Jump to content
Corsair Community

Threadripper 39xx + H150i PRO XT - What are your temps?


onyx64

Recommended Posts

Just installed the H150i PRO XT on the threadripper 3960x, and I am seeing temperatures hitting the 95C limit when running prime95 (Small FFTs, 48 threads). Although the "square plate" of the cooler doesnt cover the entire IHS, I dont expect it to hit the thermal limit, or reach near throttling (which is around 87) if the H150i is really meant for threadripper.

 

Anyone have a threadripper 39xx + H150i combo that see this?

Link to comment
Share on other sites

What are your voltages when running Prime95?

Have you configured PBO? Reset the thermal limit?

Also, what are is the temperature of the coolant? Ambient temp?

Typically, Ryzen processors will downclock before it hits that temp limit. However, that can be overridden.

Prime95 is brutal on CPUs and not a realistic benchmark that approximates to anything close to real world usage. It's notorious for overheating all CPUs (including Intel).

Link to comment
Share on other sites

What are your voltages when running Prime95?

Have you configured PBO? Reset the thermal limit?

Also, what are is the temperature of the coolant? Ambient temp?

Typically, Ryzen processors will downclock before it hits that temp limit. However, that can be overridden.

Prime95 is brutal on CPUs and not a realistic benchmark that approximates to anything close to real world usage. It's notorious for overheating all CPUs (including Intel).

 

- No overclock. BIOS is set to defaults.

- PBO is again the default on auto

- Temperature of the coolant is always sub 40C. I wonder if the sensor is working at all

- ambient is 23C, and air going into the rad is the same.

- Yes i know CPU can throttle upon nearing thermal limit. But the whole point of having an efficient cooler is to not let it reach that state.

- Prime95 is the ultimate heat test - similar to burn-in tests. It is a test that tells whether the AIO is capable of dissipating all the heat produced when the chip is working at TDP.

 

This is not a question of intel vs AMD. This is a question of how good the AIO is for a CPU. All said and done, the H150i does not cover the entire IHS. It should NOT be advertised as an AIO cooler that supports threadripper.

Link to comment
Share on other sites

Well Prime95 is a heat test for sure, but it produces way too much heat, more than you'd see in the harshest real world usage. All it does is to show you how your cooler reacts to Prime95..

I bet even on custom loop you'd struggle to keep it under TjMax

 

That said, there's no silicon under the edges of the IHS in the 3960x. they are centered under it in a rough square pattern, so, full coverage isn't that important for that particular threadripper.

 

Coolant at 40° makes sense. Prime stresses the chips so hard that the heat transfer starts to be the bottleneck rather than the cooler efficiency. in such extreme cases, you may see a bit of improvement by using high grade thermal pastes like kryonaut or conductonaut, but again, it will be only a few degrees.

 

What happens when you run a blender benchmark, or a looped Cinebench?

 

 

Edit : just for example, cinebench R20 makes my 10900k pump 200w continuous at roughly 70°C . Prime95 small FFT 275w @ 84°C, on custom loop...

in heavy use i rarely see the CPU go above 60.. That's just to tell how irrelevant it is to real use. It's good to test how your radiators behave when heat soaked maybe but that's about it.

Edited by LeDoyen
Link to comment
Share on other sites

- Prime95 is the ultimate heat test - similar to burn-in tests. It is a test that tells whether the AIO is capable of dissipating all the heat produced when the chip is working at TDP.

 

No it doesn't. If you are running small FFT on auto settings you are testing the physical material of the CPU and heat transference from CPU to cold plate. It is likely you hit that 95C instantly and the cooler has little to say about that.

 

A test of the cooling system's ability to dissipate heat is you start Test A and the CPU temp is 70C and 30 minutes later it's 73C (excellent) or you start at 70C and 30 minutes later its 95C (poor). The CPU cooler prevents (or tries to prevent) the CPU temperature from getting worse. It can't prevent the heat from passing through the CPU in the first place. You can also measure this in terms of coolant temperature rise (H150i Temp). The 30 minutes is just an arbitrary number, but heat dissipation involves some element of time elapsed in relation to watts in and watts out.

 

The voltage still matters because at your auto settings, you can put 10 different 3960x CPUs in and get 10 different auto voltages. It will vary between different tests as well. Without the voltage to give the CPU temp some context, it's hard to say if you have a weak CPU, an over-aggressive motherboard, poor cooling, or anything else. Unless you really intend to hunt Mersenne Prime numbers with your machine, what Prime 95 small FFT is really good for is quickly binning a bunch of processors. 30 seconds on and next one up. That's not a test of the cooling system. It's a test of the CPU. Whether you can run the actual program you need for the length of time you need it without added temperature penalty is a test of the cooling system.

Link to comment
Share on other sites

No it doesn't. If you are running small FFT on auto settings you are testing the physical material of the CPU and heat transference from CPU to cold plate. It is likely you hit that 95C instantly and the cooler has little to say about that.

 

A test of the cooling system's ability to dissipate heat is you start Test A and the CPU temp is 70C and 30 minutes later it's 73C (excellent) or you start at 70C and 30 minutes later its 95C (poor). The CPU cooler prevents (or tries to prevent) the CPU temperature from getting worse. It can't prevent the heat from passing through the CPU in the first place. You can also measure this in terms of coolant temperature rise (H150i Temp). The 30 minutes is just an arbitrary number, but heat dissipation involves some element of time elapsed in relation to watts in and watts out.

 

The voltage still matters because at your auto settings, you can put 10 different 3960x CPUs in and get 10 different auto voltages. It will vary between different tests as well. Without the voltage to give the CPU temp some context, it's hard to say if you have a weak CPU, an over-aggressive motherboard, poor cooling, or anything else. Unless you really intend to hunt Mersenne Prime numbers with your machine, what Prime 95 small FFT is really good for is quickly binning a bunch of processors. 30 seconds on and next one up. That's not a test of the cooling system. It's a test of the CPU. Whether you can run the actual program you need for the length of time you need it without added temperature penalty is a test of the cooling system.

 

From what you're saying it looks like you measure the efficiency of the cooling system from the cold plate on to the radiator and out. A good cooling system also takes into account the coverage of the cold plate on to the IHS. That would be the difference between prime95 shooting upto 95 or something lower, like 85, for eg. all else being equal. Clearly the bottleneck is the heat transfer between the IHS and the coldplate ;)

Link to comment
Share on other sites

The bottleneck is the CPU material itself. Conductivity is part of the process, but you don't get dramatically different results between this piece of metal or that. Internet AIO testing gets a bit convoluted since many reviews don't standardize settings and they certainly aren't trying to assess the cold plate functionality. However, if you dig up any review of a custom CPU water block, you will see more scientific merit and also that first to worst is usually in the neighborhood of 2C. I assume that is not the difference you are looking for and representative of how hard it is to make gains in that area. On the other hand, +-0.05v could be 10C apart depending on how close you are to the end of the voltage/temp curve.
Link to comment
Share on other sites

×
×
  • Create New...