MSI Vector 16 with Intel Ultra7 255HX - Missing some FIVR control option

[xyz_cityhunter]

I'm missing some FIVR control option as in screenshot below: (Missing Intel GPU, iGPU unslice, CPU E Cache)

Is there something that I missed which cause this missing FIVR Control option?
Confirmed that VBS is OFF

Other BIOS setting:
Intel Virtualization Technology - Disabled
Power & Performance/Overclocking Lock - Disabled
Power & Performance/CPU - Power Management Control/Configure CPU Lock Options/CFG Lock - Disabled
Advanced/Overclocking Performance Menu/Overclocking Feature - Enabled
Advanced/Overclocking Performance Menu/Undervolt Protection - Disabled

Meanwhile the Core Isolation looks like this:

Not sure if BIOS setting related which I'm not aware of which cause this?
Hope someone can help on this, thank you.

 

[unclewebb]

Everything VBS related looks OK. Your CPU Core and P Cache voltages are both being applied correctly. The offset values are both showing up in the FIVR monitoring table.

Is it possible that the Intel GPU is disabled? Sometimes there is an option in the BIOS to disable the Intel GPU. You might have to install MSI Center to toggle the Intel GPU on and off. That is my best guess at why both the Intel GPU and iGPU Unslice are not showing up in the FIVR window or in the FIVR monitoring table.

I saw a FIVR screenshot from a Lenovo laptop with a 275HX. I an pretty sure it was showing the Intel GPU and iGPU Unslice. It was not showing the E Cache voltage. My best guess for this is that Intel removed access to this E Cache voltage control for Core Ultra. The Core and P Cache undervolt settings seemed to be working correctly.

I do not have access to any FIVR documentation. Intel is not interested in giving their full documentation to individual programmers like myself. I also do not own or have access to any new Core Ultra hardware for testing purposes. I have no plans at the moment to update ThrottleStop for the new Core Ultra mobile processors. I have not seen anything yet that I should or could be doing differently.

Try running GPU-Z, Does it report that you have an Intel GPU?

Edit - I would set Power Limit 4 to the max, 1023. Having this limit set to only 80 is probably going to cripple performance.

 

[xyz_cityhunter]

Everything VBS related looks OK. Your CPU Core and P Cache voltages are both being applied correctly. The offset values are both showing up in the FIVR monitoring table.

Is it possible that the Intel GPU is disabled? Sometimes there is an option in the BIOS to disable the Intel GPU. You might have to install MSI Center to toggle the Intel GPU on and off. That is my best guess at why both the Intel GPU and iGPU Unslice are not showing up in the FIVR window or in the FIVR monitoring table.

I saw a FIVR screenshot from a Lenovo laptop with a 275HX. I an pretty sure it was showing the Intel GPU and iGPU Unslice. It was not showing the E Cache voltage. My best guess for this is that Intel removed access to this E Cache voltage control for Core Ultra. The Core and P Cache undervolt settings seemed to be working correctly.

I do not have access to any FIVR documentation. Intel is not interested in giving their full documentation to individual programmers like myself. I also do not own or have access to any new Core Ultra hardware for testing purposes. I have no plans at the moment to update ThrottleStop for the new Core Ultra mobile processors. I have not seen anything yet that I should or could be doing differently.

Try running GPU-Z, Does it report that you have an Intel GPU?

Edit - I would set Power Limit 4 to the max, 1023. Having this limit set to only 80 is probably going to cripple performance.

Thanks, confirmed the switching from Discrete Graphic Mode back to MSHybrid Graphics Mode does make the Intel GPU shows up

Noted on the PL4, didn't knew that was a problem as it was the default value. I have update it to 1023.

1 question on the Offset voltage on CPU Core and CPU P Cache: Does it really need both to be same?
I noticed like some older Intel system is able to have like 60mV offset on CPU P cache while having much larger 150mV offset on CPU Core so I was thinking if I misunderstood that both need to be same?

And if I don't see "HOT" or "POWER" show up beside the Limits button after 10minutes of Cinebench R23, does this means there's no power or thermal throttling happened?
Like this one the Max temperature after 10minutes of Cinebench, it recorded 102C but the duration not long until it trigger the thermal throttling?

 

[unclewebb]

Try using ThrottleStop 9.7.3. It is the latest version that in theory should best support a Core Ultra 7 255HX.

Intel® Core™ Ultra 7 Processor 255HX (30M Cache, up to 5.20 GHz) - Product Specifications | Intel

Intel® Core™ Ultra 7 Processor 255HX (30M Cache, up to 5.20 GHz) quick reference with specifications, features, and technologies.

www.intel.com

The maximum safe operating temperature for these CPUs is 105°C. Some manufacturers set a value called PROCHOT Offset. It appears your CPU has set PROCHOT Offset to 5 which tells the CPU to start thermal throttling at 100°C instead of the full 105°C. If this value is not locked by the BIOS, it can be adjusted in the TPL window of TS 9.7.3. Having this set to 5 is OK.

switching from Discrete Graphic Mode back to MSHybrid Graphics Mode does make the Intel GPU shows up

Thanks for confirming that. Some CPUs have separate FIVR settings for the Intel GPU and the iGPU Unslice. Some only have a single setting for the Intel GPU. My 14900HX has both settings which I believe need to be adjusted equally. I set them both to an offset of -100 mV. It did not crash. I did not do any testing beyond that.

Undervolting the Intel GPU in theory is not that important. If you are playing games in Hybrid mode, the Nvidia GPU will be doing all of the hard work. The Intel GPU will rarely if ever be significantly loaded. There is usually little to be gained by undervolting the Intel GPU.

When PROCHOT 100°C lights up red, that means your CPU did some thermal throttling. It might have only needed to thermal throttle for a few milliseconds. That is enough to trigger this flag within the CPU. Even if ThrottleStop is not running at the time, this flag can still get triggered. As soon as you start ThrottleStop it will show if there was any thermal throttling. You can click on where it says PROCHOT 100°C at anytime to reset this flag.

And if I don't see "HOT" or "POWER" show up beside the Limits button after 10minutes of Cinebench R23, does this means there's no power or thermal throttling happened?

HOT or POWER showing up beside the Limits button is not always the most reliable indicator. That data is only being updated once per second. It is possible that you can have some very brief thermal or power limit throttling and these indicators will never be triggered.

I like running a log file when I am testing. That provides a good record of any throttling so you do not have to constantly look at ThrottleStop. The CORE column of Limit Reasons is also worth watching when Cinebench is running.

Does it really need both to be same?

I have never tested a 255HX or anything similar. I do not know if the core and P cache offset voltages need to be set to the same offset value. The voltage being fed to the core and the P cache is often times coming from the exact same voltage rail within the CPU. Requesting a much larger undervolt for the CPU Core might not make any difference. The values in the ThrottleStop FIVR window are only voltage requests. The CPU gets to decide what to do with your requests. If you request something really crazy or out of spec, the CPU will likely or hopefully ignore your request.

If you can prove that setting these two offset voltages to different values is making a difference then continue doing what you are doing. Watch for any changes in temperature or improvements in performance. Everything else is not important.

 

[xyz_cityhunter]

Try using ThrottleStop 9.7.3. It is the latest version that in theory should best support a Core Ultra 7 255HX.

Intel® Core™ Ultra 7 Processor 255HX (30M Cache, up to 5.20 GHz) - Product Specifications | Intel

Intel® Core™ Ultra 7 Processor 255HX (30M Cache, up to 5.20 GHz) quick reference with specifications, features, and technologies.

www.intel.com

The maximum safe operating temperature for these CPUs is 105°C. Some manufacturers set a value called PROCHOT Offset. It appears your CPU has set PROCHOT Offset to 5 which tells the CPU to start thermal throttling at 100°C instead of the full 105°C. If this value is not locked by the BIOS, it can be adjusted in the TPL window of TS 9.7.3. Having this set to 5 is OK.


Thanks for confirming that. Some CPUs have separate FIVR settings for the Intel GPU and the iGPU Unslice. Some only have a single setting for the Intel GPU. My 14900HX has both settings which I believe need to be adjusted equally. I set them both to an offset of -100 mV. It did not crash. I did not do any testing beyond that.

Undervolting the Intel GPU in theory is not that important. If you are playing games in Hybrid mode, the Nvidia GPU will be doing all of the hard work. The Intel GPU will rarely if ever be significantly loaded. There is usually little to be gained by undervolting the Intel GPU.

When PROCHOT 100°C lights up red, that means your CPU did some thermal throttling. It might have only needed to thermal throttle for a few milliseconds. That is enough to trigger this flag within the CPU. Even if ThrottleStop is not running at the time, this flag can still get triggered. As soon as you start ThrottleStop it will show if there was any thermal throttling. You can click on where it says PROCHOT 100°C at anytime to reset this flag.


HOT or POWER showing up beside the Limits button is not always the most reliable indicator. That data is only being updated once per second. It is possible that you can have some very brief thermal or power limit throttling and these indicators will never be triggered.

I like running a log file when I am testing. That provides a good record of any throttling so you do not have to constantly look at ThrottleStop. The CORE column of Limit Reasons is also worth watching when Cinebench is running.


I have never tested a 255HX or anything similar. I do not know if the core and P cache offset voltages need to be set to the same offset value. The voltage being fed to the core and the P cache is often times coming from the exact same voltage rail within the CPU. Requesting a much larger undervolt for the CPU Core might not make any difference. The values in the ThrottleStop FIVR window are only voltage requests. The CPU gets to decide what to do with your requests. If you request something really crazy or out of spec, the CPU will likely or hopefully ignore your request.

If you can prove that setting these two offset voltages to different values is making a difference then continue doing what you are doing. Watch for any changes in temperature or improvements in performance. Everything else is not important.

Thanks for the explanation.
Seems like version9.7.3 is indeed better as I encounter issue where I cannot click and view the Turbo Group of Efficiency Cores in version 9.6

I will mostly move to discrete graphic only as I'm pretty much using it as a desktop replacement, and seems like the TBT5 Type-C is actually output from the NVIDIA GPU so would like to connect to G-sync capable monitor to see if that can work.

Hm, kind of interesting that the PROCHOT 100°C will have red flag when I tried 10minutes Cinebench run and the highest temp record was only 91°C max.

I did press the CLR button to get fresh data log before starting the 10minutes run.

Still need to spend some time to find optimum setting, the -80.1 offset is capable to run the 10minutes but not the 30minutes.
Still want to experiment to see if the Turbo ratios can help on stability but seems like this new 200 series is not so great to undervolt as I read some comment in Reddit where the limit for higher Ultra9 275HX is only -50mV.

 

[unclewebb]

kind of interesting that the PROCHOT 100°C will have red flag when I tried 10minutes Cinebench run

I will assume that you clicked on PROCHOT and reset that flag before you started running Cinebench.

There are multiple temperature sensors within the CPU package. If PROCHOT lights up red during Cinebench, it is possible that a temperature sensor not located on one of the cores could have triggered this. That sensor may or may not be as accurate as the core temperature sensors. It could be more accurate.

Try watching Limit Reasons while testing. Does THERMAL ever light up yellow or red during a Cinebench run? Check the ThrottleStop Log File box and run a log file when testing. It would be interesting to see if there is any indication of any throttling showing up in the log file. Exit ThrottleStop when finished testing and attach a log file to your next post.

If PROCHOT Offset is not locked in the TPL window, you can set that to 0 for a little more thermal headroom. I usually lock this setting to 0 or to 1.

I encounter issue where I cannot click and view the Turbo Group of Efficiency Cores in version 9.6

Thanks for reporting that bug. When I do not have access to new hardware, minor bugs like that can happen. It is difficult to fix bugs if they never get reported. TS 9.7.3 seems to be working correctly.

the -80.1 offset is capable to run the 10minutes but not the 30minutes.

Does your computer crash or does Cinebench crash while the CPU is fully loaded or partially loaded? Did you set V/F Point 1 to 80 or 100?

where the limit for higher Ultra9 275HX is only -50mV.

That is actually a good thing. That is typical of most Intel CPUs. The default voltage for some of the 14th Gen HX processors was not just too high. It was WAY too high! I can undervolt my 14900HX by -150 mV . That is a good indication that someone at Intel created a voltage curve that was well beyond what the CPU actually needs to run stable. No surprise that Intel had so many problems with their 13th and 14th Gen CPUs. "Just give 'er some more voltage" was probably not the best solution.

 

[xyz_cityhunter]

I will assume that you clicked on PROCHOT and reset that flag before you started running Cinebench.

There are multiple temperature sensors within the CPU package. If PROCHOT lights up red during Cinebench, it is possible that a temperature sensor not located on one of the cores could have triggered this. That sensor may or may not be as accurate as the core temperature sensors. It could be more accurate.

Try watching Limit Reasons while testing. Does THERMAL ever light up yellow or red during a Cinebench run? Check the ThrottleStop Log File box and run a log file when testing. It would be interesting to see if there is any indication of any throttling showing up in the log file. Exit ThrottleStop when finished testing and attach a log file to your next post.

If PROCHOT Offset is not locked in the TPL window, you can set that to 0 for a little more thermal headroom. I usually lock this setting to 0 or to 1.

View attachment 401728

Ok, will make try the log and see what I will get. Not sure if I get that from previous run and I just didn't noticed it. And I've already set the PROCHOT Offset to 0.

Does your computer crash or does Cinebench crash while the CPU is fully loaded or partially loaded? Did you set V/F Point 1 to 80 or 100?

Totally no issue when running the 10minutes run, I was away when I tried the 30minutes run and came back to see it BSOD. I believe the V/F for 800Mhz was set to 160 but will reconfirm later when I got the time to test again.

I really need to make a table of what setting tested vs outcome to avoid overlap testing...

 

[xyz_cityhunter]

Finally found some stable voltage region for 30minutes run, -75.2mV is able to complete 30minutes run without issue but still can observe some sluggish response in game menu (Peak at single core I guess) so go down to -70.3mV and can see some response improvement.
Here's my current setting, I reduce the default P core boost range of 52 - 50, down to 49 - 46 for undervolt stability.

While only reduce the E core ratio of constant 45, to 45 - 44 for good temperature and also maintain performance level.

From my testing, I found that most test all P core with 49-46 ratio is stable when undervolting, I tried pushing it to 50-47 but will see Cinebench crashed in 30minutes run instead.
While for E core, I noticed that ratio 45 can provide uplift in performance but output really high temps while ratio 44 still can maintain decent performance with lower temps.
Can see the temps below that when I tried to lock ratio 45 for all E cores, it end up reporting Max temperature, while P cores (Ratio 46) are mostly 98 and below:

I believe this is a good stopping point, next would be to optimize the GPU
Do share if you can think of anything else to better optimize this.

 

[xyz_cityhunter]

@

Does your computer crash or does Cinebench crash while the CPU is fully loaded or partially loaded? Did you set V/F Point 1 to 80 or 100?

I got questions on the V/F Point, is the points fixed or is can be edited to more MHz points?
My current one seems like limited to just these few, would like to add more like 5000MHz, 4900MHz and so on:

And the Cache V/F seems like different MHz value, should they be the same?

 

[unclewebb]

questions on the V/F Point

The V/F points are set by Intel and cannot be changed. You cannot add or delete V/F points and you cannot modify the MHz. The Core and Cache V/F points are set to different frequencies. I am not sure why Intel does this. It is possible for the highest V/F point to change frequencies when overclocking but that is about it.

Think of V/F points like individual points on a curve. If you add or remove a significant amount of voltage at one point along the curve, that can add or remove voltage at other near by V/F points on the curve. This curve will automatically be adjusted so the VID voltage is constantly increasing as the MHz increases.

Here is an example of what ThrottleStop does when adding voltage at V/F Point 1 at 800 MHz. It gives the voltage curve a little bump up at low MHz. This has proven to help stability when undervolting.

 

[xyz_cityhunter]

The V/F points are set by Intel and cannot be changed. You cannot add or delete V/F points and you cannot modify the MHz. The Core and Cache V/F points are set to different frequencies. I am not sure why Intel does this. It is possible for the highest V/F point to change frequencies when overclocking but that is about it.

Think of V/F points like individual points on a curve. If you add or remove a significant amount of voltage at one point along the curve, that can add or remove voltage at other near by V/F points on the curve. This curve will automatically be adjusted so the VID voltage is constantly increasing as the MHz increases.

Here is an example of what ThrottleStop does when adding voltage at V/F Point 1 at 800 MHz. It gives the voltage curve a little bump up at low MHz. This has proven to help stability when undervolting.

View attachment 402332

Ok, I guess the Offset Voltage will move the whole curve down while this V/F Point will compensate per operating frequency.
Suppose like in FIVR, I put -70.3mV as Offset Voltage with Turbo Ratio range from 52 to 46 but my previous testing shows that the upper 52-50 range isn't stable:

I then compensate some mV in the V/F Point, let's say I need 70mV to be have it stable when it's in 50-52 range, do I input it like this?

Suppose I leave the 4800 as empty then it should follow the -70.3mV from default value while having gradient increase to -0.3mV at 5100?

And I assume the cache having same number of V/F levels so I should duplicate the same mV compensate on Core to Cache...

Not sure if it's a display bug but somehow I noticed the V/F Core MHz can change when I was typing this while the ThrottleStop is in view of another monitor.

I also read some mentioned on the Per Core Ratio Limits, I believe this will limit per core operating frequency but how to know which P Core number?
Do we need to use HWinfo while stress test to know which core can boost higher while also monitor the core temperature along with surrounding E core?

 

[unclewebb]

testing shows that the upper 52-50 range isn't stable:

If you are trying to use a -70 mV global undervolt and it is not 100% stable when the CPU is loaded then I would reduce that undervolt. Some people with similar CPUs are not stable at anything beyond -45 mV to -50 mV.

Suppose I leave the 4800 as empty then it should follow the -70.3mV from default value while having gradient increase to -0.3mV at 5100?

I have never tested a Core Ultra CPU. I spent a day or two figuring out my 14900HX. It has been very stable with V/F Point 1 set to 150 and the Core and the P Cache set to -150 mV. Trying to come up with the perfect settings and figure out how everything works is no longer that important to me. I like to keep things simple. Set it, test it and forget it is what I do.

Using a -70 mV global undervolt and then adding all of that back in by entering a 70 mV V/F Point value seems to cancel your undervolt. Canceling the undervolt at 800 MHz makes sense because many 14th Gen HX processors are not stable when lightly loaded when trying to use a really big global undervolt. Boosting the low MHz voltage helped to balance things out. The Core Ultra is a different beast. Users will have to figure out what works best.

And I assume the cache having same number of V/F levels so I should duplicate the same mV compensate on Core to Cache...

The core and cache V/F points occur at different frequencies. Duplicating all of the values does not seem necessary.

All of these various settings are just voltage requests. The CPU uses some advanced algorithm to determine how much voltage should go to the core and the P cache. I do not know anything about this Intel formula. That means it is all trial and error to find out what works best.

a display bug but somehow I noticed the V/F Core MHz can change when I was typing this while the ThrottleStop

It could be a bug or it could just be how these new CPUs work. Always keep in mind that ThrottleStop was not designed for Core Ultra processors. ThrottleStop has never been fully tested on these new processors. If it works, great. If any feature does not work correctly then do not use that feature or do not use ThrottleStop at all. Try using Intel XTU instead. Their programmers have access to the full Intel documentation and lots of modern hardware for testing purposes.

I am quite happy with my 14900HX. It could be years before I ever buy a Core Ultra HX processor and update ThrottleStop for these.

how to know which P Core number?

I would put a load on the CPU and then I would change the Per Core values one at a time. You can run HWiNFO and watch to see if changing these values limits some cores of the CPU and you can observe what cores get limited. You can also look at the main ThrottleStop monitoring table to see this data.

ThrottleStop shows that your two best cores appear to be P Core 0 and P Core 2. At default settings, these two should be loaded first since these are your best cores. They can reliably run a little faster compared to all of the other P cores.