10850k weird bench

[Imannudein]

Hello all,

 

After i have updated to bios 2004 on the maximus 12 hero wifi, my 10850k will bench on cpu z at approx 7200 upon fresh start. Cinebench r23 will churn 16200 score. I have all limits remove with MCE enabled but no OC. 
 

After a while using the computer, when i try to bench again, cpu z will give 6900 and CB r23 benched at 15300. I have checked via HWinfo64 and saw no throttling whatsoever. Max temp is 73C at 4.8ghz all core on both runs. 
 

how can this be and can someone explain to me what is happening? 
 

Your feedback and input is much appreciated. 

[jones177]

You should run at 4.8ghz all cores all the time without MCE.

With MCE it should be at 4.9ghz or more all the time.

 

I think that MCE is Brocken on the Hero XII since most of the time I can only get 4.9ghz all cores out of it and only once did I get 5ghz. Both are pointless if your CPU is cool enough to use TVB.   My i9 10900k is stable at 5.2 and 5.3ghz so why not even 5.1ghz with MCE.

 

My i9 10900k does 16202 in R23 stock at 4.9ghz all the way through and CPUz is 7463 so you are about right for 4.8ghz.

 

I don't use MCE because TVB beats it in most games since my temps are below 70c but in CPU benches it is pointless.

 

Using TVB is working out for me now but at some point I will do a 5.2 or 5.3ghz all core manual overclock.

 

 

[unclewebb]

@Imannudein At default settings, my 10850K runs fully loaded at 4800 MHz and I get these CPU-Z results. A difference of 100 points in the Multi Thread test is normal variation.

 

 

If your results are decreasing after you run your computer for a few hours, open the Task Manager, go to the Details tab and see if there is anything significant running in the background. I get consistent results because System Idle Process always shows 99% when I begin testing. 

 

[Imannudein]

1 hour ago, unclewebb said:

@Imannudein At default settings, my 10850K runs fully loaded at 4800 MHz and I get these CPU-Z results. A difference of 100 points in the Multi Thread test is normal variation.

 

 

If your results are decreasing after you run your computer for a few hours, open the Task Manager, go to the Details tab and see if there is anything significant running in the background. I get consistent results because System Idle Process always shows 99% when I begin testing. 

 

Thank you guys for the feedback. So in essence, i should reset CMOS and press F1 upon boot? Or just change MCE to auto? 
 

FYI, on the maximus, upon first boot ir after bios update, we are presented with an option to F1 for default intel or F3, all limits removed. However, settings in extreme tweaker remains the same in terms of MCE. 

[unclewebb]

@Imannudein What do you want your CPU to do? Do you want it to run at the same speed (4800 MHz) whether 1 core or all 10 cores are active? Do you want it to use the default turbo ratios so it clocks up to 5200 MHz when 1 core is active? Here is an example of a 10850K using its default turbo ratios during a 1 Thread test.

 

 

These CPUs are extremely flexible and you can set them up to run however you want them to run. First you have to decide how you want your CPU to run.

 

You can also lower the voltage and lower the power consumption and temperatures significantly. Here is an example of what is possible. 

 

 

 

[jones177]

3 hours ago, Imannudein said:

Thank you guys for the feedback. So in essence, i should reset CMOS and press F1 upon boot? Or just change MCE to auto? 
 

FYI, on the maximus, upon first boot ir after bios update, we are presented with an option to F1 for default intel or F3, all limits removed. However, settings in extreme tweaker remains the same in terms of MCE. 

I have not been able to find any decimation on this feature. 

I have it set to F3 since I watched this video just before I built the computer.

 

[Imannudein]

@unclewebbso you're saying that I should run MCE at Auto instead of enabled? What I'm looking for the CPU is no OC, 4.8Ghz all core all the way since I have proper cooling via a Kraken x63.

[unclewebb]

@Imannudein The 10850K has the ability to use the 52 multiplier when lightly loaded. That is not overclocking. Those are the default specs for a 10850K. It was designed by Intel to automatically use a higher multiplier when lightly loaded. You are not going to hurt this CPU by using it like this.

 

If you are not interested in using your CPU at default specs, choose Sync All Cores in the BIOS, set that to 48 and your CPU will use the 48 multiplier whether 1 core is active or all 10 cores are active.

 

[Mister Woof]

13 hours ago, jones177 said:

You should run at 4.8ghz all cores all the time without MCE.

With MCE it should be at 4.9ghz or more all the time.

 

I think that MCE is Brocken on the Hero XII since most of the time I can only get 4.9ghz all cores out of it and only once did I get 5ghz. Both are pointless if your CPU is cool enough to use TVB.   My i9 10900k is stable at 5.2 and 5.3ghz so why not even 5.1ghz with MCE.

 

My i9 10900k does 16202 in R23 stock at 4.9ghz all the way through and CPUz is 7463 so you are about right for 4.8ghz.

 

I don't use MCE because TVB beats it in most games since my temps are below 70c but in CPU benches it is pointless.

 

Using TVB is working out for me now but at some point I will do a 5.2 or 5.3ghz all core manual overclock.

 

 

 

I don't know what's going on with my 10900k, but at default settings, I RARELY ever see 5.1ghz even, with game temps in the 50s. Generally it's at 4.9ghz 99% of the time.

 

Never even seen 5.2ghz, 5.3ghz for more than a blip for a millisecond. TVB is enabled, otherwise default.

 

I get better results from manual 5.1ghz OC. Not sure what I'm doing wrong at stock settings.

 

Doing CBR23 single core right now at default settings, core 3 hit 5.3ghz for a split second then stabilizes at 4.9-5.1ghz

 

 

Cpu package is 44c.

[unclewebb]

40 minutes ago, Mister Woof said:

at default settings, I RARELY ever see 5.1ghz

There are high performance counters within the CPU that run at billions of cycles per second. These timers allow the CPU multiplier to be precisely monitored. Most monitoring programs have decided to cut corners and are not accurate enough to show you what a CPU is really doing. By not using the Intel recommended monitoring method, they are not capable of showing you what percentage of time a 10900K is using the 53 multiplier.

 

For a quick test, I set my 10850K to the same default turbo ratios that a 10900K uses.

 

 

A 10900K at default specs can use the 53 multiplier when 1 or 2 cores are active. This decreases to the 51 multiplier when up to 3 cores are active and it will use the 50 multiplier when up to 5 cores are active. When between 5 and 10 cores are active, the 10900K will use the 49 multiplier.

 

Here are the results while a 1 Thread benchmark test is in progress.

 

 

It is obvious that the CPU is using the 53 multiplier for the vast majority of time. Windows is running this 1 thread load on the first core so it is easy to see the maximum multiplier. A 1 thread load on a CPU that supports 20 threads means the CPU needs to spend 5% (1/20) of its time in the C0 state processing this task. The other 0.1% is used to process the various Windows background tasks. 

 

Intel CPUs definitely work correctly. Too bad most monitoring software has convinced users that there must be something wrong with their processor.

[Mister Woof]

8 minutes ago, unclewebb said:

There are high performance counters within the CPU that run at billions of cycles per second. These timers allow the CPU multiplier to be precisely monitored. Most monitoring programs have decided to cut corners and are not accurate enough to show you what a CPU is really doing. By not using the Intel recommended monitoring method, they are not capable of showing you what percentage of time a 10900K is using the 53 multiplier.

 

For a quick test, I set my 10850K to the same default turbo ratios that a 10900K uses.

 

 

A 10900K at default specs can use the 53 multiplier when 1 or 2 cores are active. This decreases to the 51 multiplier when up to 3 cores are active and it will use the 50 multiplier when up to 5 cores are active. When between 5 and 10 cores are active, the 10900K will use the 49 multiplier.

 

Here are the results while a 1 Thread benchmark test is in progress.

 

 

It is obvious that the CPU is using the 53 multiplier for the vast majority of time. Windows is running this 1 thread load on the first core so it is easy to see the maximum multiplier. A 1 thread load on a CPU that supports 20 threads means the CPU needs to spend 5% (1/20) of its time in the C0 state processing this task. The other 0.1% is used to process the various Windows background tasks. 

 

Intel CPUs definitely work correctly. Too bad most monitoring software has convinced users that there must be something wrong with their processor.

Well, benchmarking didn't fare well despite lapses in monitoring software.

 

On stock settings, I was only scoring 560 single thread and 7002 multi in cpuz with 99idle. Temperatures were good, 60s in multicore and 40s single. Monitoring reported 4.9-5.1ghz during the single core and 4.8-4.9ghz multicore.

 

Same test manual locked all core 5.1/4.8 cache, 605 single/7586 multi.

 

Maybe gigabyte bios is just bad at boosting.

 

Googling this 4.9ghz wall, I see tons of hits. 

 

Manual overclockimg resolves the issue, but it's a bit frustrating.

 

[unclewebb]

55 minutes ago, Mister Woof said:

I was only scoring 560 single thread and 7002 multi in cpuz

Maybe your Gigabyte board did not set your CPU up correctly to the default Intel specs. A 560 single thread test score is way below what it should be.

 

One interesting thing is that If you examine the CPU-Z Single Thread test closely, you will see that it is actually loading 2 threads, not 1 thread. The screenshot shows the first task is fully loading one thread of the CPU. The second thread of this test is bouncing between 2 other cores so the CPU is mostly using the 51 and 52 multiplier during the CPU-Z Single Thread test. 

 

 

Cinebench R23 is a true single thread test. It only loads one thread of the CPU. The screenshot shows an average multiplier of 52.74 while this test is in progress. That means the CPU was using the 53 multiplier 74% of the time and the 52 multiplier the other 26% of the time. Definitely proof that the 53 multiplier exists and is being used.

 

 

 

[Imannudein]

1 hour ago, unclewebb said:

Maybe your Gigabyte board did not set your CPU up correctly to the default Intel specs. A 560 single thread test score is way below what it should be.

 

One interesting thing is that If you examine the CPU-Z Single Thread test closely, you will see that it is actually loading 2 threads, not 1 thread. The screenshot shows the first task is fully loading one thread of the CPU. The second thread of this test is bouncing between 2 other cores so the CPU is mostly using the 51 and 52 multiplier during the CPU-Z Single Thread test. 

 

 

Cinebench R23 is a true single thread test. It only loads one thread of the CPU. The screenshot shows an average multiplier of 52.74 while this test is in progress. That means the CPU was using the 53 multiplier 74% of the time and the 52 multiplier the other 26% of the time. Definitely proof that the 53 multiplier exists and is being used.

 

 

 

What is this software and how can i get hold of it? Tonight i'll be testing the following:

 

1. Reset CMOS

2. Press F3 upon boot

3. Enable XMP1 for RAM

 

Will post back. 

[Mister Woof]

I ended up giving up on stock function and all-core OC

 

Just put my cache at 48, vcore fixed to 1.335, and changed the turbo limits

 

1 core 53

2 core 53

3 core 53

4 core 53

5 core 53

6 core 52

7 core 52

8 core 51

9 core 51

10 core 51

 

seems to bench a lot better (624 single, 7550 multi) in cpuz (5.1-5.2ghz), and maintains 5.3ghz in cinenbench single core (1420) and 5.1ghz multicore (17200)

 

Stability testing on all-core workloads remains the same as when I had the 5.1ghz all-core dialed it. Just gives it some more oomph when i've got lightly threaded tasks.

 

Also re-enabled all the power saving C-states, EIST, etc, so it powers down when idle.

 

right now according to hwinfo64 im at 7.9w cpu package power and 800mhz

 

Edits: messed up the names in the benches

[Imannudein]

@unclewebb

So i have reset the BIOS and did the following:

1. Press F3 upon boot

2. Enabled XMP 1

3. MCE left at Auto Let BIOS Decide

 

And here are the results. All core boosting to 4.8Ghz all the way till i stop the test. Relatively low compared to others. 

 

[Imannudein]

@unclewebbHere's another screenshot showing package power 

 

[Imannudein]

New problem tonight. Booted with MCE enabled and XMP1. While on desktop, computer shutdown and tried to reboot with Q Code 7F. Cut power, reboot and all seems normal now. Weird. 

[Mister Woof]

I did MCE on my gigabyte board and it applied 5.3ghz all core and 1.56v VROUT

 

Undid that with the quickness

[Imannudein]

6 hours ago, Mister Woof said:

I did MCE on my gigabyte board and it applied 5.3ghz all core and 1.56v VROUT

 

Undid that with the quickness

I think that was the problem too. 

[Mister Woof]

11 minutes ago, Imannudein said:

I think that was the problem too. 

I think the motherboards all just go nuts doing their own thing. IMO best to manually adjust it or run at default without any enhancements.

 

Part of me thinks its intentional to burn out your CPUs so you have to buy more motherboards and upgrade sooner.

[Imannudein]

Just now, Mister Woof said:

I think the motherboards all just go nuts doing their own thing. IMO best to manually adjust it or run at default without any enhancements.

 

Part of me thinks its intentional to burn out your CPUs so you have to buy more motherboards and upgrade sooner.

So I know now how to manually OC with fixed multiplier and fixed vcore. How do you do the OC as per your post above where you sort of put a boost table to it. 

[Mister Woof]

4 minutes ago, Imannudein said:

So I know now how to manually OC with fixed multiplier and fixed vcore. How do you do the OC as per your post above where you sort of put a boost table to it. 

Not sure about your mobo but this is what i did on mine:

 

 

[Imannudein]

Thank you for the quick reply. will try later. 

[xg32]

8 hours ago, Mister Woof said:

I did MCE on my gigabyte board and it applied 5.3ghz all core and 1.56v VROUT

 

Undid that with the quickness

if i can stomach 1.5v vid at idle i'd be running 5.3 24/7, some research indicates it's fine as long as it's not actual vcore, but uhh not for me . might not even me the VROUT ur talking about but sounds about right with MCE.

[Mister Woof]

Just now, xg32 said:

if i can stomach 1.5v vid at idle i'd be running 5.3 24/7, some research indicates it's fine as long as it's not actual vcore, but uhh not for me 

i don't know for sure, but first time i booted up i tried MCE - saw 5.3ghz all core and VR VOUT at 1.56v and hit restart immediately lol