I have a 10900X and am running BIOS 1.21.
I'm using PerCore mode have have set all cores to 47x with 0 AVX offsets. Yet, in HWiNFO64 I see the following:
Turbo Ratio Limits - IA/SSE, Fused: 45x (1-2c), 43x (3-10c)
Turbo Ratio Limits - IA/SSE, Resolved: 47x (1-4c), 43x (5-10c)
Turbo Ratio Limits - AVX2, Fused: 40x (1-2c), 38x (3-10c)
Turbo Ratio Limits - AVX2, Resolved: 40x (1-2c), 38x (3-10c)
Turbo Ratio Limits - AVX-512, Fused: 40x (1-2c), 38x (3-4c), 37x (5-8c), 34x (9-10c)
Turbo Ratio Limits - AVX-512, Resolved: 40x (1-2c), 38x (3-4c), 37x (5-8c), 34x (9-10c)
 
When loading more than 4 threads, the ratio drops to 43x in accordance to the above. It is incorrectly generating a second 43x (5-10c) ratio limit and ignoring the AVX offsets completely.
 
If I use RatioLimit mode and set all cores to 47x, it works fine and HWiNFO64 shows the following as expected:
Turbo Ratio Limits - IA/SSE, Fused: 45x (1-2c), 43x (3-10c)
Turbo Ratio Limits - IA/SSE, Resolved: 47x (1-10c)
Turbo Ratio Limits - AVX2, Fused: 40x (1-2c), 38x (3-10c)
Turbo Ratio Limits - AVX2, Resolved: 47x (1-10c)
Turbo Ratio Limits - AVX-512, Fused: 40x (1-2c), 38x (3-4c), 37x (5-8c), 34x (9-10c)
Turbo Ratio Limits - AVX-512, Resolved: 47x (1-10c)
 
Another issue is each time I change one of the PerCore ratios, it resets the PerCore Adaptive Voltage to the VID value instead of my last setting.

Monstieur
If I leave the BIOS in RatioLimit mode and use XTU to configure PerCore mode, all the ratios work correctly without throttling.

I have no experience using XTU for tuning, I use it just to check on health (throttling, …). Could you please elaborate how you did this?
 
As a side note, your AVX offsets might be bit too aggressive.
 
I’ve too started with offset of 0 on “per core ratio” multiplier of 43 because everything was indicating I should be able to. System was passing number of tests using AVX for hours without any problem. Then I realized throughput of those tests seems to be significantly lower with offset of 0 than it was before I changed offset away from default of 3. I changed back to 3 and overall throughput counter-intuitively went back up.
 
So I’ve decided to leave it at default until I understand things better. I’ve been able to complete 12+ hours of large set OCCT:CPU and I’m 10+ hours into large set Prime95 without any issues and with very reasonable temperatures.
 
However when I use small sets my CPU very soon starts to thermal throttle. I have a feeling root of that is in my CLC 360 being set to use “coolant temperature curve” and thus being “laggy” behind CPU temperatures, that I might have success if I switch to “CPU temperature curve”.
 
However, CLC 360 fans at full speed sound like a jet at take off, I wouldn’t be able to stand nearby for longer time, so I might try it again once I have better cooling and/or noise control.
 
Point I am trying to make is if I am reaching thermal throttle so fast even with AVX negative offset of 3 on multiplier of 43 I can only imagine what your offset of 0 on multiplier of 47 is doing to your CPU.

Monstieur
Though the UEFI was set to Per-Core mode, XTU first launched and showed that the system was in Package mode. Both Windows installations are running the Balanced power profile.

I would like to double check what my system says which mode it is in when I do same. Where in XTU it says is it currently in package or per core mode?

Monstieur

ZoranC
I would like to double check what my system says which mode it is in when I do same. Where in XTU it says is it currently in package or per core mode?

It does not explicitly tell you whether the system is currently in Package or Per-Core mode, but the Advanced Tuning settings had Package mode preselected.

 
Yes, it seems when you open 'Tuning' page what XTU displays is its starting defaults, not current BIOS values, -BUT- it also seems (I need to double check that) it -DID- read some of values from current ones in BIOS ...
 

Monstieur
When I selected Per-Core mode, it showed the various CPU cores with default multipliers of 47x, 46, and 45x. Yet it would still throttle to 43x in all-core load. After changing to Per-Core mode in XTU and rebooting (not necessary for changing settings usually) it showed all cores at 47x as configured in the UEFI. So it looks like clicking Per-Core mode in XTU did something and make the UEFI values work.

That is interesting. When I now went to XTU and switched tab from 'Package' to 'Per core' mode it showed all of my core multipliers as 43x, exactly the way I have them set in BIOS, not the way you had them (45/46/47). Only time I had it way you did is when I first time switched in BIOS from 'package' to 'per core'. Defaults BIOS offered at that moment were ranging from 45 to 47 but I didn't accept defaults, I reduced them all to 43 (Intel's official 'all core boost' for this CPU).

I’ve worked some more on my settings and here are results:
 
1. With ‘per core’ multiplier of 43x, both AVX offsets of 3 and C States disabled I was able to pass Prime95 large set for 24 hours without any throttling of anything. However, attempt to run smallest set was resulting in occasional XTU reports of PMAX throttling and temperature gradually increasing toward TjMax and then thermal throttling. Attempt to do small set was resulting in practically immediate thermal throttling.
 
2. That was with CLC360 on auto fan curve default (based on coolant temperature). As a next step I’ve tried CLC360 fan on fixed full speed. Smallest set Prime95 was now taking much longer to reach thermal throttling but it still would reach it. Small set would still result in practically immediate thermal throttling.
 
3. Next I’ve went back to BIOS and enabled back C States. That resulted in better temperatures (they were not far from TjMax but were stable)  and now I was able to run smallest set without any thermal throttling. It also took longer for small set to result in thermal throttling.
 
To me that indicated disabling C States might not be impacting regular loads but having them enabled helps significantly with thermal management of AVX loads so I should leave it enabled.
 
4. That also made me think about thermal impact of AVX loads, made me ask myself “Is offset of 3 on x43 too aggressive, what will happen if I make AVX run at base multiplier?”.
 
I changed offset to 6 and now for the first time ever both smallest and small set were able to run without any throttling.
 
5. As a very last step I have put CLC360 back to auto fan profile and I am still able to run small set Prime95 without any throttling. Actually, I’v been running it for over 2 hours now and temperature didn’t exceed 75C with a room temperature of 25C, it was holding practically flat. Now I am running blend set and temperature is oscillating between 55 to 70C without any throttling in sight.
 
All that indicates to me, as uneducated layman, need to be careful around anything that ties into behavior when under AVX loads, that default ‘per core’ negative offset of 3 on x43 multiplier seems too aggressive, and that this will require revisit using XTU and benchmarks that use AVX loads.
 

Monstieur
On my 10900X with just +35 mV I can do ...

To which setting exactly did you add +35 mV, please?

Monstieur
The temperature stays around 100 C and does not throttle.

Intel's official TjMax for this CPU is 94C so you are pushing past it. Is your BIOS CPU TjMax set to Auto? If yes there is a X299 Dark BIOS bug that results in incorrect value higher than what it should be.

ZoranC
To which setting exactly did you add +35 mV, please?
 
Intel's official TjMax for this CPU is 94C so you are pushing past it. Is your BIOS CPU TjMax set to Auto? If yes there is a X299 Dark BIOS bug that results in incorrect value higher than what it should be.

I set Adaptive Voltage to Auto with +35 mV Offset. This gives me 1.335 V + 35 mV on the worst core at 4.9 GHz. With Adaptive Voltage on Auto I can run AVX-512 at only 4.0 GHz since it receives VID at 4.0 GHz + 35 mV.
VID at 4.4 GHz + 35 mV is insufficient for AVX-512. To run AVX-512 at 4.4 GHz I must set Adaptive Voltage to 1.2 V with +35 mV Offset so it receives 1.235 V, which is stable.
Setting Adaptive Voltage to 1.2 V has no effect on the voltage at 4.9 GHz because the VID is already higher on Cascade Lake-X.
 
I use 105 C TjMax only for stability testing. I'm happy with 3 hours of Prime95 stability on SSE / AVX2 / AVX-512, and Linpack Xtreme. I can run -3x / -5x offsets at 4.9 GHz and keep temperatures below 105 C. In my applications it rarely exceeds 70 C.
 
The "worst" core is actually stable at 1.335 V, but even at 1.370 V it's still cooler than the rest. I'm trying to use PerCore mode to tune each core and avoid the unnecessary +35 mV Offset on all cores.

Just read through this whole thread... I dont have this CPU so I cant directly speak to any bug that may or may not be there. However I did see that you are changing the TJMax value to 105C? One thing of note here, that TJMax value is used by the motherboard, CPU, and software to calculate and report CPU temperatures. Artificially increasing TJMax will result in the reported temperature being higher than it actually is. Artificially decreasing the TJMax will result in the reported temperature being lower than it actually is. I would not recommend setting this value to anything other than default settings, especially when you are trying to determine if there is a bug with the CPU/BIOS.  
 
 

Monstieur
I have identified the bug in the BIOS and how to reproduce it.
 
Common Settings:
Multiplier Control = Manual - PerCore
Core 1 to 10 = 47
 
Bug Produced (PerCore ratios completely ignored and CPU reverts to Intel defaults):
Case 1:
Vcore = Adaptive
Adaptive = Any value
Offset = Any Value
 
Case 2:
Vcore = PerCore
Core 1 Adaptive = Any value
Core 1 Offset = Any Value
Core 2-10 Adaptive = Auto
Core 2-10 Offset = +0

Core 2-10 values must be set to Auto / +0 for the bug to be produced.
 
Working Correctly:
Vcore = PerCore
Core 2-10 Adaptive = Fixed Value or Core 2-10 Offset = Fixed Value
 
Setting a value on anything other than Core 1 prevents the bug. But if you want to alter only Core 1 and leave the rest at Auto, it's not possible. Similarly if you want to set PerCore ratios, but not use PerCore voltages, it's not possible.
 
HWiNFO64 also confirms when it's working correctly:
Turbo Ratio Limits - IA/SSE, Fused: 45x (1-2c), 43x (3-10c)
Turbo Ratio Limits - IA/SSE, Resolved: 47x (1-10c)

Could you reproduce the bug again using both scenarios above and post a link to the .CSV file for HWINFO64? Would like to take a look at the data, leave all sensors active for the test. Also could you post of the full specs of your PC? 

jasoncodispoti
You can reproduce this yourself... The easy way is by right clicking on the CPU temp listed in HWINFO64 and change the TJMAx value, take note of the change in the CPU temp when doing so. 

The CPU only reports Distance to TjMax. Temperature is calculated and reported as TjMax - Distance to TjMax.
 
You are speaking of outdated software that required the user to manually enter TjMax. If you override the automatically detected TjMax value, obviously the temperature will be reported incorrectly. Modern software detects the TjMax value you have configured in the UEFI - they give you the option to override it in case it was not detected correctly.
 
The only place you should be changing TjMax is in the UEFI. The CPU uses this value to thermal throttle when Distance to TjMax reaches 0.

s.redkin
In my PerCore mode, the core frequency cannot be higher than the frequency of the first core. What to do about it?

What is happening? The UEFI lets you enter each core ratio separately.

s.redkin
I don't know. The OS robot  finds the optimal values, sets them, but in fact, when working, all frequencies are no higher than the frequencies of the first core. 

It could be OC Robot, combined with the bugs in PerCore mode. Just set all the cores manually and observe them in HWiNFO64. You can use XTU to change the ratios within Windows, then enter your final OC in the UEFI.

s.redkin
I manually entered, anyway. For example, core 7 is 48, and core 1 is 46, then core 7 will only be 46 in operation.

Have you entered values for Adaptive Voltage on all the cores? Look at the ratio for each core in XTU in Windows. If the UEFI values are bugged, XTU will show 80x or something instead of your value.

s.redkin
All adaptive per Core. XTU shows correctly.

You have to enter a voltage even though it's set to Adaptive. If you don't it breaks. Just enter the stock VID, or disable and re-enable PerCode mode to populate the stock VID.
 
You can try using just XTU. Set the UEFI to RatioLimit mode and switch to Per Core mode in XTU. This always works.