CPU:7900x
Motherboard: x299 FTW K
Bios: v.1.22

Hi,
I've tried to overclock CORE 0, per core.
Input Voltage: 1.850v
4.8ghz 1.213v NO Throttling (OCCT small data set 1Hour stable)
4.9ghz 1.275v Throttling
It drops by 100-200Mhz.

I figured out, anything set over 1.24v, starts throttling, doesn't matter if 4.8 or 4.9ghz.
The higher I set the voltage, the higher is the drop. 4.8ghz 1.35v, drop by 500mhz.

Temperature stays at 85 C max.

I've already disabled C-States and set Windows Power Plan from Balanced to High, but nothing changed.
Both EPS power cables, 8+8, are connected to the motherboard.

The Bios has no current limit and no LLC settings.

Input Voltage should be 1.95+ V. You should also increase the long / short duration power limit in addition to current limit. Adaptive voltage does not increase the voltage of CPU cores with higher VIDs. Some cores have VIDs close to 1.35 V at 4.9 GHz. Always set an Offset to increase core voltage at Turbo frequencies.

Monstieur
Input Voltage should be 1.95+ V. You should also increase the long / short duration power limit in addition to current limit. Adaptive voltage does not increase the voltage of CPU cores with higher VIDs. Some cores have VIDs close to 1.35 V at 4.9 GHz. Always set an Offset to increase core voltage at Turbo frequencies.

I've already had the Input Voltage set to 2.0v, hasn't change anything.

Where in the Bios can I find the "long / short duration power limit" setting?

Generally, I'm overclocking with a fixed voltage until it's stable and then I'm switching the voltage mode to offset voltage.

Make sure your AVX offset is set to 0 as well.

If you read reviews like here at Tom's Hardware, you'll see that those things run notoriously on the hot side..Intel opted to use thermal paste (Dow Corning stuff, I presume) between the CPU die and IHS instead of Indium-based solder like on some other CPUs.
 
You should try liquid metal TIM such as Thermal Grizzly Conductonaut (with a whopping 73 W/mk thermal conductivity) between your cooler and IHS, but then the Intel stock paste would act as a barrier..Ideally, you'd want to delid the thing and use Conductonaut between the die and IHS too.

Other than that, go for the best cooler you can get and/or try getting the Vcore down as low as possible, or get another/better mobo that has Vdroop control/better voltage control in the BIOS.
 
Just food for thought here.

bob16314
If you read reviews like here at Tom's Hardware, you'll see that those things run notoriously on the hot side..Intel opted to use thermal paste (Dow Corning stuff, I presume) between the CPU die and IHS instead of Indium-based solder like on some other CPUs.

You should try liquid metal TIM such as Thermal Grizzly Conductonaut (with a whopping 73 W/mk thermal conductivity) between your cooler and IHS, but then the Intel stock paste would act as a barrier..Ideally, you'd want to delid the thing and use Conductonaut between the die and IHS too.

Other than that, go for the best cooler you can get and/or try getting the Vcore down as low as possible, or get another/better mobo that has Vdroop control/better voltage control in the BIOS.

Just food for thought here.

Agree with this 100% as well, you are going to run into overheating issues long before other issues with that CPU... You mentioned above that you are hitting max temp of 85C. If that is the max temp when running OCCT Large Data Set with AVX disabled... than your CPU is running HOT. Generally speaking X299 does not like temps above 80C, I would aim to keep makes temps at or below 80C. 

Sorry, I did not see anything in your original post regarding AVX settings. At least with the paid version of OCCT the dataset does not dictate the instruction set used, you can choose any instruction set. Also AVX is not really used that much in rendering AVX2 and AVX-512 are used most often. From the conversations and testing that I have done there are NO current/power limits on the X299 platform as whole, both are set to unlimited. I know for sure that the X299 Dark from EVGA does not have any power limits configured at the motherboard level I would say that neither does the board that you are using, but you can check with EVGA. 
 
I would say that you have one of the following or a combination of the following...
- Overclock is just not stable
- You are running into the Microcode issue with AVX-512 and either need to use an older BIOS or reach out to EVGA
 
I can help you determine if its the Microcode issue if you want, but you may need to purchase the paid version of OCCT. I would need to go back and look at my notes to see the best way of reproducing the issue. 
 
Edit: 
Corrected post to show that X299 as a whole does not have power limits. 

Ki2Is
It wouldn`t bother me to use an older Bios, but anything below Bios v.1.20 has no PerCore voltage control for Vcore.

Do this first... Submit a ticket to EVGA support and lets get an official answer on the "power limits" for the FTW K. Lets just make sure that there are no power limits on that board before moving forward with other testing. 

Which Country are you in? 

Hmm, I would think its the same for Germany as it is for the US. Register your motherboard on the EVGA site if you have not done so already. Than you should be able to go Member >> Service >> Contact Technical Support.  

Ki2Is

jasoncodispoti

Ki2Is
I'm using OCCT 4.5.1 free version

There is a known issue with the Intel Microcode for X299 CPUs that will cause the CPU to throttle when AVX-512 instructions are being used in combination with voltage overrides. This issue is discussed at length over in this post, . This MAY be what your experiencing as I am pretty sure the free version of OCCT uses AVX instructions by default. The only way to now for sure is for you to purchase the paid version of OCCT and do additional testing, which I would recommend doing regardless otherwise you are not fully testing the CPU and will not know if your overclock is stable. 

OCCT 4.5.1 has different options that you can choose, Small Data Set = uses AVX, and Large Data Set= NO AVX use.
I'm rendering a lot, so I need to test the stability with the option Small Data Set.

I've read the thread above.
I have the x299 FTW K motherboard, not the x299 Dark.
So, that means that issue doesn't come up with Bios 1.17 ?

As Jason noted, this is a known issue with SKX processors and the newest Intel microcode.  I'll send you the file for the E297 FTW K.  It's based on BIOS 1.22, but with older microcode that does not generate this issue.  Keep in mind that this is provided as-is, and EVGA does not guarantee that the processor (or system) will be protected against vulnerabilities patched with newer microcode released by Intel.
 
The alternative is to use adaptive/offset voltages, instead of override voltages.  The trigger for this is when you manually use override voltages.

@EVGATech_LeeM
Thanks for reworked BIOS, worked as designed.
Perfect for overclocking purposes on the 2nd BIOS chip.
The only things I've noticed:
On BIOS 1.22, single CORE#0 4.8Ghz overclock was stable @1.212v.
When I was running the reworked BIOS, I'd to bump the voltage to 1.275v to get a stable overclock again.
So, it appears, not just impacting stability over 1.24v, but the overall stability by using override mode on BIOS 1.22, either.

Realy fast and reliable support.
Appreciate it.