TLDR
Its looking that Ryzen Gremlins are not allowing me to get consistent results, looking like just leave PBO on AUTO and tweak CO ....
Warning, if you want to waste your time read further
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Note:
Some info below is relevant to my CPU which is a 5600X, those who have higher core count CPUs hope you can test the methodology outlined below to see if it holds true, same goes for other 5600X users, would be great if other could validate my findings
Prerequisites
1. Stable DRAM set
2. CO (Curve Optimiser) set to per core and each core is tuned to its most stable value while using "Boost Override" set to 200 mhz.
3a. CPU LLC: Auto
3b. CPU NB LLC: Auto
4a. vCORE: Auto
4b. vSOC/vDIMM/vDDP/vDDG: Whatever you require for your stable DRAM set
5. PPT/TDC: The maximum value your BIOS supports
5. PPT/TDC, set to AUTO, see amendment to section 7. CPU VDD Telemetry Offset Value
6. EDC
This needs discussion, as im water-cooled im not concerned about restricting the "Thermal Design Current (TDC)" hence the reason for setting it to max value, also im not concerned about maximum power efficiency so again i set the "Package Power Total (PPT)" to its maximum value. I have found through my testing that setting these at maximum values does not negatively influence, in terms of performance, the maximal obtainable performance from my system when running x/y/x/ benchmark.
6. EDC
The above (striked out comment) is valid only if you have a specific use case scenario, though in such scenarios you would be better off using an all core overclock, not PBO.
All we need to be concerned about is tuning the right value for EDC and not worry about anything else.
So we go through the process of, set an EDC value, reboot, rinse and repeat, we will use CB23 multi-core and Linpack Xtreme (selection: 1, 3, 5, enter, enter) to compare results between different EDC settings.
For the record, my optimal EDC value for my 5600X sample is 110A, there is a bit of leeway between 105 to 115 so if you have OCD you can try each increment, hahahahahaha, but it wont tell you anything so why bother
7. CPU VDD Telemetry
After you got your EDC value you do the following, set EDC to double your value, yeah I can hear you thinking, "but why do we want to do that", well now we are going to use the CPU VDD Telemetry section
7a. CPU VDD_SoC Current Optimisation: Custom Setting
7b. CPU VDD Full Scale Current: The optimal EDC setting for your system as previously tested.
7c. CPU VDD Telemetry Offset Value: This needs more discussion, see below...
7d. CPU Soc Full Scale Current: Im yet to play with this...
7e. CPU Soc Telemetry Offset Value: Im yet to play with this....
CPU VDD Telemetry Offset Value
Basically I found that this value effects the balance between optimal all core performance and optimal single core performance, set the value too low and multi-core will suffer but single core will boost further, if you find the right balance you can get the best of both Worlds. For the record im not yet finished tuning this value yet buts its looking to be a relatively low value (comparative of the max input value which is 10000mA), play in the range of 8mA to 30mA if on 5600X for higher core count CPU you need to raise this range
On further testing the relationship is not as above but as follows.
The VDD telemetry offset effects the maximum all core frequency achievable dependant on the load that is being requested.
For example, Prime95 Small FFTs using AVX2, is more "heavy" on the CPU than running the CPU-Z bench test.
What this means in real numbers ??
Effect on Prime 95 Small FFTs (AVX2)
Offset: 1mA, All Core Frequency is around 4430-4460 mhz
Offset: 2999mA, All Core Frequency is around 4530-4560 mhz
Ughhh, this is getting worse, after a reboot the above was no longer valid, more Ryzen Gremlins !!
As I was in the process of getting results for screenshots I was going into BIOS and changing setting to run the tests.
I had already tested 1mA offset with AUTO PBO over two reboots and it was knocking 80-100 mhz off Prime95 Small FFTs.
But now with AUTO PBO and 1mA offset it is only loosing 20 mhz from all core frequency
Different types of loads are effected in different ways, you need to test for your specific use case.
After this amendment, I had to re-investigate on how PBO works and from preliminary testing it seems that the most flexible solution is to leave PBO on AUTO and tweak only the "CPU VDD Full Scale Current" and "CPU VDD Telemetry Offset Value" values.
Think that's it, I will update this post with images and links.
For the record
CB23: Between 1215x - 122xx
Linpack Xtreme: AVG 671 Gflops (3800/1900 14-15-14-14-26-40-240-1T)
Once ive fully tested my optimal values will run more benchmarks, want to now discuss why I went down this avenue as its pretty important if you value your time
Because of the recent information @Veii kindly shared we now have a tool that tells us a maximum value of the most important setting that is associated with how far you can extract the maximal performance of your CPU while being confined (as we cant change these 🤬🤬🤬 ) to AMD "safe" range and that is "Maximum CPU VID" and "ProcHot"
You cant aim any higher than what the maximum VID of the CPU is without loosing performance and gaining instability. You can set higher voltages, but they dont do anything to help things, just output more heat with less performance. I have not found a way to increase the maximum VID, maybe I am missing something really simple.......
![Font Rectangle Material property Parallel Pattern]()
Id imagine if you were able to increase the max VID value you could then tweak EDC slightly higher while re-tweaking your CO values while using a small BCLK increase. You wont need much BCLK as the CPUs are already near the maximum of what can be achieved from the silicon, but there is headroom for a small increase in CPU performance using BCLK since the maximal value of the CPU "Boost Override" has been gimped to 200mhz
And we must also talk about ProcHot as I saw Veii highlighted this value,at idle it is 65, but on load its limit is set to 4850 see image below there are two readings one is set to 65 (assuming thats Celcius), so I am unsure if this really is a limit to how far we can push the processor when the CPU is thermally challenged, i.e. running a standard AMD heatsink/fan.
![Font Rectangle Material property Parallel Pattern]()
Its looking that Ryzen Gremlins are not allowing me to get consistent results, looking like just leave PBO on AUTO and tweak CO ....
Warning, if you want to waste your time read further
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Note:
Some info below is relevant to my CPU which is a 5600X, those who have higher core count CPUs hope you can test the methodology outlined below to see if it holds true, same goes for other 5600X users, would be great if other could validate my findings
Prerequisites
1. Stable DRAM set
2. CO (Curve Optimiser) set to per core and each core is tuned to its most stable value while using "Boost Override" set to 200 mhz.
3a. CPU LLC: Auto
3b. CPU NB LLC: Auto
4a. vCORE: Auto
4b. vSOC/vDIMM/vDDP/vDDG: Whatever you require for your stable DRAM set
5. PPT/TDC, set to AUTO, see amendment to section 7. CPU VDD Telemetry Offset Value
This needs discussion, as im water-cooled im not concerned about restricting the "Thermal Design Current (TDC)" hence the reason for setting it to max value, also im not concerned about maximum power efficiency so again i set the "Package Power Total (PPT)" to its maximum value. I have found through my testing that setting these at maximum values does not negatively influence, in terms of performance, the maximal obtainable performance from my system when running x/y/x/ benchmark.
6. EDC
The above (striked out comment) is valid only if you have a specific use case scenario, though in such scenarios you would be better off using an all core overclock, not PBO.
All we need to be concerned about is tuning the right value for EDC and not worry about anything else.
So we go through the process of, set an EDC value, reboot, rinse and repeat, we will use CB23 multi-core and Linpack Xtreme (selection: 1, 3, 5, enter, enter) to compare results between different EDC settings.
For the record, my optimal EDC value for my 5600X sample is 110A, there is a bit of leeway between 105 to 115 so if you have OCD you can try each increment, hahahahahaha, but it wont tell you anything so why bother
7. CPU VDD Telemetry
After you got your EDC value you do the following, set EDC to double your value, yeah I can hear you thinking, "but why do we want to do that", well now we are going to use the CPU VDD Telemetry section
7a. CPU VDD_SoC Current Optimisation: Custom Setting
7b. CPU VDD Full Scale Current: The optimal EDC setting for your system as previously tested.
7c. CPU VDD Telemetry Offset Value: This needs more discussion, see below...
7d. CPU Soc Full Scale Current: Im yet to play with this...
7e. CPU Soc Telemetry Offset Value: Im yet to play with this....
CPU VDD Telemetry Offset Value
On further testing the relationship is not as above but as follows.
The VDD telemetry offset effects the maximum all core frequency achievable dependant on the load that is being requested.
For example, Prime95 Small FFTs using AVX2, is more "heavy" on the CPU than running the CPU-Z bench test.
What this means in real numbers ??
Offset: 1mA, All Core Frequency is around 4430-4460 mhz
Offset: 2999mA, All Core Frequency is around 4530-4560 mhz
Ughhh, this is getting worse, after a reboot the above was no longer valid, more Ryzen Gremlins !!
As I was in the process of getting results for screenshots I was going into BIOS and changing setting to run the tests.
I had already tested 1mA offset with AUTO PBO over two reboots and it was knocking 80-100 mhz off Prime95 Small FFTs.
But now with AUTO PBO and 1mA offset it is only loosing 20 mhz from all core frequency
Different types of loads are effected in different ways, you need to test for your specific use case.
After this amendment, I had to re-investigate on how PBO works and from preliminary testing it seems that the most flexible solution is to leave PBO on AUTO and tweak only the "CPU VDD Full Scale Current" and "CPU VDD Telemetry Offset Value" values.
Think that's it, I will update this post with images and links.
For the record
CB23: Between 1215x - 122xx
Linpack Xtreme: AVG 671 Gflops (3800/1900 14-15-14-14-26-40-240-1T)
Once ive fully tested my optimal values will run more benchmarks, want to now discuss why I went down this avenue as its pretty important if you value your time
Because of the recent information @Veii kindly shared we now have a tool that tells us a maximum value of the most important setting that is associated with how far you can extract the maximal performance of your CPU while being confined (as we cant change these 🤬🤬🤬 ) to AMD "safe" range and that is "Maximum CPU VID" and "ProcHot"
You cant aim any higher than what the maximum VID of the CPU is without loosing performance and gaining instability. You can set higher voltages, but they dont do anything to help things, just output more heat with less performance. I have not found a way to increase the maximum VID, maybe I am missing something really simple.......
Id imagine if you were able to increase the max VID value you could then tweak EDC slightly higher while re-tweaking your CO values while using a small BCLK increase. You wont need much BCLK as the CPUs are already near the maximum of what can be achieved from the silicon, but there is headroom for a small increase in CPU performance using BCLK since the maximal value of the CPU "Boost Override" has been gimped to 200mhz
And we must also talk about ProcHot as I saw Veii highlighted this value,
