i9-10850k running hot even with 2 CPU and 4 case fans

maktos

n00b
Joined
Dec 26, 2020
Messages
4
I just built a new PC around the i9-10850K CPU. Motherboard is ASUS ROG MAXIMUS XII HERO Intel Z490.
Cooler is the famous EVO 212, single fan version, but with a 2nd Noctua fan added. Nice fan BTW for $20.

I have another (cheap) 120mm case fan right "upwind" of the CPU, on the back of the case sucking air in, blowing onto the CPU. Then one (cheap) fan above the CPU sucking air out. 2 more case fans in the front sucking air in, which came with the case.

Everything is running smoothly on the new rig, except for the CPU temps. I have installed Linux Mint 20 (my Linux distro of choice).
I use the program "sensors" to report current motherboard temperatures.

My CPU test of choice is to do some CPU mining. All 10 cores keep hitting 100C or close to it (it's clearly "auto throttling" to preserve the CPU which is inefficient for mining).
If I quit the CPU mining, the cores instantly drop to the 40's C, and then drift down to the 30's from there.

When I first tested this PC, I was getting poor mining performance -- on par with my 5-year-old i7-5820K. I went into the bios and basically followed the first few steps for overclocking: set MCE to enabled, the CPU to 4.9 GHz and turned on "Synch all Cores".
This was to get a baseline before doing any real overclocking -- to get a feel for what your rig's thermals are at the start.
By the "first few steps of overlocking", I'm referring to this article: https://www.overclockers.com/how-to-overclock-the-i9-10900k-a-guide-for-taming-the-beast/

Well, my thermals are pretty bad. It's already throttling the CPU as the cores continually hit 100C. Why? My cooler seems quite beefy and has 4 copper tubes, etc. -- it's a real classic for air cooling.

Is air cooling simply insufficient to get decent performance out of this CPU (assuming no overclocking)?

Here's my issue: my CPU mining performance is better than it was originally, but it's still significantly slower than my Ryzen server which, as you can see below, has only 72.8% the CPUmark score of this new rig. And call me crazy, but CPU mining should be a pretty good indicator or benchmark for real-world CPU performance in other software as well, right? (video editing, graphics, compiling software, creating ZIP archives, etc., all the things I do on a daily or weekly basis)

I'm not looking to overclock. I'd just like to get that 23,316 CPUmark performance, nothing more, nothing less. But so far I'm not seeing it. Is my CPU defective? Do I need to set something different in the BIOS?

My old main PC, built 5 years ago:
Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz
13,012 CPUmarks

Recent server build:
AMD RYZEN 7 2700X 8-Core 3.7 GHz
16,985 CPUmarks

Brand-New PC built 12/15/20:
i9-10850K
23,316 CPUmarks
 
Last edited:
am i ready that right, youre trying to hit 4.9GHz with a 212? thats not gonna work. set it back to stock, make sure your fans are push/pull and not mounted with both blowing into the heat sink and flip the back fan to exhaust.
 
am i ready that right, youre trying to hit 4.9GHz with a 212? thats not gonna work. set it back to stock, make sure your fans are push/pull and not mounted with both blowing into the heat sink and flip the back fan to exhaust.

Yes, that's right. As I said, I'm just after "decent" performance, which I define as "23,316 CPUmarks". I would like to think this CPU can deliver way more performance than my 5-year-old i7-5820, even with this 212 air cooler.

The CPU is currently push/pull. The "push" is the stock fan that came with the 212. The "pull" is the Noctua which is blowing much more air (out of the CPU, towards the blu-ray DVD drive).

My case has ventilation on the top of it -- I removed the silly "dust screen" to improve airflow. I have a cheap 120mm fan attached to the EXTERIOR of the case, sucking air out of the case and blowing it upward. I could do this with a second fan as well, if I wanted.

Before I set the CPU to 4.9GHz in the bios, the thing was staying at 800 MHz or something, giving me horrible CPU performance, even under load. I think the BIOS reported a minimum 800 MHz and a max 3200 MHz. But whatever it was set to, I was getting equal-to-worse performance compared to my i7-5820. That machine was built in Nov 2015 and it wasn't state-of-the-art even then. So "the same as my old PC" just isn't going to do it for me. Is this CPU *that bad* as far as overheating while not delivering the promised performance?

I wish my Ryzen 3900x would have worked out -- I kept getting freeze ups. But that thing really ran when it wasn't locking up. That CPU/MOBO was the same price, but 32K CPUmarks!
 
Last edited:
Yes, that's right. As I said, I'm just after "decent" performance, which I define as "23,316 CPUmarks". I would like to think this CPU can deliver way more performance than my 5-year-old i7-5820, even with this 212 air cooler.

The CPU is currently push/pull. The "push" is the stock fan that came with the 212. The "pull" is the Noctua which is blowing much more air (out of the CPU, towards the blu-ray DVD drive).

My case has ventilation on the top of it -- I removed the silly "dust screen" to improve airflow. I have a cheap 120mm fan attached to the EXTERIOR of the case, sucking air out of the case and blowing it upward. I could do this with a second fan as well, if I wanted.

Before I set the CPU to 4.9GHz in the bios, the thing was staying at 800 MHz or something, giving me horrible CPU performance, even under load. I think the BIOS reported a minimum 800 MHz and a max 3200 MHz. But whatever it was set to, I was getting equal-to-worse performance compared to my i7-5820. That machine was built in Nov 2015 and it wasn't state-of-the-art even then. So "the same as my old PC" just isn't going to do it for me. Is this CPU *that bad* as far as overheating while not delivering the promised performance?

I wish my Ryzen 3900x would have worked out -- I kept getting freeze ups. But that thing really ran when it wasn't locking up. That CPU/MOBO was the same price, but 32K CPUmarks!

I hardly know where to begin. First off, who bases their purchase decisions on Passmark CPU scores?

The CPU will report 800MHz or around there at idle. This is normal and this is fine. It will ramp up when needed. That's not what's impacting performance. The clock speeds are dynamic and change based on the workload. And using something like Passmark CPU marks to define performance is ridiculous. It's an arbitrary number. Intel hasn't come a long way in terms of IPC since then so you will only see a meaningful difference in performance in specific use case scenarios. Setting the CPU to 4.9GHz also will prevent it from turbo boosting to higher frequencies which impact single threaded performance. I've tested the Core i9-10900K extensively and leaving the thing at stock clocks (allowing for up to 5.3GHz on a 10900K) usually results in better performance than a 5.1GHz all core overclock. Of course, this depends on your actual workload. You know, the thing people actually use to define performance.

Also, Ryzen's are perfectly stable. If it isn't, there was an issue there which isn't likely processor related. It could have been motherboard, firmware, memory, or whatever. And yeah, it was going to get more "CPU marks" because it has a higher core and thread count than the 10850K does.
 
The CPU is currently push/pull. The "push" is the stock fan that came with the 212. The "pull" is the Noctua which is blowing much more air (out of the CPU, towards the blu-ray DVD drive).
thats really gonna mess with with airflow. put both noctuas on it. the 212 is also not a good enough cooler for any chip at 4.9.

Before I set the CPU to 4.9GHz in the bios, the thing was staying at 800 MHz or something, giving me horrible CPU performance, even under load. I think the BIOS reported a minimum 800 MHz and a max 3200 MHz. But whatever it was set to, I was getting equal-to-worse performance compared to my i7-5820. That machine was built in Nov 2015 and it wasn't state-of-the-art even then. So "the same as my old PC" just isn't going to do it for me. Is this CPU *that bad* as far as overheating while not delivering the promised performance?
that was probably your power saving setting, and is normal. in windows you set it to max performance to keep the cpu at full speed.
 
thats really gonna mess with with airflow. put both noctuas on it. the 212 is also not a good enough cooler for any chip at 4.9.


that was probably your power saving setting, and is normal. in windows you set it to max performance to keep the cpu at full speed.

I don't do Windows. I've been 100% Linux since 2011, and not about to stop. Zero reason or incentive for me to use Windows. I was a Windows user up through Windows XP. When M$ tried to force me to buy all new software for absolutely no reason, I said, "I'd rather go open source -- I'll still have to learn all new software, but at least the cost will be zero."

How do you set max performance in Linux? I didn't have to do that with my Ryzen...
 
I hardly know where to begin. First off, who bases their purchase decisions on Passmark CPU scores?

The CPU will report 800MHz or around there at idle. This is normal and this is fine. It will ramp up when needed. That's not what's impacting performance. The clock speeds are dynamic and change based on the workload. And using something like Passmark CPU marks to define performance is ridiculous. It's an arbitrary number. Intel hasn't come a long way in terms of IPC since then so you will only see a meaningful difference in performance in specific use case scenarios. Setting the CPU to 4.9GHz also will prevent it from turbo boosting to higher frequencies which impact single threaded performance. I've tested the Core i9-10900K extensively and leaving the thing at stock clocks (allowing for up to 5.3GHz on a 10900K) usually results in better performance than a 5.1GHz all core overclock. Of course, this depends on your actual workload. You know, the thing people actually use to define performance.

Also, Ryzen's are perfectly stable. If it isn't, there was an issue there which isn't likely processor related. It could have been motherboard, firmware, memory, or whatever. And yeah, it was going to get more "CPU marks" because it has a higher core and thread count than the 10850K does.

Are there any examples where CPUmark scores aren't a good indication of CPU performance? For example, where Processor A has a lower score than Processor B, but is actually faster? I believe this particular benchmark tracks raw CPU power quite well.

Regarding the Ryzen, I researched the issue and apparently random lockups at idle is a "thing", and has been for a couple of years. AMD hasn't fixed it yet. It has to do with certain of its CPU states; I don't remember all the arcane details. But when I discovered threads from 2018 with other people describing the problem I was having, I decided to return my mobo and CPU. I tried 2 different motherboards, and the second one was worse than the first. I can't have a system that locks up randomly like that, not even if it's 50% faster than the equivalent Intel for the same price (which it was!)

It sounds like I need to set it back to stock speeds (auto on everything) but I still want the CPU to rev up to 100% under load. I don't believe it was doing that at first. I'll repeat, I expect better performance (CPUmark or not) than my 5 year old PC, or why bother with the $1200 expense of a new PC build?

P.S.
As an aside, just so you know where I'm at in this game -- I know my title on this forum is "n00b", but I'm hardly a n00b when it comes to PC hardware. I built my first PC in 1997, when I was 21 years old. I have built many PCs since then. That having been said, things change with every new chipset and I haven't worked in PC hardware professionally for many years -- I switched to software dev for a career path years ago. So there's plenty about Gen 10 chips, overclocking, etc. that I don't know. You have to be neck deep in hardware to become expert or truly good at it. Hard to do that when you're only outfitting your household with PCs. ;)
 
Last edited:
I don't do Windows. I've been 100% Linux since 2011, and not about to stop. Zero reason or incentive for me to use Windows. I was a Windows user up through Windows XP. When M$ tried to force me to buy all new software for absolutely no reason, I said, "I'd rather go open source -- I'll still have to learn all new software, but at least the cost will be zero."

How do you set max performance in Linux? I didn't have to do that with my Ryzen...
it was an example, not a conversion attempt. no need for a rant.
idk, i use windows which is why i used it as an example. the low MHz is normal speed stepping or whatever they call it these days. the other way is via bios setting, turn off all power saving, c states etc.
 
Disable c-states in BIOS, that should override the power savings - but you'll run hotter when you do.

CPU mining - can you find out if that uses a lot of AVX?
 
I don't do Windows. I've been 100% Linux since 2011, and not about to stop. Zero reason or incentive for me to use Windows. I was a Windows user up through Windows XP. When M$ tried to force me to buy all new software for absolutely no reason, I said, "I'd rather go open source -- I'll still have to learn all new software, but at least the cost will be zero."

How do you set max performance in Linux? I didn't have to do that with my Ryzen...
Right, what distro / version / config / did you dick with your kernel / do you know how to allocate resources to the software you are using?

Simple is to run a Pi script with top open and see how many threads it's using.

Less simple is the Phoronic test suite tuned to your build.

Least simple is running EKS-D saturating your deployment as if you were manually running a test suite.

You sound like a person that's never had to budget Canonical or Rhel enterprise support, orchestrate licenses, or had to go thru a formal feature request.
 
Are there any examples where CPUmark scores aren't a good indication of CPU performance? For example, where Processor A has a lower score than Processor B, but is actually faster? I believe this particular benchmark tracks raw CPU power quite well.

Regarding the Ryzen, I researched the issue and apparently random lockups at idle is a "thing", and has been for a couple of years. AMD hasn't fixed it yet. It has to do with certain of its CPU states; I don't remember all the arcane details. But when I discovered threads from 2018 with other people describing the problem I was having, I decided to return my mobo and CPU. I tried 2 different motherboards, and the second one was worse than the first. I can't have a system that locks up randomly like that, not even if it's 50% faster than the equivalent Intel for the same price (which it was!)

It sounds like I need to set it back to stock speeds (auto on everything) but I still want the CPU to rev up to 100% under load. I don't believe it was doing that at first. I'll repeat, I expect better performance (CPUmark or not) than my 5 year old PC, or why bother with the $1200 expense of a new PC build?

P.S.
As an aside, just so you know where I'm at in this game -- I know my title on this forum is "n00b", but I'm hardly a n00b when it comes to PC hardware. I built my first PC in 1997, when I was 21 years old. I have built many PCs since then. That having been said, things change with every new chipset and I haven't worked in PC hardware professionally for many years -- I switched to software dev for a career path years ago. So there's plenty about Gen 10 chips, overclocking, etc. that I don't know. You have to be neck deep in hardware to become expert or truly good at it. Hard to do that when you're only outfitting your household with PCs. ;)

I don't know. There may be. My point is that the numbers it generates are arbitrary and do not directly translate to anything useful. It's the same complaint I have with 3DMark. What influences synthetic benchmarks is not necessarily what influences actual workloads. It's a poor way to define what constitutes good performance.

If your gaming, 32,000 CPU marks is meaningless as Ryzen's multithreaded performance advantages do not translate to gaming performance. It's like trying to build a machine for Photoshop and using passmark to decide that rather than actual performance in the application. It doesn't make sense.
 
There is only a half dozen air coolers that can cool this chip within reason 212 evo isn't one if them. Linus tried using a 212 evo and it just roasted the 10900k at full load.
 
There is only a half dozen air coolers that can cool this chip within reason 212 evo isn't one if them. Linus tried using a 212 evo and it just roasted the 10900k at full load.

And the 10900k has better silicon through binning. The 10850k is likely going to boost less and run hotter than a 10900k.

That's not to say it isn't a good CPU and a good value compared to the 10900k... It's just a 10900k that didn't pass the boost test up to the rated speeds with a reasonable power draw.
 
You case fan airflow setup sounds broken. You have 3x inlet fans, and 1x outlet fan. When you have two opposing fan directions like you have in your case (one back inlet + 2 front inlet), you need a similarly-strong outlet to prevent a deadzone at the center of that case.

I would add another case fan to the top of your case (it's hard to force an airflow 90-degree change-of-direction with 1/3 as many outlet fans). Or if that is not possible maybe try 2x inlets in front, and use the top plus rear as the outs?
 
Last edited:
Garbage cooler. Nocatua if you can afford it.
 
Last edited:
  • Like
Reactions: drklu
like this
A 212 EVO is not going to cut it for that OC no matter how many fans you have.
Agreed. The 212 is a fine cooler, but it's not going to cut it for these new higher end cpu's.
 
I don't know. There may be. My point is that the numbers it generates are arbitrary and do not directly translate to anything useful. It's the same complaint I have with 3DMark. What influences synthetic benchmarks is not necessarily what influences actual workloads. It's a poor way to define what constitutes good performance.

If your gaming, 32,000 CPU marks is meaningless as Ryzen's multithreaded performance advantages do not translate to gaming performance. It's like trying to build a machine for Photoshop and using passmark to decide that rather than actual performance in the application. It doesn't make sense.
100% this. If you want to use those programs use a bunch of them and then compare them to legit reviewers to see if you're in the same ballpark as them. That way you know what cooling set up they are using and can more accurately gauge if you are getting the performance you should be getting or not.
 
With the boost behaviours of these new CPUs, cooling and power are what dictate your performance. Same as it's always been for overclockers, just now they do it on their own. So the latest from Intel run very hot and suck a bunch of juice in the process. A serious AIO water cooler 240/280mm, or very large tower air cooler, D15 etc... are required to prevent throttling. Of course ambient air temps matter as always, but a Hyper212 is just no where near enough for that CPU, or any higher teir chip in todays market. It will get heat soaked instantly, and only a leaf blower for a fan could get heat away from it fast enough, maybe not even that.

If you can swing it, budget and case allowing, a good 240/280mm AIO is going to do it best for the price, unless you have room and air flow for a Noctua D15.

In addition, power supply and motherboard can cause throttling as well, if the power delivery is lacking.
 
Back
Top