Lousy i7-7700k sample for thermals, or am I missing something?

[Sabs]

Hi, this will be a bit long but thanks for reading.

I'm using an i7-7700k on an ASUS STRIX Z270H with Crucial DDR4-3200. I started out with factory clocks and XMP enabled but wasn't happy with my temps.

I'm using a ThermalRight Le Grand Macho RT inside a Fractal R5 case with an extra 140mm front fan (fans at 12V). I picked the CPU cooler as reviews indicate it should perform equal/better to a Noctua D15 with a bit lower noise (on a 6700k, load <60C stock and <70C OC). Review 1, 2, 3

I've noticed that even with XMP off, setting the RAM clock to 3200 bumps up the VCCIO and VCCSA. So I reset defaults and here is what I get while gaming in CS:GO with no OC and the RAM at auto settings (DDR2133):



Seems pretty high to me for stock. I was hoping to run a mild long term OC, maybe 4.7 or 4.8GHz, and of course run my RAM at the rated DDR3200. Here's how ugly it gets with 4.7GHz, 1.25V, DDR3200.



My concern is that this is the result at about 23C ambient, but as I live in a hot climate it'll be more like 27C ambient in the summer.

I still have more settings to test but the best I can do stable is 4.7GHz @ 1.205V and 4.6GHz at 1.18V. I'm sure I can go with a lower manual voltage at stock clocks to keep temperatures in the mid-high 70's under load, but with a massive $80 CPU cooler and stock clocks would be pretty disappointing. Not to mention, once the ambient temp is higher I suspect I'd be in the mid-high 80's.

I've reapplied thermal paste (Thermalright CF III) and noted the cooler was making good contact from the spread of the paste the first time around. The Fractal R5 isn't a super high airflow case with its quiet fans, but from tests I've seen using a i7-6700k rather than 7700k it seems more than adequate paired with a good cpu cooler.

Any suggestions for voltages or anything to check with my hardware? I'd rather not delid if I don't have to - I'm not planning to OC very high and would rather not spend the time/money/take the risk.

 

[trents]

Take you overclock voltage settings off of Auto or off of adaptive. They typically assign way more voltage to cores, IMC and cache than is necessary for stability and generate way higher temps than necessary in the process.
1. Set the vcore to Manual and give it 1.25 volts
2. Return the core speed multiplier ("ratio") to the stock frequency and start the overclock from scratch using stock frequencies to start.
3. Raise the core speed multiplier (or "ratio") by 1x increments until the computer won't boot into Windows or is not stable in Windows.
4. Go back into bios and add .025 to the vcore.
5. Boot back into Windows and start doing stability testing with AIDA64 Extreme or Intel Extreme Tuning Utility (XTU) or Realbench. Montitor core and or package temps which you want to keep under 90c. Run these stress tests for 1 hr.
5. Add vcore in .01 increments if not quite stable.
7. If temps permit, you can try increasing the overclock by the same process.

This should get your temps and voltages under control. AIDA64 Extreme and XTU should be run overnight to confirm stability once you get the overclock roughed in. If using Realbench, confirm the final overclock with a four hr. run.

Be patient with the process.

After getting a stable overclock on Manual mode you can try Offset mode which will allow you to use somewhat lower vcore when not under load. But I would start with Manual to see what is the vcore needed to get your overclock stable under load. That value will still hold true in Offset mode.

 

[Sabs]

I set everything at stock and manually set the CPU core/cache voltage to 1.25V. When using DDR 3200 I've set the IMC to 1.35V; I can try using DDR 2133 and 1.2V but again I'd like to run the modules at DDR 3200. Is 1.35V where I should leave the IMC voltage or should I experiment with lowering it? For the most part it looks like the auto memory voltage is staying <1.2V at DDR 2133.

Here are the results with a manually set 1.25V, stock everything else.

 

[trents]

Your core temps are under control now, peaking at 86c but I would not push it any more without better cooling. Is the system stable? What have you stress-tested it with and for how long? 3200 mhz RAM would require an IMC voltage supplement, yes, but I don't know how much. I have never tried to run fast RAM like that. But I am wondering how much a 1.35 voltage for that component to make the RAM stable is hurting you from a temp standpoint? I can tell you this, the amount of performance increase realized with running high frequency RAM like that is very, very minor. Almost nothing in most apps. You would be much better off from a performance standpoint with lower RAM frequency and a higher CPU clock if it would lower temps enough to allow you to get one more notch or overclock out of the CPU.

 

[Sabs]

Thanks for the replies. I agree the RAM speed isn't too important and there are only a couple games/apps I run which would gain a small % increase from (and probably less than that of a 300MHz or so CPU OC).

The system is stable. I'm just really surprised that my CPU cooler (Le Grand Macho RT), which in tests matches or beats many liquid coolers & high-end air coolers like the Noctua D15, is hitting 85C at stock in gaming.

 

[trents]

I'm guessing you are pretty much topped out on your overclock while using an air cooler, even a good one.

 

[trents]

I don't know how much of a noise freak you are but that huge hunk of CPU cooler heatsink begs for a stronger fan or at least a pull fan mounted on the backside. Where did you mount the extra 140mm case fan?

- - - Updated - - -

I also note that your package temps at idle are 36c. That seems a little high to me. On a 14 nm CPU with only 1.25 vcore I would expect idle package temps to be in the 20s. Are you sure the cooler is seated properly and the TIM is spread properly? You might try redoing the mount and the tIM.

 

[Sabs]

My goal was to have a pretty quiet system with a mild OC, though I'd settle for no OC with quiet fans and decent temps if I have to. I'd like to see what I can do using the existing fans, since from all the reviews I've read (mostly based on the 6700k) I should be able to see pretty good temps. The fan on the CPU cooler is rated for 73 CFM; I would think to get something that moves noticeably more air would add a lot of noise.

There are two 140mm fans up front as intake and one 140mm as exhaust right behind the CPU cooler. The front fans are filtered so the exhaust moves quite a bit more air than the fronts do individually. Pic below.

The idle temps seemed awfully high to me too. I just got done playing with the fan curve for the CPU and I've made some improvement by having it run closer to 50% speed at idle instead of 30% or so. I combined that with using adaptive CPU voltage instead of manual and I'm at 30C idle. I also got the load temps down a bit by having the fan hit full speed at a lower temp.

As far as I can tell the cooler is making good contact as I only spread paste on the CPU itself and it imprinted fully against the cooler once removed. I did redo the entire mount and reapplied paste.

Although things are looking better, I read up and decided to take on the task of delidding the CPU using a kit. Hopefully all goes well. I won't have it for about another week, but I'll report back with the impact it makes. Thanks for all of your input!

 

[EarthDog]

Jebus...

3200mhz should NOT require any SA increase...I needed 1.25 SA for 3866.. 3200 should be good at stock or damn close...

 

[trents]

If you are going to the trouble of delidding I certainly would recommend using Liquid Ultra as the TIM. Otherwise the temp improvement you experience may disappoint you.

Liquid Ultra is a liquid metal product. It must be applied with care but gives significantly better results than conventional pastes. I have used it in several delidding projects without issue.

 

[Sabs]

Jebus...

3200mhz should NOT require any SA increase...I needed 1.25 SA for 3866.. 3200 should be good at stock or damn close...

Good to know - I'll play around with manual settings for VCCSA and VCCIO. It appears on this ASUS board that voltage increases are applied automatically simply by selecting a higher DRAM frequency.

If you are going to the trouble of delidding I certainly would recommend using Liquid Ultra as the TIM. Otherwise the temp improvement you experience may disappoint you.

Liquid Ultra is a liquid metal product. It must be applied with care but gives significantly better results than conventional pastes. I have used it in several delidding projects without issue.

I went with the RockIt 88, relid kit, and Liquid Ultra. There are some good videos out there so I'm no longer too intimidated by the process... I intend to tape off the area around the die and outer area of the lid. Have you used anything like nailpolish to insulate the area around the die? Hopefully all goes well and I'll see some good results using CLU.

 

[Woomack]

On my board I have to set ~1.05V SA for 3200 memory clock ( 0.95V is stock SA ). 1.2V for 3600, 1.3V+ for 3866+.
VCCIO can be ~0.05V lower than SA.
When you keep cache clock low ( ~4.2GHz ) then cpu voltage can be lower. On my CPU it's 4.5GHz@~1.15V, 4.8GHz@~1.18V, 5GHz@~1.285V. 1.25V was still working fine on small air cooler ( I wasn't checking tests with AVX ).
Auto cpu voltage is usually ~1.25V at stock 7700K and SA is 0.95V but some boards set it higher when you start to play with cache and memory clocks. Auto SA at 3600+ memory clock is usually 1.35V or more.

 

[trents]

Good to know - I'll play around with manual settings for VCCSA and VCCIO. It appears on this ASUS board that voltage increases are applied automatically simply by selecting a higher DRAM frequency.



I went with the RockIt 88, relid kit, and Liquid Ultra. There are some good videos out there so I'm no longer too intimidated by the process... I intend to tape off the area around the die and outer area of the lid. Have you used anything like nailpolish to insulate the area around the die? Hopefully all goes well and I'll see some good results using CLU.

Funny you should mention taping off around the die because I was going to suggest that. Yes, I have used nail polish to insulate the components around the die. But I think if you just tape around the die you will be okay. My other suggestion from experience is to shorten the bristles on the applicator brush. They are kind of wimpy and shortening them makes them stiffer and gives you more control. Cut off about 40% of that length and it will make the application much easier.

Use a plastic credit card edge to scrape off the old glue from the die and the lid edges before you apply the liquid ultra. A little 90% isopropyl alcohol helps loosen that crud. And if you accidentally splash some liquid metal over onto components on the die around the core, don't freak out. With alcohol and the scrub pad (or an old toothbrush works better) that comes with the Liquid Ultra kit the liquid metal is removed very easily. Just do it over a trash can or cardboard box because it's very messy. Don't do it over the bathroom sink or your wife will be very upset.

I had a delidded Haswell that I cleaned and reapplied Liquid Ultra to several times so don't sweat it if you have to redo the application. The Liquid Ultra is easily removed if necessary.

When you go to reinstall the CPU, drop the naked CPU in the socket first and then add the lid. That minimizes the chances the lid will slide around on you before you get the CPU in place in the socket.

And you will find that when you go to reinstall the CPU in the socket the retaining clamp will slide the lid forward on the die a little as you lower the lever. Allow for that by placing it a little to the rear of it's normal position to start with.

 

[Sabs]

Thanks for all the input and info trents, and thanks Woomack for the voltages - I'm able to do SA 1.05, VCCIO 1.0, and DRAM 1.25 @DDR3200.

The delid & relid is done with no problems, and I've finished testing. Previously, I wasn't able to attempt 5.0GHz since temps got too high, while 4.9GHz was stable at 1.33V but reached low 90's.

After the delid and changes to VCCIO/SA/DRAM I'm seeing about a 20C drop, so just what I was hoping for. My chip isn't anything special, but I was able to do 5.0GHz at 1.37V stable and will run 4.9GHz at 1.30V long term.

Thanks again for all the help!

p.s. not sure why DRAM voltage is reading high, but the power in watts seems the same as before.

 

[custom90gt]

Makes me want to delid mine...

 

[trents]

Excellent! Did you use Liquid ultra on both sides of the lid or just on the underneath? Most of the temp drop comes from what goes on the underside. Using it on the top side of the lid gives another 2-4c drop in my experience but can be risky if you ever remove the cooler. Liquid metal flakes can fall on motherboard and video card circuitry and create shorts.

Now we know that your CPU had very poor TIM application between the die and lid from the factory. This is when we see those dramatic temp drops from delidding/Liquid Ultra like you got. Not all are quite that dramatic.

Thanks for reporting back.

 

[R_Pierce]

Hmmm, I just bought an i7-7700k as well, and am quite shocked by the temps as well. Although within the "normal" range, seems high to me. Mine is peaking in the mid 80's on stress tests at 5.00 ghz and 1.385 volts. (I have not started to drop the voltage down yet until unstable, will do that tonight) with a Swiftech 320x2 watercooler on it!

I might also try the delid and see how much it helps out. Ill also play around with RAM tweaking like you have above.

 

[Mandrake4565]

Hmmm, I just bought an i7-7700k as well, and am quite shocked by the temps as well. Although within the "normal" range, seems high to me. Mine is peaking in the mid 80's on stress tests at 5.00 ghz and 1.385 volts. (I have not started to drop the voltage down yet until unstable, will do that tonight) with a Swiftech 320x2 watercooler on it!

I might also try the delid and see how much it helps out. Ill also play around with RAM tweaking like you have above.

What stress test? Mid 80's at 1.385 volts doesn't sound high at all.

 

[R_Pierce]

Realtest and Prime 95 (the mixed test). I guess I figured it would run a bit cooler with a water cooler. My i5 6500 ran at about mid to upper 70's at 1.435 volts with a cryorig R1 cooler.

 

[R_Pierce]

Alright!! I have gotten the voltage down to 1.350 at 5.0ghz. I can't get it to boot at 5.1 with voltage all the way up to 1.385. So I think it's safe to say that 5.0 is about my chips limit. With the reduction in voltage, I am now seeing a peak of 75*c which running the HWBOT RealBench test. Average of 49*c.