Poor Performance with the H100i Pro? - You Decide

[jayhall0315]

Hello Veterans! - I am new to the forum and recently purchased a Corsair H100i Pro 240 AIO system to tame my 8700k chip when overclocking. The only problem is that I can tell very little to zero difference over the previous CryOrig H7 it replaced. My home (and the room I am using) are at 26 deg C. With the H7 and my 8700k at stock, I was getting 35 to 41 deg C during idle and about 93 deg C during Prime95 runs over 30 min. Running the chip at 5 GHz and 1.35 volts was impossible. It would idle at about 41 deg C and go up 100 deg C during Prime95 runs. It would fail Intel XTU CPU stress tests and could not complete a Prime95 run.

 

Therefore, I ordered the H100i Pro but my temps remain about the same. (I have used custom watercooling loops with Heatkiller, EK and XPSC components in the past, but don't want to fool with the setup this time) I have set the 2 Corsair fans that come with the H100i Pro to push against the radiator and expel hot air out the top of the case. The coolant temperatures for the H100i Pro are typically about 31 to 32 deg C during idle with the 8700k at 5 GHz (and with Speedstep enabled in the BIOS). During idle at 5 GHz and 1.35 volts, I run about 35 to 39 deg C and I can complete some Intel XTU CPU stress tests (on some days oddly but not others) where the temps fluctuate between 66 or so and about 86 deg C (see image 1). When I run Prime95 overclocked, I cannot complete even 5 min. (see image 2) At about the 2 min mark with blend, my temps which stay initially stable at 75 deg C roughly, quickly move up to the high 90s deg C range and then blue screen or instability. Quite frankly for a liquid cooling solution this strikes me as poor. I am uncertain if the fault lies in the chip, the H100i Pro or something else. Perhaps I lost at the Silicon Lottery :(

 

I have not delided the chip and don't plan to do so. I have remounted the H100i Pro copper plate/ pump head three times and checked the pattern of the thermal paste (using Grizzly Kryonaut) but it appears fine and normal. I can see no kinks in the tubing and the pump is not making any strange sounds that would indicate a blockage or problem. The temperatures ramp up and down normally under Prime95 and as angulated sine wave in XTU testing. Not sure if this is correct for the H100i Pro. I am not using the Link software, I am using iCUE 3.4.95. From iCUE, I have tried quiet, balanced and extreme profiles with little actual difference to the processor temps. I also see no way to ramp up the pump rpms and fans as CPU temps increase (or at least I see no way to do this in iCUE). In the ASUS BIOS, I have set CPU_FAN header to disabled and it reads under Q-Fan as PWM and 100%. (it would be really nice Corsair, if in your documentation you could mention how to set the BIOS for the H100i Pro)

 

Curiously, iCUE always shows the pump RPMs as about 1080. This never changes no matter what I do. Is this the correct RPM range for this H100i Pro pump? Seems low to me. Finally, does this seem par for the course or does this strike you as being off? I expected with liquid cooling that I would be able to hit 5 GHz at 1.35 volts with idle temps in the 30s, gaming in the 50s deg C and Prime95 runs in the high 80s but I am not anywhere close to that. Any help would be appreciated, or I have to return this product.

[jayhall0315]

Incidentally, if any gurus of Corsair or their engineers come across this thread, you can easily improve the performance of this H100i Pro unit by having the hoses come down straight into the pump and ditching the 90 degree swivel connectors that impede flow. You will see a 2 to 4 deg C positive improvement if you ditch those 90 degree bends. Also, the tubes should be 3/8 or 1/2 inch. You can hold more water, and more water can dissipate more watts. Perhaps I am overlooking some obvious simple solution, but iCUE needs a simple fan profile setup (like ASUS Q-Fan allows) for how to manage the fan curves on the radiator.

 

Maybe I missed it somewhere but I can find no place that lists the specs for this pump? If I can get the pump up to 2500 rpm or so, cooling will be greatly improved. Allowing this pump and the fans in connection to ramp up from something like 500 rpms at idle to 2400 rpms at full load under Prime95 would do wonders for this unit.

[DevBiker]

Can you show the iCue configuration page for the pump? Also, what fan curve are you using? And what are your coolant temperatures?

[jayhall0315]

My coolant temps currently sit at about 31.5 with the 8700k overclocked at 5 GHz at 1.35 volts (but SpeedStep and Turbo mode are enabled in the BIOS and C-States are off). So please remember that the CPU is cycling through various frequencies and temps depending upon load. (Please see the attached image) As to fan curve, I am currently using "balanced". Although, I have used extreme and fixed at 100% and it didn't make a great deal of difference. (Fixed at 100% with the fans spinning full out max at about 2450 rpms produced CPU temperatures about 3-5 deg C cooler but Prime95 runs still fail)

[jayhall0315]

I will test the coolant temp and take a screenshot under Prime95 here in a little bit and post that too, to help out.

 

(Also, I am well aware that temps would be cooler if I had the fans pushing cool air from outside the top of the case down onto the radiator, but then roughly 14 to 150 watts of waste heat would be sent right onto the motherboard and GPU, so I settled for this trade off with the fans in a push config, expelling waste heat out of the top of the case)

[Zotty]

Personally.... run the pump on balanced for normal use and extreme when stressing the system... I tend to run Fixed RPM's for the Fans my self... fixed 850rpm for light loads and fixed 1200rpm for stressing/gaming

[jayhall0315]

During a Prime95 run, the coolant temps got up to 38.7 deg C on "balanced" profile and were still going up before I killed the workers. (if I don't kill Prime 95 by about this point, I get a frozen screen and have to reboot) See attached image and look on the far right for the coolant temp as reported by iCUE 3.4.95

[jayhall0315]

Running the H100i Pro at full out max settings with "Fixed" enabled in iCUE and set to run fans at 100%, things are not much better. The coolant temps were 37.6 and rising before I killed the workers. (please see attached image) With "Fixed" at 100%, I could run Prime95 for about 5 min and 10 seconds before I got to the mid 90s deg C. With "balanced", I could run Prime95 for about 2 min and 50 seconds before reaching the mid 90s deg C. (For both runs the 8700K was overclocked to 5 GHz at 1.35 volts with SpeedStep and Turbo enabled, and C-States off)

 

Note that no matter what profile I select, the pump always runs at ~ 1080 rpms. What is up with that? If the pump ran up to perhaps, 2500 rpm, it would increase performance greatly. (at the cost of longevity to some degree)

 

I am not sure what the heat load of an 8700k at 5 GHz is, ...perhaps 120 to 155 watts? Not sure, but I thought the H100i Pro could handle up to 150 watts? Am I incorrect? The H100i Pro should be able to handle my 8700k and if I take an off the cuff heat equation (PDE) with 150 watts as a baseline, I approximate I should be at about 85 to 87 deg C and then max out there.

[c-attack]

For starters, take the pump out of the silly quiet mode to run Prime 95. That is for sitting on the desktop reading your emails and is going to set you back 2-3C at load. The pump is not dynamically adjustable. It has three preset speeds: Quiet (1100), Balanced (2100), Extreme (~2900). It works like the fans and you need to click the preset on the left, then apply it to pump. The pump will highlight yellow when changed. On the bigger surface area H115i/H150i PRO models, 2100 vs 2900 makes no difference at all. 1100 usually does show a small penalty. We haven't seen a lot of test user data for the 240mm. 1100 is most certainly going to cost you. It is possible there is small 1-1.5C difference between 2100 and 2900 on the H100i Pro, but obviously that is not the change you are looking for. Since the pump is not dynamically adjustable, my recommendation is to park it on Balanced and forget about - except maybe when doing this stuff.

 

However, neither pump speed or fans is your real problem. I can hook 720mm or 10ft radiator up to this and it isn't going to change the outcome. When you start XTU or Prime and the full voltage kicks in, you see a jump from 35 to 75C (or whatever it is for you). I am not sure it has an official name, but I often refer to it as the power on delta T or sometimes in terms of liquid temperature as the CPU to coolant differential. This value is entirely dependent on your CPU model, its physical construction, conductivity through the CPU lid/TIM/cold plate sandwich, and of course... the voltage applied at the pins. The last one is the only thing you have any control over. The radiator size, pump speed, and fan settings have no effect on this at all. This stage 1 part of the cooling process is all conductive and if you were building your own custom loop with unlimited funds, you could hunt down a cold plate with exotic metals to increase the conductivity. Those often promise 1-2C improvements over the more common copper and copper allow plates. Again, not what you are looking for.

 

So if your power on delta T at 1.35v is +43C and the coolant temp is 30C, then the only way to make it any lower is to reduce to the room temp by 10C and everything else inside the case. That may happen two months from now, but in the meantime the only thing you can do is lower the voltage (since delidding is off the table). In your Prime 95 graph you can see the issue. Your CPU package temp is a flat line. If the cooler was being overwhelmed, the CPU temp would tick up +1C every minute or so. Water coolers offer you a greater heat capacity over time compared to air towers, but neither can alter the stage 1 instant voltage on temp change.

 

The cooler is Stage 2, which involves getting rid of the heat faster than the new heat enters the system. With a water cooler, you can see this clearly as the coolant temp changes. With the larger 280 and 360mm versions, fan speed isn't much of an issue. With the 240mm, blasting the fans isn't going to help much, but there likely is a few degree difference between base minimum and moderate speed. The fans help dissipate heat from the radiator. That's it. The theoretical maximum reduction is coolant temp is back to the starting point. So if you started your test at 30C coolant temp and it is 35C now, the most you could reduce the temp is by 5C. Nothing is 100% efficient so you will never get all the way back to starting value, but that is the general idea. With a proper pump speed and moderate fans, I think you can take the coolant down to +5C delta. That will be a 3C improvement. I typically see real use coolant deltas of 3-5C with my larger 280mm at 1.30v on the 8700K. Benchmarks about +5C with only about 1000 rpm on the 140mm fans.

 

There are a few things you should do in the Asus BIOS to help with both this testing and your everyday use. If you have not seen it, take a look at the Asus OC guide for Kaby Lake (same things for Coffee). It's a bit technical and not all of it is necessary information, however there are a few key things. The most critical involve managing the Vcore beahvior. Asus has all these modifiers in the Z370 BIOS that can ruin your day. You have not mentioned whether you are testing with a fixed/manual Vcore or on adaptive. Adaptive will require some additional tweaks to keep the voltage in check. Both will be affected by your LLC setting. To that end, while you may have set 1.35v in the BIOS, that does not mean that is the actual end result. Double check this in one of the monitoring program. If it is ballooning you up to 1.40v, that would be a definite shift in temperature, albeit possibly necessary for stability. Either way, excessive overshoot is not what you want here.

 

**Almost forgot. BIOS Q-Fan settings. I am not sure we have talked about this for the H100i PRO. If that thing has a SATA connector you hooked up, that is your power and the BIOS/motherboard/Qfan end doesn't matter. The motherboard lead only sends a speed back. If there is no SATA connection, then the power comes down that 3 pin line and it needs to set at 100%/Full Speed or "disabled" in the BIOS. Since your pump and fan speeds are not being affected, it seems you are sending the necessary power. When you don't, it's like sliding a dimmer switch down with pump and fans dwindling.

[jayhall0315]

Thank you very much C-Attack. I looked carefully (or thought I did) in the iCUE software and did not see a way to increase the pump speed. Now, I see what you mean. Will slowly digest what you mentioned and get back to you here in a bit! Thank you and thank you!

[jayhall0315]

I am using 1.35 volts set to manual for now C-Attack. I may play with adaptive mode in a few days. Also, I am using an offset for AVX instruction sets of -3 or -2. For now, I just want to stress test and see what the 8700k I have is capable of.

 

With the pump on "extreme" at about ~ 2950 rpms and the fans set to "extreme" as well (each fan was at about ~ 2460 rpms), I was able to run Prime95 for 30 min although temps did reach into the mid 90s deg C. The coolant temps were typically about 35 to 36 deg C during this run. But it did stay stable at 5 GHz at 1.35 volts. I am guesstimating here, but I would say the 8700k at 5 GHz puts out about 150 watts and that is about the thermal load capacity of the H100i Pro.

 

I am reading the article on Kaby Lake now... thank you again C-Attack.

[jayhall0315]

I would like to mention for any Corsair staff or engineers who happen to read this:

 

AT NO PLACE IN THE INCLUDED INSTRUCTIONS THAT COME WITH THE H100I PRO IS THERE ANY MENTION OF 3 DIFFERENT PUMP SPEEDS OR HOW TO SET THEM!

 

There is also no mention in the included instructions of how to setup your H100i Pro with Link or iCUE and no information on how to adjust settings or set up the unit for overclocking. Since the pump by default is set to quiet ( ~ 1080 rpms) and iCUE is not clear about how to turn on the higher speeds, many folks buying the H100i Pro are likely to be upset about their thermals and consider returning the unit. IMHO, without being mean to Corsair, that is very poor indeed. If it had not been for C-Attack above, I likely never would have known how to set the pump speed to another setting and may have returned the unit. Please consider adding a page or two to the instruction guide that comes with the H100i Pro, so that new guys get their fans and pump set up correctly in either Link or iCUE.

[DevBiker]

Take a look at my signature and "iCue for the Link User".

All of the Pro series pumps are the same.

 

But ... I can't argue with the complaint. It's not the first time that important details have been left out of the manual either (like setting the fan header to 100% in the v2 series coolers.)

[tiessar]

There is nothing you can do if you don't want to delid. Run the CPU at stock and you will have more than acceptable thermals.

[mrpetrov]

There is nothing you can do if you don't want to delid. Run the CPU at stock and you will have more than acceptable thermals.

 

I second this. After some trepidation, I delidded and repasted my 8086K about three days ago and my temps dropped 25-30C under P95 load. In hindsight, delidding is easy, it really is not difficult, nor require particular manual dexterity. Repasting is marginally more tricky, but watching a couple of videos and taking things slowly make it a lot less daunting. Still, if you won't delid and relid, then best of luck - you'll just have to lower your voltages and see how stable you can get it. No cooling solution can change that.

[jayhall0315]

Petrov, what thermal paste or liquid metal did you use during the delidding?

[mrpetrov]

Petrov, what thermal paste or liquid metal did you use during the delidding?

 

I relidded using Grizzly Conductonaut Liquid Metal between the CPU and the IHS. Between the IHS and the AIO cooler block I used Grizzly Kryonaut. I used Permatex 22072 Ultra Black Maximum Oil Resistance RTV Silicone Gasket Maker between the CPU PCB and the IHS (just on two of the sides - very small amounts) instead of any sort of Locktite or super glue.

[c-attack]

^^^^^ As above. Note he used the liquid metal (Conductonaut) under the CPU lid only, then used the more traditional paste (Kryonaut) between the CPU and cooler cold plate. Liquid metal products will strip the finish on the cold plate and CPU in less than 30 days. It’s not worth it and after delidding a 1-2C difference will be irrelevant anyway.

[jayhall0315]

Just to understand C-Attack, you are saying that what Petrov did is correct and to use liquid metal on the die after delidding but not on top of the IHS that makes contact with the H100i Pro pump cold plate? I was unaware that liquid metal would create a reaction with the copper baseplate of the H100i Pro pump head.

[c-attack]

Yes. It doesn’t take very long. I’ve done it with new CPUs and new coolers. The plate and lid start to show change in a matter of weeks. I’m not suggesting this leads to performance degradation or that it will ruin either product. However the visual change is noticeable and if you read the warranty on some products, it could be viewed as a violation of terms. More practically, most CPU vendors will not buy a used CPU that has been in contact with liquid metal. This is mostly about appearance, but there are consequences. *(In addition to the living nightmare of losing a single drop of liquid metal anywhere inside your case)

[jayhall0315]

Just wanted to pop back in to say that I did go ahead and delid my 8700K and had a huge temperature drop. I was expecting something like 8 to 15 deg C lower if I did the process correctly, but I got 35 deg C lower!!! (at load)

 

I followed this process here (along with watching some other vids as well):

 

[ame]

[/ame]

 

(The only step he leaves out is to apply some clear nail polish to the gold contacts on the 8700k. And then make sure to place the gasket maker compound fully around the "bottom" orientation of the IHS, in the very, very slim chance that any liquid metal should slide down.)

 

Previously, using the H100i Pro with the pump and fans on "balanced", I was still getting up to 93 to 95 deg C on the 8700K at 5 GHz at 1.35 volts with Prime 95 blend at the 5 min mark roughly. After delidding and using Grizzly Conductonaut liquid metal on the cpu die, and then using Grizzly Kryonaut thermal paste between the IHS and H100i Pro cooling plate, my temps dropped to only 58 to 60 deg C at the 10 min mark with Prime95 blend!!! What is even more surprising is that when I delidded the IHS off the 8700k, the thermal paste Intel chose, seemed to be applied evenly with good contact (of course it is thermal paste and not liquid metal, ... but still).

 

Now, with the H100i Pro, I am getting about 30 to 31 deg C on the 8700k at 5 GHz but idling with only Chrome open (in comparison to about 37 deg C previously)(I kept SpeedStep, Turbo Mode and C-States all enabled in the BIOS). Compiling code (all cores used) and my temps never go above about 49 deg C. Gaming multi-core and my temps never go above about 51 deg C and with Prime95 at the 10 min mark, they are about 60 deg C.

 

I then decided to see what the combination of delidding and H100i Pro could achieve. After fooling around for some time, I was finally able to settle on 5.3 GHz at 1.41 volts with a stable Prime95 run for 30 min with temps at about 80 to 83 deg C. Above that or with more voltage and things quickly went south. Still, 5.3 GHz is way beyond what I expected. I turned it back down to 5 GHz and this is what I will run day to day.

 

So, ..... huge difference for me. Obviously, if you have a "K" cpu and delid it, you are going to see improvements. I am just astounded I see a 35 deg C difference under load!

[DevBiker]

Pretty astounding results. I'd guess that you had a particularly poor TIM paste on your processor - not only silicon lottery but paste lottery as well?

 

The good news is that Intel is going back to solder TIM for the next round.