RealBench temperature way too high

[Apollo190]

I'm still testing my new setup to see how low I can undervolt while stable in prime95, realbench, etc...

My setup:

• CPU: i9-10900K
• Cooler: bequiet Dark Rock Pro 4 (with both fans installed)
• Thermal paste: Thermal Grizzly Kryonaut
• GPU: Asus TUF RTX 3080 OC
• RAM: 2x16GB 3600MHz HyperX Fury
• Motherboard: Asus TUF Z490-PLUS
• Case: Asus TUF GT501 (3 intake fans at the front panel, 1 exhaust fan on the back, and perforations are all around the case)

BIOS settings:

• XMP: XMP I (DDR4-3603 17-21-21-39-1.35V)
• ASUS MultiCore Enhancement: Enabled - Remove all limits
• SVID Behavior: Auto
• AVX Instruction Core Ratio Negative Offset: 0
• CPU Core Ratio: Auto
• Power-saving & Performance Mode: Auto
• CPU Load-line Calibration: Level 4 - Recommended for OC
• BCLK Aware Adaptive Voltage: Enabled
• CPU Core/Cache Voltage: Manual Mode
• CPU Core Voltage Override: 1.365 V
• DRAM Voltage: 1.35000 V
• CPU Core/Cache Current Limit Max: 255.75 A
• Long Duration Package Power Limit: 4095 Watt
• Package Power Time Window: 56 Sec
• Short Duration Package Power Limit: 4095 Watt
• Intel SpeedStep: Enabled
• Intel Speed Shift Technology: Enabled
• Intel Turbo Boost Max Technology 3.0: Enabled
• CPU C-states: Enabled

I'm still not 100% sure if the 1.365V is stable, but 1.36 resulted in BSOD in prime95, even though it survived realbench and blender. I'll have to test more.

However, I couldn't believe how high the realbench pushes the temperature, especially considering I used to run realbench for 6 hours on my gaming laptop (i7-4720hq|gtx 980m) and temps were never even close to this. From what I've seen on many forums, prime95 is considered to be the hardest cpu stress test, but I had no issues with thermals when running prime95 blend.

RealBench results after 5 minutes of stress testing (Room temperature: 25C):

 

Prime95 settings & results after 5 minutes of stress testing (Room temperature: 25C):

 

Blender Classroom benchmark results (Room temperature: 25C):

 

CPU-Z Benchmark results (Room temperature: 25C):

 

As you can see, cpu was running at 4.9GHz on all cores. Blender was running a bit hotter than prime95, but still acceptable, unlike realbench. I didn't feel comfortable leaving realbench running any longer than 5 minutes.Is this normal and is there anything I can do about ithinking about reapplying the thermal paste, but I'm pretty sure I did a good job last time around.

[Bombastinator]

*attempts to find the temperature levels complained about*
*fails*
 

[ShrimpBrime]

It's a 10 core 20 thread 14nm monster on 4.9ghz OC probably getting real close to 300w.

 

Yeah, it's gonna run hot with a Dark Rock 4. 

 

[SAVE-12-HK]

96c peak? looks normal on intel

how often will all cores at load?daily use will just use 20~30%

[Apollo190]

5 hours ago, ShrimpBrime said:

It's a 10 core 20 thread 14nm monster on 4.9ghz OC probably getting real close to 300w.

 

Yeah, it's gonna run hot with a Dark Rock 4. 

 

The peak was 209.415W actually, you can see it in the blender screenshot, and dark rock pro 4 claims "250w tdp" whatever they mean by that. And isn't 4.9GHz the stock boost for all cores? (with thermal velocity boost though, I just removed the temperature requirement for the boost, which is 70C on stock I believe)

Anyways, how am i supposed to test my overclocks and undervolts if I can't run stress tests longer than 5 minutes (unless I want to fry the cpu)?
EDIT: forgot to mention that fan curves were stock, meaning they reach 100% rpm at 70C.

[Bombastinator]

56 minutes ago, Apollo190 said:

The peak was 209.415W actually, you can see it in the blender screenshot, and dark rock pro 4 claims "250w tdp" whatever they mean by that. And isn't 4.9GHz the stock boost for all cores? (with thermal velocity boost though, I just removed the temperature requirement for the boost, which is 70C on stock I believe)

Anyways, how am i supposed to test my overclocks and undervolts if I can't run stress tests longer than 5 minutes (unless I want to fry the cpu)?
EDIT: forgot to mention that fan curves were stock, meaning they reach 100% rpm at 70C.

You can’t fry the cpu unless you turn off a bunch of stuff designed to shut the thing down after it reaches critical temp.  You could make it shut down I suppose by interfering with temp throttling. If stuff isn’t messed with though the cpu will reach throttle temp (95°c? I forget) and then just run slower to maintain just below the throttle temp.

[Apollo190]

55 minutes ago, Bombastinator said:

You can’t fry the cpu unless you turn off a bunch of stuff designed to shut the thing down after it reaches critical temp.  You could make it shut down I suppose by interfering with temp throttling. If stuff isn’t messed with though the cpu will reach throttle temp (95°c? I forget) and then just run slower to maintain just below the throttle temp.

Mine didn't throttle even when some cores reached 99°c, not sure which setting in bios could prevent the throttle. But still, if it throttles during stress test, I can't conclude if my oc is stable.

[ShrimpBrime]

3 hours ago, Apollo190 said:

The peak was 209.415W actually, you can see it in the blender screenshot, and dark rock pro 4 claims "250w tdp" whatever they mean by that. And isn't 4.9GHz the stock boost for all cores? (with thermal velocity boost though, I just removed the temperature requirement for the boost, which is 70C on stock I believe)

Anyways, how am i supposed to test my overclocks and undervolts if I can't run stress tests longer than 5 minutes (unless I want to fry the cpu)?
EDIT: forgot to mention that fan curves were stock, meaning they reach 100% rpm at 70C.

Wattage used and BTU produced are kinda different things..... But relative to each other.

 

This rig would be better suited with a nice custom loop for overclocking head room. IMO.

[Bombastinator]

2 hours ago, Apollo190 said:

Mine didn't throttle even when some cores reached 99°c, not sure which setting in bios could prevent the throttle. But still, if it throttles during stress test, I can't conclude if my oc is stable.

Part of the problem is there are levels of stability.  I OCed a 4770k and I could make it run for an hour or two on furmark and prime95 (what were all that was available to me at the time) with no crashes at 4.5ghz.  So by one consideration level stable.  The problem was it would still crash occasionally playing games.  To get really no crashes ever I had to go down to 4.0ghz.  Took a few months of play to get it there and confirm it.  It’s a chip that hasn’t been made in years though and was a weird and not very good example of it in the first place, so it may not make a very good example.  You’re not trying to OC a kind of lead 4770k.

Edited November 28, 2020 by Bombastinator

[Apollo190]

On 11/28/2020 at 5:21 PM, ShrimpBrime said:

This rig would be better suited with a nice custom loop for overclocking head room. IMO.

Which one would you recommend?

[Apollo190]

Since my current air cooler, dark rock pro 4, isn't able to support anything over 4.9GHz in heavy workloads on my 10900k, I'm now considering the liquid cooling. It hits 99C after 5 minutes of asus realbench stress test, but it's better in games. In battlefield 1 for example, it hits ~80C.

I've never had a liquid cooler before, so I have a couple of quesitons:

1. How much thermal difference can I expect from 360mm aio cooler compared to the one I have now?

2. How much more could I overclock with a 360mm aio cooler, and would the performance difference be worth it?

3. How much louder would the 360mm aio cooler be compared to dark rock pro 4?

4. Which 360mm aio cooler is the best right now? My priorities are performance, noise level, price, in that order. From what I've seen on reviews so far, I like the nzxt kraken z73 (especially the lcd display) and arctic liquid freezer 2. But I know very little about them, so feel free to share your experience, opinions, etc...

5. Would I have to replace the stock fans with better ones? If so, which fans should I get?

6. I also read good things about IceGiant ProSiphon Elite, but there's very little information about it online. Is it even out yet? Or is it still at prototype stage? Is it better than 360mm aio coolers? Any information would be appreciated.

[Letgomyleghoe.]

Before spending your money on expensive coolers try remounting your cooler as a dark rock pro 4 should handle a 10900k pretty well. Buildzoid, a overclocker, managed to get 5.3 ghz on a 10900k with a 280mm aio which will preform about same as a dark rock pro 4 without hitting the thermal junction point.

[ShrimpBrime]

1 hour ago, Apollo190 said:

Which one would you recommend?

A few sites to look at:

https://www.frozencpu.com/cat/l1/g30/Liquid_Cooling.html?id=eb8ohoY4

https://www.ekwb.com/shop/kits

https://www.primochill.com/

https://www.xs-pc.com/

 

There's a lot of sites beyond this short list. 

Look at what you like and research the products from there.

These do have water loop kits and AIO as well. 

[Bombastinator]

One poster that knows a lot about water cooling product specifics is @For Science! he might have suggestions.  He isn’t around a whole ton though.  A reply might take time.

[Apollo190]

8 hours ago, Letgomyleghoe. said:

Before spending your money on expensive coolers try remounting your cooler as a dark rock pro 4 should handle a 10900k pretty well. Buildzoid, a overclocker, managed to get 5.3 ghz on a 10900k with a 280mm aio which will preform about same as a dark rock pro 4 without hitting the thermal junction point.

In gamer nexus review of 280mm arctic liquid freezer 2, it seems to be 5C cooler than nh-d15, and dark rock pro 4 is almost identical to nh-d15 I think. However, 5C wouldn't change anything in my case since I'd still be over 90C in realbench stress test. I'm kind of avoiding taking it apart since it's difficult to install dark rock pro 4, I needed someone to hold it in place while tightening the screws last time, in order to avoid coldplate sliding around and ruining the paste contact.

Edited November 30, 2020 by Apollo190

[For Science!]

10 hours ago, Apollo190 said:

Which one would you recommend?

Would setting an AVX negative offset resolve the issues here?  (Say -2)

[Apollo190]

22 minutes ago, For Science! said:

Would setting an AVX negative offset resolve the issues here?  (Say -2)

I'll give it a try and report back. Also, I noticed that prime95 gave bsod only when it switched to heavier test, and temps skyrocketed from 80C to 100C, so maybe voltage is actually stable, and it was thermals that caused bsod, if that's possible idk.

[Apollo190]

@For Science! Ok, I set an AVX negative offset to 2 and ran realbench again. Cores were now running at 4.7GHz of course. Temperatures are in the picture below. Still over 90C but much better than before, it actually managed to complete the full 15 minute stress test this time. Before avx offset it took only 5 minutes to reach 98-99C on some cores.

This still makes me unable to test my overclocks however, for example if I wanted to test a 5.0 or 5.1GHz overclock stability, it would run at 4.8-4.9Ghz in the stress test. And if I disable the avx offset, I'm out of thermal headroom for any kind of overclock .

[narrdarr]

What case are you using?

[For Science!]

47 minutes ago, Apollo190 said:

@For Science! Ok, I set an AVX negative offset to 2 and ran realbench again. Cores were now running at 4.7GHz of course. Temperatures are in the picture below. Still over 90C but much better than before, it actually managed to complete the full 15 minute stress test this time. Before avx offset it took only 5 minutes to reach 98-99C on some cores.

This still makes me unable to test my overclocks however, for example if I wanted to test a 5.0 or 5.1GHz overclock stability, it would run at 4.8-4.9Ghz in the stress test. And if I disable the avx offset, I'm out of thermal headroom for any kind of overclock .

I think the way to look at it is that you do not need a rock solid overclock for all operations. What this is telling you at the moment, is that with your silicon and motherboard combo, the current voltage setting is more suitable for something like a 4.7 GHz for a incredibly heavy AVX workload like RealBench. The same voltage may not required to hold the same frequency for a lighter load (e.g. Cinebench R15, or Gaming). So I think it is important to test for the load you are interested in, for example if you are interested in gaming, choose a handful of stressers that are games (e.g. TimeSpy) and dial in an overclock that you are satisfied with, then at the same voltage, find out how much you need to dial back the OC (with negative AVX offset) that works with RealBench.

[Apollo190]

2 hours ago, narrdarr said:

What case are you using?

Asus TUF GT501

[Apollo190]

1 hour ago, For Science! said:

I think the way to look at it is that you do not need a rock solid overclock for all operations. What this is telling you at the moment, is that with your silicon and motherboard combo, the current voltage setting is more suitable for something like a 4.7 GHz for a incredibly heavy AVX workload like RealBench. The same voltage may not required to hold the same frequency for a lighter load (e.g. Cinebench R15, or Gaming). So I think it is important to test for the load you are interested in, for example if you are interested in gaming, choose a handful of stressers that are games (e.g. TimeSpy) and dial in an overclock that you are satisfied with, then at the same voltage, find out how much you need to dial back the OC (with negative AVX offset) that works with RealBench.

That could work, thank you! I'll try it out that way.

What do you think about my cooler though? Is this normal for dark rock pro 4, or did i mess up with thermal paste application or something?

Would there be much difference with a 360mm aio?

[For Science!]

6 minutes ago, Apollo190 said:

That could work, thank you! I'll try it out that way.

What do you think about my cooler though? Is this normal for dark rock pro 4, or did i mess up with thermal paste application or something?

Would there be much difference with a 360mm aio?

Hard to know exactly, but I think its unlikely that the application/mounting is catastrophically bad, a 360 mm AIO may get you a couple degrees less, and also generally take longer to heat up (and also cool down). If you unmount the cooler and can see that most of the thermal paste is squished out and you only get lightning-like threads of thermal paste, then you can be sure the pressure was adequate. If its more like a blob in the middle, more pressure needs to be applied. 

 

 

I haven't worked with a 10900K personally, so cannot comment whether you temperatures are normal for a DRP4 or not. Usually this is more to do with voltage than anything else in my experience./. Another thing that just came to my mind as I was reading your post in a bit more detail, I would recommend you to consider using offset mode instead of manual, this won't change your OC stability, but may save you on power consumption and idle temps as you don't need to maintain 1.365 volts all the time.

 

 

[Apollo190]

1 hour ago, For Science! said:

a 360 mm AIO may get you a couple degrees less, and also generally take longer to heat up (and also cool down).

Now that you mention it, with drp4, as soon as I stop the stress test, temperature drops from ~95C to 50-60C range almost instantly. I found that a bit strange, hopefully such abrupt changes in temperature won't damage anything.

1 hour ago, For Science! said:

If you unmount the cooler and can see that most of the thermal paste is squished out and you only get lightning-like threads of thermal paste, then you can be sure the pressure was adequate. If its more like a blob in the middle, more pressure needs to be applied. 

 

Hmm, I was hoping I wouldn't have to go through installing this cooler again, but I guess I'll have to in order to find out .

1 hour ago, For Science! said:

I haven't worked with a 10900K personally, so cannot comment whether you temperatures are normal for a DRP4 or not. Usually this is more to do with voltage than anything else in my experience./. Another thing that just came to my mind as I was reading your post in a bit more detail, I would recommend you to consider using offset mode instead of manual, this won't change your OC stability, but may save you on power consumption and idle temps as you don't need to maintain 1.365 volts all the time.

I haven't tried the offset mode yet, it drops the voltage while idle, like auto mode? That would be great, I was actually worried about idling at constantly high voltage all the time.

I assume I'll have to add +60 mV since default seems to be 1.3v, at least that's what it shows in bios as current value on the left.

[For Science!]

8 minutes ago, Apollo190 said:

Now that you mention it, with drp4, as soon as I stop the stress test, temperature drops from ~95C to 50-60C range almost instantly. I found that a bit strange, hopefully such abrupt changes in temperature won't damage anything.

Hmm, I was hoping I wouldn't have to go through installing this cooler again, but I guess I'll have to in order to find out .

I haven't tried the offset mode yet, it drops the voltage while idle, like auto mode? That would be great, I was actually worried about idling at constantly high voltage all the time.

I assume I'll have to add +60 mV since default seems to be 1.3v, at least that's what it shows in bios as current value on the left.

I don't think you need to be worried about damage for the sudden change, its just what is being drawn or not. Yes, Offset mode is a bit more like the auto mode, I'd recommend setting it first and then finding out what you need to add (or start from like +10 mV, to be safe). I don't think it necessary starts from the same base line.