Question i9-11900F - - - low single-core Cinebench 2024 result, multi-core seems OK ?

[radosuaf]

I just bought this one used and benchmarking it I have the system as in the signature (ASUS APE2 enabled) and my Cinebech single core result is pretty poor...

CPU-Z looks somewhat similar (this is 11900F vs 11900K, but multi-core is 1.4% slower whereas single-core is 21% slower):

Is it possible that one or some of the cores don't work full-speed? Is there a way to test a particular core or do those run all the cores, but just one at a given time?

 

[uWebb429]

Intel Turbo Boost does not work correctly on the 11900F unless the C states are enabled in the BIOS. The CPU will run single threaded tasks much slower if this is the problem. Run HWiNO. Check to see what C states your CPU is using when it is idle.

Check the Task Manager for background activity. Too many background tasks will cause more cores to become active and this will reduce the CPU speed.

 

[radosuaf]

This might be the most straightforward and helpful piece of advice I got on the forums and I've been using many of them for YEARS! Didn't know about that and somehow did not come across when looking for the answer...

I assume I set this to:

C-States: Enabled
Thermal Monitor: Enabled
Dual Tau Boost: Auto

?

 

[radosuaf]

OK, so my current BIOS settings are Auto, Enabled, Disabled:

HWiNFO looks like this:

Switched the setting to Enabled and rerun Cinebench - same score (82 points). HWiNFO:

Basically through the whole test the CPU won't go over 5.0 GHz, just at the very end jumps for a second to 5.1 GHz. Temperature is not an issue, as you can see.

 

[uWebb429]

single core result is pretty poor

I have a similar 10850K. When locked to 5000 MHz, it shows the exact same 82 point Cinebench 2024 single core result. Your score might be poor compared to a new Intel 13th or 14th Gen CPU but it seems to be right where it should be for an 11900F.

There was not a lot of difference between Intel 10th and 11th Gen desktop CPUs. The 11th Gen supported the AVX512 instruction set but that is not an advantage if the software you are running does not use those instructions. Surprised that Cinebench was not compiled to take advantage of AVX512.

Basically through the whole test the CPU won't go over 5.0 GHz

Many Intel CPUs over promise and under deliver. The maximum multiplier when a single core is active is rarely used with the typical number of Windows background tasks running. If Thermal Velocity Boost is enabled, the CPU will downclock 100 MHz when it reaches 70°C. It is best to disable TVB.

Run ThrottleStop 9.6 if you want a more accurate look at what multiplier the CPU is using.

ThrottleStop 9.6 Download

ThrottleStop is a small application designed to monitor for and correct the three main types of CPU throttling that are being used on many lapto

www.techpowerup.com

ThrottleStop uses high performance monitoring timers and closely follows the Intel recommended monitoring method. No tool is more precise.

When a 10850K is set to default settings, it can use the 52 multiplier when 1 or 2 cores are active. ThrottleStop shows a multiplier of 51.74. This confirms the CPU is using the 52 multiplier at least 74% of the time. When any Windows background task needs to be performed, the CPU will switch to a lower multiplier. This is what brings down the average. Reduce the Windows background tasks and the multiplier will go higher. Double click on the ThrottleStop monitoring table to see all cores and threads.

Edit - Here is an irrelevant screenshot for comparison purposes. The 10850K has two extra cores and they are all locked to 5000 MHz so no surprise that the Cinebench 2024 multi test is higher compared to an 11900F.

 

[radosuaf]

I have a similar 10850K. When locked to 5000 MHz, it shows the exact same 82 point Cinebench 2024 single core result. Your score might be poor compared to a new Intel 13th or 14th Gen CPU but it seems to be right where it should be for an 11900F.

Funny thing is that the results that are visible on the first screen (11400F - 87, 11700 - 94) are from my machine, I guess, because I owned both of them at some point of time. So that would be the same hardware, different CPUs and:
a/ 11400F is maximum 4.4 GHz
b/ 11700 is maximum 4.9 GHz
c/ this 11900F is maximum 5.0 GHz (if HWiNFO does not lie

If we divide score by speed we get:
a/ 19.8
b/ 19.2
c/ 16.4

Seems something is way off here. Also, the MC score for 11700 was 740 and for 11900 I get 841 in my highest run, so:
a/ 11700 is 4.4 GHz all-core, 4.4/4.9 x 94 = 84.41 theoretical single-core x 8 = 675 vs 740
b/ 11900 is 4.7GHz all-core, 4.7/5.0 x 82 = 77.08 theoretical single-core x 8 =617 vs 841

Looks kinda odd... Unless it's more complicated math.

 

[radosuaf]

Played with OCCT a bit, as it allows to test particular cores:

My "favored cores" must be 2 & 3, as these are the only one reaching 5.1 GHz. With that amount of stress on one core the temp skyrockets within seconds and I guess only liquid nitrogen can somehow tame it - other cores are ~30C and the heatsink is not even warm, but Core 3 throttled.

There are 4 benchmarks in the suite:

Just the first one looks bad...

 

[uWebb429]

if HWiNFO does not lie

HWiNFO is a great program but it does not always accurately track the CPU multiplier during single core tests. Give ThrottleStop a try. Turn on its Log File option before your start Cinebench. When you are done testing you will have an accurate record of your CPU performance. Exit ThrottleStop when you are finished testing so it can finalize the log file. Copy and paste the log file data to,

www.pastebin.com

Post some ThrottleStop screenshots including the FIVR and TPL windows so I can see your settings.

While Cinebench is running, open the Limit Reasons window and watch for any throttling flags turning red. If the CPU is not throttling then it might be the cache that is throttling. If an 11400F can score 87 then you are right, something is not right.

Run msinfo. Is virtualization based security enabled? Is Windows core isolation memory integrity enabled? Let ThrottleStop run for a minute before and after your Cinebench test so it can measure background activity.

 

[radosuaf]

Thanks for the willingness to help! I'll post everything when done. For sure virtualization is enabled in BIOS, but Core Isolation is set to off.

 

[radosuaf]

UPDATE:
Limit Reasons - during the test EDP OTHER & V MAX under CORE and EDP OTHER under RING were flashing yellow and red , while TVB was on constant yellow.

Running the test and will report shortly. Seems that the speed is constantly around 4.2 GHz throughout the test. Should be (hopefully) 5.2 rather? 4.2 is less than 4.8 all-core I have all the time...

Screenshots:

Log: https://pastebin.com/MNKnWBes

 

[uWebb429]

Now do you understand why I would not recommend using HWiNFO to troubleshoot a throttling problem?

The ThrottleStop log file is showing that your CPU is averaging somewhere around 4250 MHz due to EDP and VMAX throttling. EDP stands for electrical design point. High current can trigger EDP throttling and the voltage being set too high can trigger VMAX throttling. Your 11900F is performing worse than an 11400F because it is running slower than an 11400F.

For comparison I overclocked my 10850K to default 10900K speed. This tells the CPU to use the 53 multiplier when 1 or 2 cores are active. My average multiplier was approximately 52.7. No throttling. The CPU will not use the full 53 multiplier continuously during this test because the Windows background tasks are occasionally keeping more than 2 cores active.

https://pastebin.com/GhbRt4Y4

 

[radosuaf]

OK, so we know what is wrong... What should I do now?

Should I start with CPU Core Voltage Offset?

 

[uWebb429]

What should I do now?

I am not sure what Asus Performance Enhancement does on your motherboard. I do not know if this setting is good or bad. The 11900F is a locked CPU. It does not allow any overclocking. You cannot run an all core overclock at the maximum multiplier on a non K CPU so I am really not sure the point of this feature. If this feature increased the turbo power limits, that is fine. If it changed a setting and is causing the throttling that you are seeing, that is not fine.

CPU Core Voltage Offset

In the BIOS on my Asus board I went to Ai Tweaker - Internal CPU Power Management and I lowered the IA AC Load Line and IA DC Load Line settings. I think the Intel load line default is 1.10. I have lowered this to 0.25 for both. This reduced the voltage spikes and lowered the overall CPU voltage. You can use HWiNFO to monitor the actual core voltage when slowly making adjustments to this.

If you do not have this option or do not want to mess with this then you can try using a negative offset voltage.

In ThrottleStop check to see if IccMax for both the core and the cache are set to 255.75. I can see that the core is set to this value but I was not sure about the cache. IccMax for both should be the same.

If you want to eliminate VMAX throttling you can try clearing the V Max Stress box in the FIVR window. You can also clear the Thermal Velocity Boost box to eliminate any TVB throttling. There might be an option for both of these in the BIOS.

To use ThrottleStop to adjust the CPU voltage you can set a -50 mV offset for both the core and the cache. If you decide that you want to make all changes in the BIOS then delete the ThrottleStop.INI configuration file after making any BIOS changes. This will allow ThrottleStop to read the changes you made in the BIOS. It will not try to use any previous changes you might have saved to ThrottleStop.

 

[radosuaf]

I reset all the settings and turned off APE2. It helps a bit (I can see frequency hitting 4.7 GHz at times), but for sure did not resolve the issue.
I'll test IA AC/DC Load Line settings tomorrow, have to finish for today.

Thank you very much again, hope we'll manage to nail it down somehow .

EDIT: Cinebench run finished. The score... 82. So turning off APE2 does not help. As power consumption stays within 65W during single-core loads, I suppose it does not affect performance.

 

[uWebb429]

hope we'll manage to nail it down somehow

I like troubleshooting odd ball problems. I am almost as curious as you are now.

With APE2 off, boot up and delete the previous ThrottleStop.INI configuration file before running ThrottleStop. Deleting the config file allows ThrottleStop to read the turbo power limits that the BIOS has set. Is PL1 set to the Intel recommended 65W default value when APE2 is off?

The log file you posted showed that the CPU was using the 47 multiplier when it was fully loaded just like it is supposed to be using when 8 cores are active. Power consumption reached 151W with zero throttling. The VMAX throttling only happens when the CPU load is reduced. At full load and 4700 MHz the VID voltage shows 1.3135V which is fine. When the load is reduced the VID voltage shoots up towards 1.5000 V, That is what triggers VMAX throttling.

That is why I recommended playing with the loadline settings. Reducing the loadline values can help tighten up the CPU voltage so there is less voltage difference between light load and full load. Clearing the V-Max Stress box in the ThrottleStop FIVR window tells the CPU to ignore any throttling when the CPU detects that the VID voltage is high.

I just had a look at the log I posted. The VID was also sky high when running the Cinebench single core test. 1.5427. Yikes!!

The loadline mod I did shows that the actual voltage going to the CPU is far less than 1.54V. I have no worries. Actual voltage is probably somewhere around 1.35V. I will go start another single core Cinebench run to see what HWiNFO reports for actual VCore voltage. I told you I was curious! Makes one wonder how many other people have this same problem during single core testing that you have.

Edit - When running my CPU at a steady 5000 MHz, actual CPU voltage was 1.323V to 1.332V during Cinebench single core.

With the 10850K set to its default turbo multipliers, it will use the 52 multiplier when 1 or 2 cores are active which is the same as what your 11900F is supposed to be using. The highest multiplier I saw was 51.93. The VCore voltage went up to 1.439V so basically this CPU needs an extra +0.1V to run 200 MHz faster.

A CPU that required lots of voltage to run stable at high MHz was not good enough to be called a 10900K. Intel lowered the maximum clock by 100 MHz and sold these rejects as 10850K. This is a fairly low voltage chip up until 5000 MHz but after that, the voltage table goes up rapidly.

 

[radosuaf]

WOW, you're the MAN! Loaded optimized defaults, turned on XMP, turned off APE2, set IA AC/DC Load Line to 0.25/0.25 and voila! The score is... 99!

The log here:
https://pastebin.com/TsGwupRx

EDP OTHER vanished from RING, but kept on blinking yellow and red under CPU, there were also yellow and red PL1 (I guess that's CPU hitting the power limit) and TVB that was constantly yellow.

After turning APE2 back on again I got 110 in SC and 848 in MC. AMAZING! Do I do any finetuning?

 

[uWebb429]

AMAZING!

That is amazing! Most people complained when 11th Gen was released because it was not any faster than 10th Gen. Your testing just put an end to that internet myth. When these CPUs were first released, there might have been more single core throttling going on than people realized.

Do I do any fine tuning?

It sounds like you have already been doing lots of fine tuning. Did you run another ThrottleStop log file when you did your single core 110 run? The previous log file you posted showed some EDP throttling but with APE2 enabled, that might be OK now. Double check that IccMax is set to the max for both the core and the cache.

A single core score of 110 compared to 82 is a 34% increase. Not bad at all. Can you post an updated Cinebench screenshot? It is hard finding Cinebench 2024 scores to compare to. Cinebench 2024 did not exist when 11th Gen was released.

Edit - Are you sure about 110? A score of 101 makes more sense. Percentage wise that is more in line with the percentage increase in MHz that you are seeing.

 

[radosuaf]

Ahhhh... Sorry, a typo from my side, not 110, but 100. I retested and got 99 again (but this time multi-core is 849):

Log from a single-core run:
https://pastebin.com/5fimMAas

IccMax is 255.75 for both CPU Core and Cache.

There was lots of TVB throttling during the multi-core test, so I guess there is space for improvement there with proper cooling.

 

[uWebb429]

TVB throttling

Clear the Thermal Velocity Boost box in the FIVR window to eliminate TVB throttling. This feature does the opposite of what most people think it does. Checking this box causes throttling.

not 110, but 100

That makes more sense. After I thought about it, 110 sounded too good to be true. 99 or 100 is still a big improvement.

Lots of EDP throttling warnings in the log file but it is not reducing performance any significant amount. It will never be a problem during normal use. There might be a BIOS current limit that can be increased. I will check later.

 

[radosuaf]

100 from 82 is a MASSIVE improvement . Still don't believe changing two settings in the BIOS was enough.... Really impressed with your knowledge! Just wondering how it is possible that 11700 was running just fine and 11900Fwasn't... It basically the same CPU, just clocked a bit higher...

About EDP throttling - would it be that option? Should I set it to maximum?

EDIT: turns out I don't have this option, only those:

Should I play with voltage offset to reduce the current?

 

[uWebb429]

Really impressed with your knowledge!

I am the guy that wrote that ThrottleStop program that you used to troubleshoot this problem. I definitely know more about Intel throttling problems compared to most people you will ever randomly meet in a forum.

The default Intel spec loadline values caused a lot of blue screen issues for my 10850K. Reducing AC/DC loadline to 0.25 dropped my full load Cinebench temperatures significantly and improved stability at the same time.

I have my board set to Level 4 for the Load-line calibration setting. I also have Sync AC/DC Loadline with VRM set to Enabled.

You can try using a negative offset voltage. Somewhere around -50 mV is usually stable. That might help with the EDP issue. Run HWiNFO and monitor the VCore voltage during Cinebench when playing with this setting. The TS Bench 960M test works well. Any errors in this test is a good sign that your CPU needs more voltage. This test tends to report errors just before the CPU crashes so if you see any errors reported, stop the test immediately and increase the voltage.

how it is possible that 11700 was running just fine

The 11700 has a max turbo frequency of 4.90 GHz so that would mean less maximum voltage, less current and not as many or any reasons to throttle. The default voltage goes up very quickly on some 10th and 11th Gen CPUs at around 5.00 GHz. I never realized that this could trigger VMAX throttling at default settings. I would have thought that most people would not encounter this problem until they started overclocking their K series CPUs.

 

[radosuaf]

I am the guy that wrote that ThrottleStop program

I know, I did the research . Glad you're not sitting only on TPU forums .

I have my board set to Level 4 for the Load-line calibration setting. I also have Sync AC/DC Loadline with VRM set to Enabled.

Should I do the same?

You can try using a negative offset voltage. Somewhere around -50 mV is usually stable.

I started testing with Cinebench to verify how it affects performance at the same time. For sure at -50 mV is finishes the single-core test but crashes at multi-core test. -40 mV is the same situation...

 

[radosuaf]

OK, looks like -30 mV is too much, with -20 mV both tests finished - 846 & 100 respectively.

EDP throttling still there.

 

[uWebb429]

Reducing the load line values already reduces the CPU voltage. I am not surprised that there is very little additional room to undervolt further. Cinebench is not a heavy duty stress test. If you can only run Cinebench stable at -20 mV then I would leave the offset voltage at 0 mV. That is what I do. I am more interested in 100% stable than trying to come up with the perfect amount of voltage.

Should I do the same?

It hardly ever hurts to try different things.

I think the Sync Loadline BIOS option is important for when the CPU estimates power consumption. This estimate should be more accurate when sync is enabled.

Level 4 works well for me. Give it a try if you are curious. If you see no advantage, leave it as is.

The amount of throttling that EDP lighting up is causing is next to nothing. I would not worry about it. Most real world apps are not single core so this is never going to be a problem. I think the only way to get another Cinebench single core point is to reduce the background tasks as much as possible and maybe running Cinebench at a higher Windows priority. That is a trick of the pro overclockers. Doing this might lock your computer up for the entire Cinebench test. For a possible extra Cinebench point, it is worth it.

Edit - Have you tried running CPU-Z? Hopefully your Single Thread benchmark score is better now. What no one seems to know is that the CPU-Z Single Thread test actually runs two separate threads which activates two separate cores. The Cinebench single core test is a true single thread test. The CPU-Z test is not.

 

[radosuaf]

I am not chasing highest Cinebench score, this is primarily a gaming machine, so I guess I'll stop here . I just wanted the CPU to work as it should .
I have already the second fan for the cooler but waiting for Noctua to send me the clips. I don't think it will do much, but will repeat the testing when it comes.

EDIT: Yes, CPU-Z looks fine now:

I played CP2077 a bit today (which is way more fun than looking at Cinebench running ) with HWiNFO in the background and the CPU even hit 5.2 GHz on favored cores, so I guess that's the way how it should be running.
I wrote ASUS about the issue - don't expect them to do a new BIOS version for a B560 motherboard in 2024, but who knows..