster Cinema)

And here are the readings through the Prime95 test:
1
2
3
4
5
6
7 (final; cooldown. Took 25 seconds. It actually dropped to 29c a few seconds after I took the screenshot.)

As for the BIOS settings, I took some pictures of how I have it set up as well as the content inside the Hardware Monitor tab.

http://s29.postimg.org/sme77zjg7/WP_20140130_005.jpg

http://s29.postimg.org/l3v1zcs3b/WP_20140130_001.jpg (was just hovering over that option; I didn't mess with it)

http://s29.postimg.org/bkld5w4l3/WP_20140130_004.jpg (C State Options)

Hardware Monitor: http://s3.postimg.org/6rhiwvmz7/WP_20140130_006.jpg

I checked my manual for the BIOS and it does say what the options are, and from what I can see, the C State is enabled correctly.

 

[ratjacket]

all looks good, if you are still worried crank your fan profile to 100% 100% of the time


here's a pic of how long my 4770k takes to cool down after prime load, Large FFT (over clocked to 4.4GHz) from 2mins, I had a LoL spectate game going and vlc running but that shouldn't effect it to much

I have a lot of fans though but if case flow is good it should cool down fast

 

[Abakus]

Thank you! That is good to know, no more stress for me! Thought I might have gotten a bad CPU. D: So I can safely surf the web, listen to music or watch movies with the temps until I get a cooler, right?

And to set the fans always at the 100% speed, I just raise the "minimum" fan speed up to the highest it goes via the BIOS or some other way?

Thanks for all the help guys. I definitely wouldn't have gotten this kind of support calling into some computer tech site or local shop. :thumbsup:


P.S
I have a pretty old case, but it is pretty big and one extra fan. I have it set up on the table—not on the floor where dust/dirt collects—in a rather cold room. So I hope that helps the situation a bit.

 

[ratjacket]

BIOS

 

[VirtualLarry]

I meant you have to enable all of them. C-states are like sleep states for cpu, the higher the number the more parts of the cpu are clocked down or disabled.

There's no disadvantage in enabling them, when required the cpu wakes up in a fraction of a second. They're are also required for the cpu's turbo boost function to work properly.

Not entirely true. There are documented (benchmarkable) differences in SSD performance (lower) with C-states enabled.

 

[coffeejunkee]

Oh, I couldn't enable all of them. In "Package C State" I have a drop down box to select which one to enable. I'll take a better look at it.

Ah, my Asus board lets me enable or disable them one by one but MSI uses a different method. Just choose C7 from the list, meaning it will go all the way down to that state.

To really run your cpu at stock settings you need to set the items from the list in post 15 to:

Intel C-state: enabled
C1E support: enabled
Package C state limit: C7
EIST: enabled
Turbo: enabled
Enhanced Turbo: disabled

It will still get hot under Prime 95 but that's kinda normal for Haswell. Real-world programs/games don't get close to those kind of temps anyway.

Not entirely true. There are documented (benchmarkable) differences in SSD performance (lower) with C-states enabled.

I knew that but I noticed no difference myself. I think it depends on the particular model of ssd as well. Even so, it's trading cpu perf for ssd perf. Also, the question is if you will even notice these benchmark differences in real-world usage.

 

[tweakboy]

Those are fine temps. Start to worry when it riches 90c and beyond. gl

 

[Abakus]

Okay, I turned my Rosewill's speed up (max made it sound like an airplane) to near max. I figured someone would be interested in the outcome whether another person like me needed to know or just out of curiosity, so I decided to track it and post it. So I then logged into windows and did an hour idle test. These are the results: http://s2.postimg.org/6o1zgw8zd/fan_change.png What a change!

And this was with light-med surfing, 1Tab-6Tabs open at once, two programs in the background (Live Updated and Sound Blast Cinema) with Spotify and VLC media player. I also had the Intel Ulitity open to check temps (still had Core Temp/RealTemp up too) while also running the first benchmark (stayed at 65c through the whole thing and dropped down to 20c in a few seconds).

After about forty minutes later of this usage, I was still in this temp zone; http://s8.postimg.org/kndy1c5c5/temps2.png

I can't believe the change from when I first made this thread. I'm impressed and now I know my mind will be blown when I get air cooling!

Specs:
Mobo; MSI Z87-G55
CPU; I5-4670K Haswell@Stock, no Turbo Boost. EIST, C1E & C State enabled.
PSU; Corsair CX600M
RAM; Corsair Vengeance RAM, 4GBx2.
Extra; Rosewill RCX-Z90-AL 92mm Sleeve CPU Cooler & stock LED fan pushing out that came with case. Nothing else but cable management and open case. Arctic Silver 5@ Pea size application method.

@coffeejunkee

Okay, I doubly made sure to check the settings. All seems to be set as described. When I read the "enable them all" I was expecting to see checkboxes to click so the dropdown box of my MSI really confused me. lol

All is clear now though.

It will still get hot under Prime 95 but that's kinda normal for Haswell. Real-world programs/games don't get close to those kind of temps anyway.

I was just about to ask a question about that! If Prime95 stress tests the computer why doesn't it test it using the temps people would actually get in real world?

 

[coffeejunkee]

Yeah, temps are looking good now. The temp sensor in the cpu is only really accurate at high temps though, so that's why you see a value like 19C which is likely below ambient.

As for stress test temps, good question. It's like worst case scenario accelerated error detection and all that. Except it doesn't really work that way, for me anyway. I can run Prime95 for 24H but crash in minutes when video-encoding.

Also, for some reason known only to Intel engineers, Haswell automatically increases cpu voltage when running AVX loads (like Prime, Linx, Aida fpu) so that makes temps even more unrealistic.