Jump to content
Corsair Community

2950X cooled by H115i, temps too high?


Recommended Posts

I think my temps are too high for my AMD Ryzen Threadripper 2950X not overclocked and running SETI on all 16 cores (i.e. CPU at 100%). I realize actual (Tdie) max temp for Ryzens is 68C. I have taken a screen shot of several programs running: CPU-Z to show CPU is not overclocked; iCUE to show that the H115i (actual AIO cooler type is in system stats) is maxxed out in fan & pump speed; HWiNFO showing both die temps (graphs showing idle temps (no SETI) and then 100% load temps (SETI on)). Of course room temp is normal, ~22C.

 

attachment.php?attachmentid=36918&stc=1&d=1573696770

 

I had a system installer do the install and except for the H115i not covering the entire Threadripper CPU case (as expected), all mounts and thermal pastes were applied correctly.

 

attachment.php?attachmentid=36920&stc=1&d=1573697870

 

Any ideas? :sigh!:

 

PS. Corsair sells coolers but there is no entry in the system stats for a cooler description? I placed mine as the second video card so it could be seen... :bigeyes:

178017646_Captureofhot2950Xstats.thumb.PNG.05145fc3760cce14bc9552771ba98761.PNG

2085509648_2019computerwithflash_sm.thumb.jpg.3ea7f8c97e5afa29995ac3130fda05f1.jpg

Link to comment
Share on other sites

A coolant temp of 26C is pretty much normal idle for a 22C room (and certainly with a 16 core CPU). The cooler is working or that 26C would keep ticking upwards into the 50s and 60s.

 

The next thing to look at is if heat is actually being transferred. Since that is all conductive between the cold plate and CPU, a gap there (even very slight) ruins the thermal transfer. CPUZ is not a strenuous test and I don't think you should run one of those until this is settled. However, you should get come coolant temp creep upwards. At 150W, I would expect a coolant increase of approximately 6C in 10 minutes. One way to assess this is the lock the fans at a nominal level (say 700 rpm) and then initiate CPU-Z Bench/Stress test. It is a linear load, so the CPU temp should hold even, except when coolant temp increases. +1C coolant temp = +1 CPU temp. Let it run for 5-10 min and see if the coolant goes up. If you see the coolant hold steady by the CPU temp keeps going up, stop immediately. That indicates either a contact problem or some bizarre power loading issue. Other signs of poor thermal contact are very jump idle temps or elevated idle temps (45-55C, etc), but a lot of CPUs are pretty jumpy these days so it can be hard to tell.

 

Unfortunately, I don't have any data for expected coolant to CPU package temp differential. Most CPUs are 35-50C above the coolant when at 100%, but this is massively affected by voltage. I need someone with more Ryzen experience to comment on whether 1.395v is an appropriate level for your stock set-up. Unmodified Auto voltage is usually pretty heavy handed.

Edited by c-attack
Link to comment
Share on other sites

A coolant temp of 26C is pretty much normal idle for a 22C room (and certainly with a 16 core CPU). The cooler is working or that 26C would keep ticking upwards into the 50s and 60s.
Understand the coolant temp is only 26C but that is not the CPU temp of which for a 2950X it has two Tdie temps (actual temperature of each die in a Threadripper) and I show them to be around 68C which is the maximum AMD recommends for a Threadripper. These Tdie temps are what I am asking about, if they are too high for a flat maxxed out H115i.

 

The next thing to look at is if heat is actually being transferred. Since that is all conductive between the cold plate and CPU, a gap there (even very slight) ruins the thermal transfer. CPUZ is not a strenuous test and I don't think you should run one of those until this is settled.
Actually I mentioned the CPU load is 100% by SETI, not CPUZ. The CPUZ screen is just to show I am not overclocking.

 

However, you should get come coolant temp creep upwards. At 150W, I would expect a coolant increase of approximately 6C in 10 minutes. One way to assess this is the lock the fans at a nominal level (say 700 rpm) and then initiate CPU-Z Bench/Stress test. It is a linear load, so the CPU temp should hold even, except when coolant temp increases. +1C coolant temp = +1 CPU temp. Let it run for 5-10 min and see if the coolant goes up. If you see the coolant hold steady by the CPU temp keeps going up, stop immediately. That indicates either a contact problem or some bizarre power loading issue. Other signs of poor thermal contact are very jump idle temps or elevated idle temps (45-55C, etc), but a lot of CPUs are pretty jumpy these days so it can be hard to tell.
So the coolant value has steadied out as you can see the Tdie temps have been showing in the graphs for about two minutes (26C is the steady state coolant temp when I let it sit for 10s of minutes at load). The H115i's fans are already maxxed out by the display of iCUE showing the extreme profile which is maxxed out fan and pump speed (which minimizes the time to achieve a steady state coolant temp as well).

 

Unfortunately, I don't have any data for expected coolant to CPU package temp differential. Most CPUs are 35-50C above the coolant when at 100%, but this is massively affected by voltage.
I have a H100i running for 6 years (continuously due to SETI) on an 8 core AMD FX 8350 and never saw 50C deltas, more like 20C.

 

I need someone with more Ryzen experience to comment on whether 1.395v is an appropriate level for your stock set-up. Unmodified Auto voltage is usually pretty heavy handed.
Since I do not overclock I stay at stock for all settings, as you noted.
Link to comment
Share on other sites

That max temp specified is for TCase (on top of the CPU, the IHC), not TJunction/TDie.

TCtl is a manufactured value intended to ensure consistent fan curves.

 

The primary temperature reporting sensor of the AMD Ryzen™ processor is a sensor called “T Control,” or tCTL for short. The tCTL sensor is derived from the junction (Tj) temperature—the interface point between the die and heatspreader—but it may be offset on certain CPU models so that all models on the AM4 Platform have the same maximum tCTL value. This approach ensures that all AMD Ryzen™ processors have a consistent fan policy.

 

Typical max vCore on a Ryzen/Threadripper is 1.4V.

 

I don't see anything of concern here. The H115i can handle the 2950X pretty well and, with PBO enabled, it'll push voltages into the 1.45-1.47 range (HydroX will allow it to push up to 1.5V or so). The key thing is to ensure a supply of cool air to the radiator. That's dependent on your installation more than anything else.

 

And .. I ran my 2950X on an H115i Platinum for some time before switching over to HydroX.

Link to comment
Share on other sites

Earendil3, Those readings seem a bit high. My 2920X tops around 50C with a Cooler Master ML360. What I get are crazy high readings with some temp monitors. On the Bios and CPUID HWMonitor the readings are correct showing <30C in idle but on Corsair link and MSI Afterburner I get >68C also in idle and well over 100 on burn. Could it be that the temps reading is not correct?
Link to comment
Share on other sites

Understand the coolant temp is only 26C but that is not the CPU temp of which for a 2950X it has two Tdie temps (actual temperature of each die in a Threadripper) and I show them to be around 68C which is the maximum AMD recommends for a Threadripper. These Tdie temps are what I am asking about, if they are too high for a flat maxxed out H115i....

 

You've completed missed the point at every turn. You could have completed a genuine test with actual relevant reference points in the amount of time it took to write that lengthy response. The cooler is working. Whether you have an installation issue, a voltage management problem, or a testing protocol problem is still unresolved. We all can see your CPU temps. That doesn't reveal the underlying reason.

Link to comment
Share on other sites

That max temp specified is for TCase (on top of the CPU, the IHC), not TJunction/TDie.

TCtl is a manufactured value intended to ensure consistent fan curves.

Understand about Tctl and thus I am not looking at it nor is that the temp I am graphing. But you mention a Tcase temp and I thought that Tdie is the closest. There are m/b temps and one identified by HWiNFO64 as the "CPU temp" (not the "CPU socket temp") which is also high and comparable to Tdie temps. Maybe that is Tcase or a close analog.

 

Typical max vCore on a Ryzen/Threadripper is 1.4V.

 

I don't see anything of concern here. The H115i can handle the 2950X pretty well and, with PBO enabled, it'll push voltages into the 1.45-1.47 range (HydroX will allow it to push up to 1.5V or so). The key thing is to ensure a supply of cool air to the radiator. That's dependent on your installation more than anything else.

 

And .. I ran my 2950X on an H115i Platinum for some time before switching over to HydroX.

I have to respectfully deny that my H115i is cooling my 2950X well enough if the Tdie and CPU temps are >= 68C. Everything is stock and I am in a chilly basement at ~22C.

 

Earendil3, Those readings seem a bit high. My 2920X tops around 50C with a Cooler Master ML360. What I get are crazy high readings with some temp monitors. On the Bios and CPUID HWMonitor the readings are correct showing <30C in idle but on Corsair link and MSI Afterburner I get >68C also in idle and well over 100 on burn. Could it be that the temps reading is not correct?
I wish I had 50C. I Googled around and HWiNFO64 was mentioned the most as the most accurate. iCUE appears to give similar, correct temps. I tried to install Asus Armoury Crate and it wouldn't work, wouldn't update and took hours to delete - 3 hrs of my life wasted. I recently started getting ANOTHER issue (extreme and fixed profiles are now not consistant) with the H115i which I'll be starting another thread for. Sigh! :[pouts:
Link to comment
Share on other sites

Understand about Tctl and thus I am not looking at it nor is that the temp I am graphing. But you mention a Tcase temp and I thought that Tdie is the closest. There are m/b temps and one identified by HWiNFO64 as the "CPU temp" (not the "CPU socket temp") which is also high and comparable to Tdie temps. Maybe that is Tcase or a close analog.

The max temp that AMD publishes is the tCase temp, NOT tDie. So that 68C max that you mentioned? That's for tCase. That's the temperature of the heatspreader. Which is pretty difficult to directly measure (there aren't any sensors on it) and there is no close analog that you can measure. Which means ... it's essentially useless as a spec temp.

 

I have to respectfully deny that my H115i is cooling my 2950X well enough if the Tdie and CPU temps are >= 68C. Everything is stock and I am in a chilly basement at ~22C.

Ummm ... OK. Well ... you'll get better temps with a custom loop. The 2950X does run warm. Here's the thing - AMD's throttling mechanism (even on Auto) is highly dependent on your cooler and temperatures. It actually does a really, really good job ... manual overclocks typically don't get much improvement over what the processor does itself, especially with PBO enabled. So a good way to gauge the effectiveness of your ThreadRipper cooling solution is to enable PBO and let it loose. Now, I did all of this with my H115i Platinum that I had on my 2950X for about six months before switching to a custom loop with HydroX. HydroX is definitely better - no question about that at all. But the H115i did a pretty respectable job. And you cannot directly compare the 2920X and the 2950X. That's like comparing thermals on an i3 to thermals on an i7.

And, while a chilly basement certainly helps, it's not always that simple. We don't know the configuration of your system or how the heat flow inside the system is going, we don't know if your radiator is configured for intake (ideal) or exhaust (less ideal). If exhaust, then you could easily (and quickly) run into issues with GPU and VRM heat flowing into the radiator. This would be significantly warmer than your 22C basement. And VRM heat with the ThreadRipper boards can actually be pretty significant.

 

I wish I had 50C. I Googled around and HWiNFO64 was mentioned the most as the most accurate. iCUE appears to give similar, correct temps. I tried to install Asus Armoury Crate and it wouldn't work, wouldn't update and took hours to delete - 3 hrs of my life wasted. I recently started getting ANOTHER issue (extreme and fixed profiles are now not consistant) with the H115i which I'll be starting another thread for. Sigh! :[pouts:

Be careful running HWINFO and iCUE at the same time. There are known issues with HWINFO interfering with iCUE devices. You need to uninstall the persistent driver and make sure that Corsair Link/Asetek support is disabled.

Link to comment
Share on other sites

OK, some of the things you are mentioning are making sense since a I have had an issue with an iCUE firmware update (posted here). For a few days my H115i has run flat out (fan noise is the most noticeable thing) and my system temps have moved down a little. Tdie on CPU node #0 is most comparable to the m/b CPU temp (a sensor on the case) and it remains mostly below 65C instead of above it like before. So I take back the claim that the H115i cannot keep the 2950X cool enough. :o:

 

Yes, I am bucking your convention of assuming Tdie node #0 & #1 and Tcpu from the m/b is not exactly Tcase but it's all I have. I have read about Tctl and I neglect it because of artificial inflation, as a test engineer I deal in data as real as possible. :nodding:

 

Sorry, I simplified things mentioning a cooler basement but the other things you mentioned apply as well. In my situation my EATX-fitting Rise computer case has lots of room for air flow and lots of fans (three in the front [inward], two in the back including the PSU one [outward] and two in the top thanks to the H115i [outward]). The Rise sits on a porcelain tile covered cement floor thus where the "cool" comes from. Just sayin'.

 

I read up on the PBO or more specifically PB2 on the 2950X. I restarted and enabled it in the BIOS. Besides my issue with H115i (I think it's bricked) at least it fails full on and thus provides maximum cooling. As my previous pics show I have those weird spikes on Tdie but with PBO enabled they are completely removed from the HWiNFO64 graphs. Since the BIOS was initially set to Auto I would guess PBO was starting and stopping but now it is enabled the temps are more stable - over 10 mins: Tdie node #0 67.5C (steady), Tdie node #1 ~67C (fluctuates a little) and m/b Tcpu at 67C (steady). CPU-Z says core speed is jumping around like a cat on a hot tin roof but then that sounds like PBO is working all the time so it's fine.

 

So now it's on to customer support... :(:

Link to comment
Share on other sites

  • 2 weeks later...

OK, I ended up posting here and eventually received a fix for the "bricked" H115i I thought I had. Thanks Corsair!

 

I also had trouble setting pump and fan speeds, they would never stay where I put them. After much clicking I now get it to stay constant but I still have to reset it after every reboot. I created a custom filter called "Max" under the "Performance" sub-menu, selected "Custom curve" and selected "2950X Package #1" (otherwise known as Tdie node#0, the hottest) as the temp sensor. I reboot and it went back to the "Extreme" filter so I had to select Max again and set up the "Custom curve" again.

 

Another thing is only the three filters "Quiet", "Balanced" and "Extreme" can set pump speed. I wonder why? At least that always stays at "Extreme".

 

So I set the "Custom curve" for max fan speed under all temps and exited iCUE. I ran HWiNFO64 all night with extensive graphing at max load (SETI on all cores). The temps are similar to when I thought the H115i was bricked and running flat out (see above). I am disappointed the CPU temps are so close to 68C. I may follow DevBiker's lead and look at the HydroX solution.

 

Another thing was the recent announcement of the 3950X. I may just let the 2950X burn itself out and thus I have an excuse to get a 3950X which has the same amount of cores, roughly the same core speeds and uses less power. Then I'd have the same as before but the H115i could keep up hoping the CPU wouldn't get so hot. Oh well. :rolleyes:

Link to comment
Share on other sites

The Pro and Platinum pumps are 3 fixed speeds. Pump speed is not overly relevant for AIO cooling with short travel distance and low to moderate resistance. The Quiet, Balanced, and Extreme "presets" that have an entirely different meaning for the fans, do double duty as Low, Medium, High in this case.

 

I suspect your fan profile may be resetting because it can't figure out why you are using Package Temp for the control variable. That is a motherboard based value. Again, not how the cooler works and you are spinning your wheels for nothing. If you need to run your program and are worried about the maximum temp, set the fans to the highest level you can stand and leave them there. You are still missing the connection between coolant temperature, the CPU package load temp, and the fans/pump. The cooler's ability to get rid of heat is not the issue if you hit your 68C mark the moment you start the program.

Link to comment
Share on other sites

  • 1 year later...

So, a year and a half later, I've done a lot of research (including this forum) and came to the conclusion:

 

The H115i does not adequately cool any ThreadRipper, because it doesn't cover the full heat spreader (or even the full die area.) It's that simple.

 

I get coolant temp up to about 50 or so, at which point the CPU hits 68C (Tdie,) at which point the CPU starts scaling back frequency to not overheat itself, and I do *not* hit max clocks at max loads with a H115i AIO cooler. The input air temperature to the fans is ambient; 21 C. The fans are running at max RPM, as is the pump. The mount to the CPU is solid and with a thin even spread of good quality paste.

 

With Threadrippers, Tdie is the *only* thing you need to worry about, and 68C is the magic limit for this. If you enable entirely custom voltage/power/heat curves in your BIOS, you can push it above this, which will shorten the life of the CPU. I have not found a way remove the 68C throttling limit on my ASUS BIOS, other than making custom P-state settings.

 

Earendil3 and c-attack and DevBiker are probably all gone, but if anyone else tries to use the Bigger Brain to look up more info about this, this is as far as I've gotten.

 

Summary: You need a Threadripper-sized cold plate / "CPU block" for Threadrippers, and the only sensor value that matters is Tdie, which the CPU uses to manage itself, and it will lower its clock frequency once it hits 68C.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...