Poor CPU performance in stress test

[Valantar]

Max core speed of 4.49 GHz IS NOT GOOD. The 10850k should be hitting 5.2 GHz on a single core at stock settings. Also, as already mentioned, a 120mm AIO is not enough to cool the 10850k. I tried using a MSI MAG CORELIQUID 240mm AIO, and it had issues keeping my 10850k cool while under full load (WCG), in a case with excellent airflow (Silverstone FT04).

If you read the OP, you'd see they are running MT tests, not single threaded ones, so ... single threaded turbo speeds are irrelevant.

OP: Whether or not this is an issue depends entirely on how much you stand to gain from a potential <10% performance increase, and whether a potential 2x CPU power increase is worth it to you. Your CPU seems to be configured to run at Intel's stock settings, i.e. a 125W power limit for long-term loads. A 120mm AIO also seems decently matched to that type of thermal output. Judging by your CPU temperatures, you're entirely fine at those settings. And your CPU is performing well - you're seeing 4.5GHz all-core boost speeds within TDP on a CPU with a base clock (=rated minimum sustained clock speed at TDP) of 3.6GHz. That's nothing to scoff at. Everything seems to be operating as it should.

Does your CPU have more in the tank, so to speak? Sure, if you're willing to give it better cooling (at least a 240mm AIO, but ideally a 280/360) and don't care about power consumption or heat output into the room. You can probably tune it to run all-core at >5GHz, but as I said, that is likely to come at a ~2x increase in CPU power consumption. 250W might be pessimistic, but you'll definitely be exceeding 200W at those speeds. Which will consume more electricity, output more heat into the room, need more expensive cooling, be noisier, etc. Your case might not even be able to handle that thermal output. And remember, this is for a ~10% increase in clock speed, meaning you'll at best be getting a 10% increase in performance, though likely less than that - few workloads scale perfectly with increased clocks. And that is only for the CPU, of course - it won't affect other parts of the system much. If your workload is GPU bound for example, it isn't likely to change much.

If this is for a mission-critical workload, I'd leave well enough alone and use it as is. If the potential for a <10% CPU performance really matters to you, and you are strongly CPU bound in your workload, then you can improve your cooling and try to tune it further. Personally, I wouldn't bother.

 

[BarbaricSoul]

If you read the OP, you'd see they are running MT tests, not single threaded ones, so ... single threaded turbo speeds are irrelevant.

Do I really have to set my 10850k to stock settings and show what the OP's CPU is suppose to be running like? I have a 10850k. I have dealt with thermal throttling, though I have not had any issues with power limit throttling. I know from experience how the 10850k runs and what it needs to run like it should. Anyway, Uncle Web has this situation handled, and knows more than I do about the 10850k, so I'm done here.

 

[docnorth]

oof power limits and thermal limits, welcome to the real word of modern intel

This video will have a lot of relevant info:

You're going to need to unlock the power in the BIOS, and upgrade the cooling if you want the 'advertised' 5GHz+ perforfmance numbers you see online

Depending on the motherboard and its VRM settings, performance varies *massively*

But even with power limits on, a result like @unclewebb posted should be expected. This PC almost crashes when hitting the powerlimit.

 

[londiste]

Do I really have to set my 10850k to stock settings and show what the OP's CPU is suppose to be running like? I have a 10850k. I have dealt with thermal throttling, though I have not had any issues with power limit throttling. I know from experience how the 10850k runs and what it needs to run like it should. Anyway, Uncle Web has this situation handled, and knows more than I do about the 10850k, so I'm done here.

If you have not had any issues with power limit throttling, you are not running it at stock. Stock would mean 125W PL1, 156W PL2 and probably a 56-second Tau given adequate cooling. This would limit the sustained frequencies by a lot. And practically any test that even decently loads all cores should hit the power limit. On the other hand I would be surprised if many motherboards actually default to stock

 

[Valantar]

Do I really have to set my 10850k to stock settings and show what the OP's CPU is suppose to be running like? I have a 10850k. I have dealt with thermal throttling, though I have not had any issues with power limit throttling. I know from experience how the 10850k runs and what it needs to run like it should. Anyway, Uncle Web has this situation handled, and knows more than I do about the 10850k, so I'm done here.

I was going to write a response to this, but @londiste beat me to it. Specs are specs, and any current or recent Intel CPU at Intel stock specs will be power limited in MT workloads.

 

[unclewebb]

CPU-Z score dropping

I have run the CPU-Z benchmark about a million and one times and I have never seen the score drop to 0 while this test is running.

I deliberately throttled my CPU as low as it could go, 800 MHz.. The results were pathetic but even then the score never locked up at 0.

I ran a separate stress test which fully load all 20 threads and then I started the CPU-Z benchmark. The benchmark scores were way down but the CPU-Z test still ran smoothly. No sudden drops to 0.

My verdict is that your computer definitely has a problem. Sometimes computers can freeze like this because the hard drive is ready to fail. Do you have any mechanical hard drives in your computer? I am not sure if SSD hard drives can have these same issues.

As for power limit throttling, here are Intel's recommended specs for their 10 core CPUs which includes the 10850K, 10900K and the 10900KF.

The datasheet shows that these are the "Recommended Value". A manufacturer or an end user can set the power limits to whatever they like. This is kind of like buying a Corvette. Of course GM is going to recommend that you always drive the speed limit. That covers their butt but who buys a high performance car to drive around at 55 mph? Hopefully no one. If you ever need someone to help you step on the gas pedal, I will show you how. You need to figure out the going to 0 problem first. At the default power limits, a computer should never drop to 0 during the CPU-Z benchmark test.

When your computer is idle at the desktop, open the Task Manager and go to the Details tab. Watch this data for a minute. Anything unusual loading your CPU?

Check to see if there are any BIOS updates available for your computer. Your screenshot shows that your CPU is using microcode 0xC8. My computer shows microcode 0xE2. The latest version might be newer than that. I would go to the Asus website to find the most updated BIOS version for your motherboard. The problem you are having might have been fixed already. The BIOS you are using is from July 21, 2020.

I used to use HD Tune to scan for bad blocks on the hard drive. The free version is good enough. There are lots of other similar tools.
When you run the HD Tune Benchmark, watch for sudden drops to 0.

13 Free Hard Drive Testing Tools (January 2025)

An updated list of the top hard drive testing programs you can use for free. Test your HDD for problems with these easy-to-use apps for Seagate, Samsung, Western Digital, Fujitsu, and other hard drives.

www.lifewire.com

does thermal throttling use power limit throttling as a means?

These two types of throttling are completely separate. Thermal throttling does not change the power limits.

When a CPU reaches 100°C, it will start to thermal throttle. This means it will only slow down as much as necessary so it does not exceed 100°C. An Intel CPU can stay pegged at 99°C for hours while it is thermal throttling. Some users think that thermal throttling is when the CPU drops to 800 MHz and sits there for a minute while the temperatures decrease. That is not how Intel thermal throttling works. The CPU temps are sampled hundreds of times per second. If the CPU can go faster without hitting 100°C, it will go faster. If it hits 100°C, it slows down a hair. This continues indefinitely.

Power limit throttling works the same way. If the power limit is set to 125W, the CPU will speed up until it reaches this limit. When it hits 125W, it slows down a hair to stay under this limit. Once again, the CPU is adjusting itself hundreds of times per second to stay within all of the limits.

 

[Dolt]

...

OP: Whether or not this is an issue depends entirely on how much you stand to gain from a potential <10% performance increase, and whether a potential 2x CPU power increase is worth it to you....

... If your workload is GPU bound for example, it isn't likely to change much.

If this is for a mission-critical workload, I'd leave well enough alone and use it as is. If the potential for a <10% CPU performance really matters to you, and you are strongly CPU bound in your workload, then you can improve your cooling and try to tune it further. Personally, I wouldn't bother.

Nice points Valantar. I do not need the extra 10% and at this point I would be satisfied with it running stable at its 3.6 GHz non-overclock. As long as it keeps my GPU fed, that's the most important part (although CPU is needed for things like pre-processing).

I have run the CPU-Z benchmark about a million and one times and I have never seen the score drop to 0 while this test is running.
...
My verdict is that your computer definitely has a problem. Sometimes computers can freeze like this because the hard drive is ready to fail. Do you have any mechanical hard drives in your computer? I am not sure if SSD hard drives can have these same issues.
...
You need to figure out the going to 0 problem first. At the default power limits, a computer should never drop to 0 during the CPU-Z benchmark test.

When your computer is idle at the desktop, open the Task Manager and go to the Details tab. Watch this data for a minute. Anything unusual loading your CPU?

Check to see if there are any BIOS updates available for your computer. Your screenshot shows that your CPU is using microcode 0xC8. My computer shows microcode 0xE2. The latest version might be newer than that. I would go to the Asus website to find the most updated BIOS version for your motherboard. The problem you are having might have been fixed already. The BIOS you are using is from July 21, 2020.

I used to use HD Tune to scan for bad blocks on the hard drive. The free version is good enough. There are lots of other similar tools.
When you run the HD Tune Benchmark, watch for sudden drops to 0.
...
Power limit throttling works the same way. If the power limit is set to 125W, the CPU will speed up until it reaches this limit. When it hits 125W, it slows down a hair to stay under this limit. Once again, the CPU is adjusting itself hundreds of times per second to stay within all of the limits.

Yes, I am confused on why the CPU-Z score drops to ~0 ten minutes into the stress test (shown below). As people have pointed out, the heat in the mid 70s seems okay. During the test, Intel XTU reports that the TSP climbs to a number higher than 125, but five minutes in when the Power Limit Throttling indicator becomes active the TSP goes to 125. The CPU-Z score at that point is and remains for several minutes near the reference score, maybe down a smidge from where it was before. Once the score drops very low (~0, shown below), I have Intel XTU history of Max Core Frequency and that number doesn't tank (range 4.41 - 4.80 GHz). Also I have had Window's built in Task Manager and Performance Monitor on, and CPU usage and reported speed remain acceptable.

I don't see why the CPU-Z score is dropping so drastically and I am stuck in figuring out what the culprit is. I suspect that raising the Long Duration Package Power Limit (PL1), disabling Turbo, and/or updating the BIOS and chipset won't change it. At this point I am wondering if I am misunderstanding this CPU-Z stress test is telling me. I have tried an older version, but the result was the same.

While idle, I don't see anything spiking on the Task Manager > Details.

I have an SSD which is where the OS, page file, and application are running. The HDD is empty right now. But I'm running the HD Tune and if it sees anything, I'll report back.

 

[unclewebb]

why the CPU-Z score drops to ~0 ten minutes into the stress test

I thought you were seeing these drops during the Bench CPU test. During the Bench test, I have never seen any drops to 0. I have never tried running the Stress CPU test for 10 minutes. I usually get bored before that.

Maybe this drop to 0 is a normal part of a long term Stress CPU test.

Now I have to run a Stress test for 15 minutes and watch for anything unusual.

*** PROBLEM SOLVED ***

After 10 minutes of the CPU-Z Stress test, the score drops to 0 and then rapidly starts counting up. The CPU speed and power consumption are all consistent when this happens. This is not any weird sort of throttling. It is just part of the Stress CPU test.

My score is back up to 2100 and counting. Now I have to wait a few more minutes to see if it peaks.

Edit - It stopped going up at 2390.1

The CPU is set to default turbo ratios and default power limits during this test. Power limit throttling at 125W is in progress.

After 10 more minutes, it started the cycle again.

Here is a ThrottleStop log file that shows consistent CPU performance during this test.
It takes a few minutes to throttle down to the 125W power limit and then runs consistently at that level.
CPU-Z suddenly restarting at 0 is not because of any throttling.

10850K CPU-Z Stress Test - Pastebin.com

Pastebin.com is the number one paste tool since 2002. Pastebin is a website where you can store text online for a set period of time.

pastebin.com

 

[Dolt]

After 10 minutes of the CPU-Z Stress test, the score drops to 0 and then rapidly starts counting up. The CPU speed and power consumption are all consistent when this happens. This is not any weird sort of throttling. It is just part of the Stress CPU test.

Yes, that is exactly what I am seeing! Good to know that a non-CyberPowerPC system shows the same thing. Thank you for digging into this!

CyberPowerPC tech support mentioned that it might be a CPU-Z behavior (bug or intentional). I tried older version of CPU-Z (in case it was a software regression) and I saw the same issue. I tried running CPU-Z on another computer and did not see the behavior of the score dropping. So I'm not totally sure that everything is okay, but if unclewebb is seeing the same thing and his PC normally runs fine, then I guess mine will too. Maybe the whole thing was just a bug in CPU-Z (or my misunderstanding how it should run).

Thanks everyone for the support on this! Lots of great ideas.

 

[unclewebb]

Hopefully some more users try running the CPU-Z Stress test. Maybe this weird thing only happens on Intel's 10 core CPUs or maybe it happens on all Intel CPUs. I know it happened right at the 10 minute mark and then again at the 20 minute mark so it is definitely not just some random bug. It is by design.

You could try contacting the programmer of CPU-Z. I am sure he would know about this feature / bug.

Now that this issue is solved, are you going to increase your turbo power limits to get some more performance out of this CPU? Running a 10850K at 125W is about as much fun as driving a Vette while following a traffic cop.

 

[Valantar]

I was about to suggest that it looked/sounded like a software issue/feature/bug/oddity - there's nothing in a CPU that ramps up as linearly and smoothly as those scores seemed to do after zeroing out, after all. If it was dropping clocks and then boosting again, it would be much more of a see-saw movement.

I'm running the CPU-Z stress test now (Ryzen 7 5800X), will report back if I see anything odd after ten minutes or so.

 

[Dolt]

Now that this issue is solved, are you going to increase your turbo power limits to get some more performance out of this CPU? Running a 10850K at 125W is about as much fun as driving a Vette while following a traffic cop.

I'm a little ashamed -- especially in an OC forum -- to admit it, but for now I am going to drive the 'vette down Main St obeying the posted limits. But once I verify that this system is stable and I don't have to worry about voiding the return policy, I would like to look into reving it a bit.

 

[mtcn77]

Hi, since you have a 'client' cpu it is not expected to work day and night at the rated tdp. Server cpus are not like that. They are expected to do so. Therefore, you often have options like 'high performance computing' HPC on client motherboards sporting this mixed use case. If you had a computer with no preference for moderate duty, you would lose a lot of bins when some of the cores were idle, so it is a compromise on either end. You either have throttle response, or high end rpm sort of speak.

 

[Valantar]

I'm a little ashamed -- especially in an OC forum -- to admit it, but for now I am going to drive the 'vette down Main St obeying the posted limits. But once I verify that this system is stable and I don't have to worry about voiding the return policy, I would like to look into reving it a bit.

Honestly, overclocking these days is pretty pointless unless it's your hobby. The days of 30-50% performance gains are long gone. Chipmakers these days know how to squeeze performance out of their silicon, and there's little left on the table. Unless you enjoy tinkering and want to spend some time and money on it, leaving it alone is the sensible thing to do. And running at 4.5GHz all-core instead of ~5 is hardly driving a sports car within the speed limits - it's still a damn fast CPU.

Edit: Just reporting back, seeing exactly the same behaviour here in the CPU-Z stress test. After ~10 minutes score falls to 0, then slowly ramps up. Likely something changing about the workload at that point.

 

[unclewebb]

Likely something changing about the workload

The workload seems very consistent when the reset to 0 happens. There is no change in load, measured power consumption or CPU speed.

My theory is that this feature was deliberately added to CPU-Z. When stress testing, the benchmark score tends to flat line at a fixed value. Anyone looking at a steady number might start wondering if their computer has frozen. By resetting this number every 10 minutes, a user will get some feedback that all is still OK.

@Valantar Thanks for confirming that it does the same thing on a Ryzen CPU.

 

[mtcn77]

Running a 10850K at 125W is about as much fun as driving a Vette while following a traffic cop.

This is not a responsible recommendation. That car should race on a drag strip, not public roads, if the driver does not want to become cheap amusement on the petrolhead center channel for onlookers...
Drive safe, no one cares if you beat the cops on your way to pick groceries.

If you want, drop cpu-z entirely and stick to P95. Depending on how you select fft size, you can fine tune your tdp to be in line with your general TDP and this enables your overclocking tests to meet daily temperature stability findings. If a cpu runs x overclock at y temperature in some burn test, you can run that cpu at x clock at y temperature for all intents and purposes without fail.

PS: I think overclocking is best when money is on the line. Without putting something at stake, the gains are even less. Say we want thermal paste performance, it is the least reliable when applied by a thin coat; however it is most susceptible to caking in this reduced film thickness state. Applying at more regular intervals becomes a necessity. It is better knowing the risks and coming to grips with them prior to setting the goals.

 

[docnorth]

Didn't understand either it was the stress test, not bench. Anyway I ran the stress test for about 12-13 min on a i7-6700 non-k with stock settings and stock cooler (HP prebuilt office PC), without the same drop after 10 min. CPU is 5 y old, CPU-Z version 1.96. Temperature reached 76 C, but it's 38 C outside and 27+ C inside.