Jump to content
Corsair Community

H115i losing performance?


rx7dude

Recommended Posts

Hello,

 

I recently upgraded my system to a Z390 Aorus Ultra motherboard and 9700K from a 7600K. It is being cooled by my nearly 3 year old H115i with ML140 pro fans.

 

I have noticed that while idling my 9700K hovers around the 29 - 35C range and while gaming never exceeds 65C when set to High Performance in windows. This is with the processor at stock settings and MultiCore Enhancment Disabled in BIOS. However, one run in Cinebench R15 produces worrying results as my processor shoots up to 85 - 90C. This is with Pump and fans set to extreme in iCue. Setting the fans to 100% drops the temps down to 75-77C but still seems rather high for a stock 9700K. The reported Voltage by CpuZ and HWInfo64 is a max of 1.248V which is relatively low.

 

Is this a sign that my H115i is losing performance or am I just unlucky and got a bad chip (I hope not).

 

Also I am using Kryonaut paste in a spread method (have reapplied twice using dot and spread both with same results) and my ambient room temperature is 23C. Cooler is mounted very firmly. The processor even with MCE disables still likes to ramp up to 4.7 - 4.8ghz across all cores, and given the 1.248 max volts it pulls, seems reasonable.

 

I have a feeling it is the H115i as gaming temps are super reasonable at 55-65C depending on ambient. Its only under 100% load it shoots up and that makes me wonder if the pump is not actually at 3200RPM. At this setting the pump also has no audible sound whatsoever.

 

Further note: Liquid temperatures idle at 28 to 30C and under gaming never exceed 37C.

Edited by rx7dude
Link to comment
Share on other sites

That's just normal for an 8 core i7 or i9 CPU 9th Gen. Mine used to do the same thing but the AIO is not such a great cooling performance option but more like a step above the HSF and looks.

Mine wasn't that good and it was outperformed by a be quiet! Dark Rock Pro 4 and way, way, way much quiet than those 140mm Corsair fans at full speed.

 

Now I have a Hydro X loop and it's much better, but at 4x the cost. Not everyone is willing to spend more than $400 for a water loop.

 

Just remember, these CPUs are not Quad-Cores with HT but double the core count so it's normal that the CPU will heat up pretty quick.

Link to comment
Share on other sites

You need to look at the coolant temperature change to assess cooler efficiency. Lots of reasons for the CPU to be one temp or another, but with the cooler it's about ambient air temp vs coolant temp change.

 

Most people will see coolant temps of +4-6C above the room temp at idle. This varies with processor, power settings, and of course case, layout, and relative room position. At load on a 9700K, you should see about +6-7C for CPU test over the first 15-20 minutes. After that, any deviance is usually down to changes in case temp.

Link to comment
Share on other sites

You need to look at the coolant temperature change to assess cooler efficiency. Lots of reasons for the CPU to be one temp or another, but with the cooler it's about ambient air temp vs coolant temp change.

 

Most people will see coolant temps of +4-6C above the room temp at idle. This varies with processor, power settings, and of course case, layout, and relative room position. At load on a 9700K, you should see about +6-7C for CPU test over the first 15-20 minutes. After that, any deviance is usually down to changes in case temp.

 

Coolant temperature is fine. Around ranging from 28 to 37C depending on task at hand. This seems correct given my ambient is around 23 to 24. I am not sure what is wrong here.

Link to comment
Share on other sites

That's just normal for an 8 core i7 or i9 CPU 9th Gen. Mine used to do the same thing but the AIO is not such a great cooling performance option but more like a step above the HSF and looks.

Mine wasn't that good and it was outperformed by a be quiet! Dark Rock Pro 4 and way, way, way much quiet than those 140mm Corsair fans at full speed.

 

Now I have a Hydro X loop and it's much better, but at 4x the cost. Not everyone is willing to spend more than $400 for a water loop.

 

Just remember, these CPUs are not Quad-Cores with HT but double the core count so it's normal that the CPU will heat up pretty quick.

 

I am highly considering upgrading to a Noctua NH-D15 Chromax Black the newly released ones. Would this give me better temperatures? I am kind of getting annoyed with AIO cooling as the results seem to be in favor of Air coolers both performance and noise wise.

Link to comment
Share on other sites

Even with MCE disabled, there is a lot more to do to get most recent motherboard BIOS versions ready to run stress tests. The manufacturers goal is to get even the worst CPU in history to load up, not get the leanest temps possible for someone running a synthetic bench. When in doubt, pile on voltage.

 

CPU temps are voltage based and change instantly when this is the cause. A cooler, regardless of type, does not stop conducting heat no matter what you do with the fans or pump. If you don't get the heat out of the cooler, it eventually becomes an add-on penalty to the CPU. Coolant temperature is the minimum possible CPU temp, so when the coolant goes +6C, you have raised the baseline by 6 and with no other changes, the CPU temps are shifted +6C upwards. However, these changes are slow! You can watch the coolant temp change up or down and 2C in a minute is about the limit either way. When you start the stress test and the CPU temp goes from 30 to 80C, that is all voltage and CPU physical properties. There isn't much you can do about the physical aspects. You can refine the voltage and power behaviors.

 

Get off Auto voltage if you are using it. You can still use adaptive voltage, it just has to be dialed in. I don't have any GA boards, so I can't walk you through their BIOS. Certainly on the Asus end there are several redundant settings that will all clock things up when you might not be expecting. There likely is a motherboard specific guide out there that can pin down everything that needs to be modified. Your temp swing at the top 85-90C then dropping to mid-70s is too much to be cooler related. The voltage may not be staying at that 1.24v level.

Link to comment
Share on other sites

Get the be quiet Dark Rock Pro 4, is better looking and performs head to head with the NH-D15. Plus it has 7 pipes vs 6.

 

I got one and from using both, aesthetics is somewhat important and the DRP4 is so much better and performance-wise both are equal I'd say.

 

IMO and experience, a CPU block that's just bare copper (like all AIO) doesn't cut it anymore for this Multi-CPUs and both, my new CPU water block in the Hydro Loop and DRP4 have Nickel-Plated base and definitely make a difference, so much so that the DRP4 does a better job cooling my i7 7700K by 10 degrees lower than the AIO 280mm RAD.

Edited by Nazgul
Link to comment
Share on other sites

I am highly considering upgrading to a Noctua NH-D15 Chromax Black the newly released ones. Would this give me better temperatures? I am kind of getting annoyed with AIO cooling as the results seem to be in favor of Air coolers both performance and noise wise.

 

Generally speaking an air cooler is usually quieter than water cooling in a couple of ways. 1) No mechanical pump makes it quieter in low noise situations. 2) lower density (thicker) radiator makes less noise when air passes through it. However, no air tower is going to be able to keep up with a 280mm radiator in terms of heat capacity. Even if you find a benchmark test that says they are only "3C apart" (or whatever), in real world use your average temps are going to be considerably higher, particularly if you intend to overclock.

 

On the cooler end, you total coolant rise is 6-8C. That is the value you are attacking with fan and pump speeds. If I gave you a 2 meter long panel with 16x120mm fans, you could still only reduce your total CPU temp by 6-8C. We are all voltage limited on our CPUs and as Intel stretches the 14nm model to the umpteenth degree, there isn't much room left. All of these are going to run close to max right out of the box. From what I have seen, delidding on the 9000 series is not exactly overwhelming, unlike the 8000 series where it almost seemed mandatory for a 20C reduction. Refining voltage behavior is about all you can do.

Link to comment
Share on other sites

Get the be quiet Dark Rock Pro 4, is better looking and performs head to head with the NH-D15. Plus it has 7 pipes vs 6.

 

I got one and from using both, aesthetics is somewhat important and the DRP4 is so much better and performance-wise both are equal I'd say.

 

IMO and experience, a CPU block that's just bare copper (like all AIO) doesn't cut it anymore for this Multi-CPUs and both, my new CPU water block in the Hydro Loop and DRP4 have Nickel-Plated base and definitely make a difference, so much so that the DRP4 does a better job cooling my i7 7700K by 10 degrees lower than the AIO 280mm RAD.

 

I think I will get one of the 2 between DRP4 and NH-D15 Black. Not only because they provide similar if not better cooling than 240/280m AIO, but because you get peace of mind of nothing besides a fan failing.

Link to comment
Share on other sites

Even with MCE disabled, there is a lot more to do to get most recent motherboard BIOS versions ready to run stress tests. The manufacturers goal is to get even the worst CPU in history to load up, not get the leanest temps possible for someone running a synthetic bench. When in doubt, pile on voltage.

 

CPU temps are voltage based and change instantly when this is the cause. A cooler, regardless of type, does not stop conducting heat no matter what you do with the fans or pump. If you don't get the heat out of the cooler, it eventually becomes an add-on penalty to the CPU. Coolant temperature is the minimum possible CPU temp, so when the coolant goes +6C, you have raised the baseline by 6 and with no other changes, the CPU temps are shifted +6C upwards. However, these changes are slow! You can watch the coolant temp change up or down and 2C in a minute is about the limit either way. When you start the stress test and the CPU temp goes from 30 to 80C, that is all voltage and CPU physical properties. There isn't much you can do about the physical aspects. You can refine the voltage and power behaviors.

 

Get off Auto voltage if you are using it. You can still use adaptive voltage, it just has to be dialed in. I don't have any GA boards, so I can't walk you through their BIOS. Certainly on the Asus end there are several redundant settings that will all clock things up when you might not be expecting. There likely is a motherboard specific guide out there that can pin down everything that needs to be modified. Your temp swing at the top 85-90C then dropping to mid-70s is too much to be cooler related. The voltage may not be staying at that 1.24v level.

 

Yes you may be right. However, HWInfo64, HWMonitor, and CPU-Z are all reporting a max voltage of 1.248 at 100% load. I don't know if there is something I am missing or what. The only reason I stay on auto-voltage is that I normally use my PC on the Balanced performance mode and have it set so that my CPU only runs at a max clock of 2.8ghz. I do not need the 4.6-4.8 turbo simply for using Firefox and Word. Auto voltage then drops the voltage due to that Balanced mode to around 0.652-0.848 volts. Do you think there is a possibility that these softwares are not reporting the right voltage? And if so, is there a way I can set a MAX voltage the motherboard is allowed to supply? Or should I just overclock my CPU to 4.8ghz across all cores?

Link to comment
Share on other sites

If you are trying to cap the frequency at lower levels (for whatever reasons), you would be better off setting up per core frequency within the BIOS. When all 8 cores are loaded 45 x 100, 4 cores 46x100, 2 cores 47, 1 x 48, or whatever your frequency targets are for your usage. Setting a specific adaptive voltage will still allow the value to drop when not fully loaded. The default "auto curve" tends to be on the high side and there are special behaviors built into. On some boards you can now alter these, but this is an area where you need GA specific knowledge. I suppose using Windows power plans is another way to do it, but software control may not be as efficient as the hardware level settings in some circumstances. The default high performance plan has a 100% minimum clock frequency, so you would be better off cloning the Balanced Plan and then raising maximum frequency back to 100% for mixed use.

 

Aside from all of that, your CPU temps are still going to be high when running at maximum stress test, particularly with AVX instructions. That's just how it is. When you start the test and your temps are 85-90C 1 second later, the cooler doesn't have anything to do with it.

Link to comment
Share on other sites

If you are trying to cap the frequency at lower levels (for whatever reasons), you would be better off setting up per core frequency within the BIOS. When all 8 cores are loaded 45 x 100, 4 cores 46x100, 2 cores 47, 1 x 48, or whatever your frequency targets are for your usage. Setting a specific adaptive voltage will still allow the value to drop when not fully loaded. The default "auto curve" tends to be on the high side and there are special behaviors built into. On some boards you can now alter these, but this is an area where you need GA specific knowledge. I suppose using Windows power plans is another way to do it, but software control may not be as efficient as the hardware level settings in some circumstances. The default high performance plan has a 100% minimum clock frequency, so you would be better off cloning the Balanced Plan and then raising maximum frequency back to 100% for mixed use.

 

Aside from all of that, your CPU temps are still going to be high when running at maximum stress test, particularly with AVX instructions. That's just how it is. When you start the test and your temps are 85-90C 1 second later, the cooler doesn't have anything to do with it.

 

Hello again friend, I decided to take your advice and adjust settings in the BIOS. Watching many guides on overclocking/undervolting. You were right. The motherboard was indeed not pushing 1.24 but rather 1.284 and nearing 1.3V for stock settings with 4.6 across all cores. 1.284V is insanely high for stock clocks so I toned it down to 1.18V as that is what got me to be stable. Temperatures are WAY down now. Before seeing an average of 83-85 and 90 sometimes, and now to high 60's and max of 72. Does that 72 spike still seem a little high for stock 4.6 at 1.18?

Link to comment
Share on other sites

No, I don't think there is a 9000 series processor out there that won't hit 70 in some peak moment. While peak temps are important to keep you out of the silicon melting and throttling zone, I suspect your average temps are going to be way down as well. Picking stray or outlier peak values out a typical data set is pretty hard, but most do have an average (per session) type of log that should give you a better sense where things are usually at. Long duration line graphs are even better.
Link to comment
Share on other sites

No, I don't think there is a 9000 series processor out there that won't hit 70 in some peak moment. While peak temps are important to keep you out of the silicon melting and throttling zone, I suspect your average temps are going to be way down as well. Picking stray or outlier peak values out a typical data set is pretty hard, but most do have an average (per session) type of log that should give you a better sense where things are usually at. Long duration line graphs are even better.

 

My average temps are indeed way down now that I undervolted to 1.18V. I was playing a Plague Tale Innocence for a good hour or so and temperatures did not go past 54C with all cores running at 4.6 to 4.8. Before I would see them in the low to mid 60's.

Link to comment
Share on other sites

No, I don't think there is a 9000 series processor out there that won't hit 70 in some peak moment. While peak temps are important to keep you out of the silicon melting and throttling zone, I suspect your average temps are going to be way down as well. Picking stray or outlier peak values out a typical data set is pretty hard, but most do have an average (per session) type of log that should give you a better sense where things are usually at. Long duration line graphs are even better.

 

Back to the bad news. Turned out my 1.188 was not stable after further testing. I decided to go the adaptive voltage route. Set an offset of -0.090 and under full 100% I have not seen anything more than 1.224V. Yay! Lower voltages than Auto...yeah the temperatures were not that friendly.

 

At 1.212 - 1.224V my 9700K at 4.6Ghz with the H115i Set to EXTREME on both fan and pump sees Cinebench reaching 80-82C and Aida64 hit 93C. Liquid temperatures rise FAST going from 31C to 43/44C almost after 2 minutes of Aida64. How the heck is a H115i not handling 1.212V??

Link to comment
Share on other sites

BLiquid temperatures rise FAST going from 31C to 43/44C almost after 2 minutes of Aida64.

 

That really shouldn't happen. Run AIDA for 5 minutes or enough to heat the coolant 6-10C. Then kill the test and note how long it takes for the coolant to come back to the original point. It should drop 3-5C in the first 3-4 minutes. The last couple of degrees usually take another 5-10 min, but the point is whether it drops or not as soon as you kill the load. If it does not, it means either the entire internal case area is also the same temp as the coolant or their is a problem with the cooler and it can't expel heat. I have never seen my coolant on my H115i Pro go +10C over hours and hours, let alone 10 min of AIDA. That would be more like 4-5C on a heavily overclocked 8700K.

Link to comment
Share on other sites

That really shouldn't happen. Run AIDA for 5 minutes or enough to heat the coolant 6-10C. Then kill the test and note how long it takes for the coolant to come back to the original point. It should drop 3-5C in the first 3-4 minutes. The last couple of degrees usually take another 5-10 min, but the point is whether it drops or not as soon as you kill the load. If it does not, it means either the entire internal case area is also the same temp as the coolant or their is a problem with the cooler and it can't expel heat. I have never seen my coolant on my H115i Pro go +10C over hours and hours, let alone 10 min of AIDA. That would be more like 4-5C on a heavily overclocked 8700K.

 

Under gaming coolant will rise about 5-7 degrees then stabilise and stay there. I test by playing 1 hour of Forza Horizon 4 and coolant went from 32 to 37 and stayed there.

 

Not even 3 minutes into Aida64 it rises from 32 to 42/43. It does however drop quite quickly. I measured it out and it took around 6 minutes for it to drop from 43 to 35.

 

I've shaken the entire unit like a madman hoping any blockages would go away. Same results. Not cool, this is exactly why a NH-D15 is better. You dont have to worry about all leaks, pump failures, or blockages. Not to mention the similar/better performance, reliability and reduced noise.

Link to comment
Share on other sites

It doesn't seem like a blockage. The gaming coolant level is right on. It's cooling off. A blocked cooler would not - one way traffic to hot. So the only operating theory I have is either an unintended AIDA configuration in the test or we are still looking at some BIOS setting that is massively increasing activity when under a synthetic load (vs the natural load of a game). I don't typically think of AIDA as overly stressful, except perhaps FPU only tests. Which version of the stress test are you using? CPU only? Or all boxes checked (blend test)?
Link to comment
Share on other sites

It doesn't seem like a blockage. The gaming coolant level is right on. It's cooling off. A blocked cooler would not - one way traffic to hot. So the only operating theory I have is either an unintended AIDA configuration in the test or we are still looking at some BIOS setting that is massively increasing activity when under a synthetic load (vs the natural load of a game). I don't typically think of AIDA as overly stressful, except perhaps FPU only tests. Which version of the stress test are you using? CPU only? Or all boxes checked (blend test)?

 

BIOS:

CPU is set to Adaptive voltage with a negative offset of -0.080. Vcore never rises more than 1.224. MCE is disabled. LLC is set to Auto. VCCIO is set to 1.05. Xmp profile 1 selected boosting ram to 3000Mhz. Nothing else has been touched.

 

Aida64:

Stress CPU, FPU, Cache, System memory are all checked. These were default settings.

Link to comment
Share on other sites

The all 4 checked is the blend test. That is what I use the majority of the time for general stability testing. That should not raise the coolant like that.

 

It's hard for me to assess the BIOS settings without having any time on that brand. However, I suspect you need to get off LLC auto. Usually the relevant guides or forum junkies talk about this incessantly. Asus uses a 1-7 scale for my board. I use 6 when clocked to 90-95% of maximum capability. If I were lower I would use 5. The problem with the Asus one is it spends all its time on 1 or 7/max. Synthetic load? Instant max. This usually shifts the voltage to some degree.

 

I don't know about negative offset vs specifically enumerated adaptive voltage with auto offset (that is how it works on Asus). So rather than guess what I think adaptive is and deduct the offset, you set the maximum and it stays there while following the curve for submaximal frequencies. Negative offsets can be difficult and require very precise tuning. Really easy to lock up/crash as lower voltages in seemingly low idle states.

 

I will try and run a blend test with fans off and see what happens.

Link to comment
Share on other sites

The all 4 checked is the blend test. That is what I use the majority of the time for general stability testing. That should not raise the coolant like that.

 

It's hard for me to assess the BIOS settings without having any time on that brand. However, I suspect you need to get off LLC auto. Usually the relevant guides or forum junkies talk about this incessantly. Asus uses a 1-7 scale for my board. I use 6 when clocked to 90-95% of maximum capability. If I were lower I would use 5. The problem with the Asus one is it spends all its time on 1 or 7/max. Synthetic load? Instant max. This usually shifts the voltage to some degree.

 

I don't know about negative offset vs specifically enumerated adaptive voltage with auto offset (that is how it works on Asus). So rather than guess what I think adaptive is and deduct the offset, you set the maximum and it stays there while following the curve for submaximal frequencies. Negative offsets can be difficult and require very precise tuning. Really easy to lock up/crash as lower voltages in seemingly low idle states.

 

I will try and run a blend test with fans off and see what happens.

 

I have tried with LLC set to Normal (I think this would be level 1 or 2) but others have informed me that I should keep LLC on auto when not overclocking. My negative offset of -0.080 has not crashed or locked up. Under gaming it's fine, running 15 Cinebench R15 runs in a row is fine, Aida64 ran fine for about 10 minutes before liquid temperatures got uncomfortably hot, and normal use is fine. I'm so lost at this point and frustrated. I don't know if it's the cooler, the cpu itself, or some bios settings. I have a very strong feeling it is the cooler because I have a friend with a Hyper 212 Evo on a 8700K at 1.39 volts and THAT doesnt exceed 90C. Whereas this H115i at 1.224 volts SHOOTS to 90 and slowly rises to 92/93 within seconds. Maybe because my previous cpu was a 7600K and didnt develop near the heat of a 9700K, is why I never noticed how much the H115i was struggling. I think the pump is not flowing liquid at 3200RPM and iCue is giving a false reading.

Link to comment
Share on other sites

I have a dual 280mm custom loop in place right now, so I cut the fans on one rad and fixed the others at 650 to try and approximate a single 280mm loop. Coolant rise for 20 minutes was +2.9C peak. My CPU temps doing the AIDA sine wave dance between 40-60C. The AIDA blend test just isn't that heavy in watts or temperature. I don't understand what you are seeing.

 

Make sure you are not running AIDA and iCUE at the same time. It is possible to configure AIDA not to interact with iCUE, but when you see improbable numbers, there is a chance they are improbable. To that end, whenever you get unexpected data back from once source, get more from somewhere else. Try another stress test. OCCT would a fair wattage equivalent, but since you seem to having difficulties, try the basic stress test from CPU-Z, bench tab. It is a linear load and relatively light. Intel XTU is another possibility and slightly lighter than AIDA, but also with the wave loading. A bad cooler is bad all the time. It doesn't care what you are doing or what program is in place. Idle, load, and everywhere in between. When you see selective bad results, it likely is a selective setting we are looking for.

Link to comment
Share on other sites

I have a dual 280mm custom loop in place right now, so I cut the fans on one rad and fixed the others at 650 to try and approximate a single 280mm loop. Coolant rise for 20 minutes was +2.9C peak. My CPU temps doing the AIDA sine wave dance between 40-60C. The AIDA blend test just isn't that heavy in watts or temperature. I don't understand what you are seeing.

 

Make sure you are not running AIDA and iCUE at the same time. It is possible to configure AIDA not to interact with iCUE, but when you see improbable numbers, there is a chance they are improbable. To that end, whenever you get unexpected data back from once source, get more from somewhere else. Try another stress test. OCCT would a fair wattage equivalent, but since you seem to having difficulties, try the basic stress test from CPU-Z, bench tab. It is a linear load and relatively light. Intel XTU is another possibility and slightly lighter than AIDA, but also with the wave loading. A bad cooler is bad all the time. It doesn't care what you are doing or what program is in place. Idle, load, and everywhere in between. When you see selective bad results, it likely is a selective setting we are looking for.

 

First of all, that's a pretty dope custom loop you got.

 

I'll try out intel XTU and CPU-Z stress tests tomorrow, however at this point I've just been so frustrated with AIO's and their minimal cooling differences between high end air coolers. I've pretty much made up my mind to go D15. It's not that I refuse to figure out what exactly the problem is with the H115i or a software setting, but I just want the added reliability, and lack of software you get with an air cooler. I dont want to worry about liquid temps and failures and all. So even if I do find the kink here, I'll still be going the air cooling route. However, I will let you know the results of the other stress tests as soon I do them.

Link to comment
Share on other sites

OK. I think it is a natural part of the process to go back to air at some point and test whether you really want the water cooling or not. However, I am concerned you are going to be looking at exactly the same issue on the air tower, possibly with less room for error. I suspect temperature is not the only factor in your decision, so perhaps that does matter quite so much. Besides, if it is not a cooler issue, then the troubleshooting is the same with either air or water.
Link to comment
Share on other sites

  • 3 weeks later...
OK. I think it is a natural part of the process to go back to air at some point and test whether you really want the water cooling or not. However, I am concerned you are going to be looking at exactly the same issue on the air tower, possibly with less room for error. I suspect temperature is not the only factor in your decision, so perhaps that does matter quite so much. Besides, if it is not a cooler issue, then the troubleshooting is the same with either air or water.

 

This thread has become old I understand. I just wanted to let you know that I swapped out my H115i for the NH-D15 chromax.black and here are my results.

H115i Extreme profile: 92 degrees in Cinebench R15 after first run

D15 Stock Bios profile: 56 degrees MAX after 10 consecutive runs.

 

Definitely was something wrong with my H11t5i unfortunately.

Link to comment
Share on other sites

×
×
  • Create New...