Yea, IIRC you can set an offset for the ring voltage (Cache SVID). But of course that may or may not be stable. I'd just raise it as high as I could without increasing vCore like I said originally. Gains are small from increasing ring, so just get the free frequency and call it good as has already been said I think.

This seems to have worked!

My Cache Svid Voltage has been just default Auto this whole time (back in the day at 46x I remember having it set to 1.2-1.25v range before but I didn't know if it did anything because I hadn't learned yet that the voltages were shared).

I set it to Adaptive, left the Turbo Additional Voltage to Auto, and just made a -0.030mv Offset to it. Because that's (close to) how much the default 50x Ring ups my R23/Load Voltage from 48x.

Booting and idling windows = OK.

R23 run... there! Load voltage back to 1.225-1.232vCore (VID 1.229-1.230), max 93c (3 hot cores, the rest are only 80c) over 3 runs with breaks in between, 234 Amps max and 285-291 Watts max - all spot on to the voltages at 46 & 48x Ring w/Auto Cache Svid.

Idling + 3 R23 runs is promising [...]
* edit
5x passes of GB6, and 2x TLOU Shader Compiling completed successfully, too (same 1.26v avg @ 205 A / 250W max as 48x Ring)!

I assume increasing the offset to 0.031, 0.032, etc. Won't make a difference because my vCore will be the one requesting the most from there.

 

My experience was the opposite, 46-48 ring didn't impact voltage requirements at all, but going to 48 to 50 caused load voltage (ie shader compiling, r23) to raise by +0.030mv. Ie. 1.26 to 1.29, and 1.225 to 1.252.

I don't really know how to force the ring to request less, I read another reply saying I could set a minus offset to Cache SVID Voltage setting (asus board) which I may try. But then it might not be stable of course, if it works, since I'd be cutting 30mv off what if defaults to requesting.

This seems to have worked!

My Cache Svid Voltage has been just default Auto this whole time (back in the day at 46x I remember having it set to 1.2-1.25v range before but I didn't know if it did anything because I hadn't learned yet that the voltages were shared).

I set it to Adaptive, left the Turbo Additional Voltage to Auto, and just made a -0.030mv Offset to it. Because that's (close to) how much the default 50x Ring ups my R23/Load Voltage from 48x.

Booting and idling windows = OK.

R23 run... there! Load voltage back to 1.225-1.232vCore (VID 1.229-1.230), max 93c (3 hot cores, the rest are only 80c) over 3 runs with breaks in between, 234 Amps max and 285-291 Watts max - all spot on to the voltages at 46 & 48x Ring w/Auto Cache Svid.

Idling + 3 R23 runs is promising [...]
* edit
5x passes of GB6, and 2x TLOU Shader Compiling completed successfully, too (same 1.26v avg @ 205 A / 250W max as 48x Ring)!

I assume increasing the offset to 0.031, 0.032, etc. Won't make a difference because my vCore will be the one requesting the most from there.

Just make sure you actually see the desired ring clocks in HWiNFO and that they are not getting throttled.

 

Just make sure you actually see the desired ring clocks in HWiNFO and that they are not getting throttled.

Yeah during R23, gaming, shader compiling, etc. I see it sitting flat at 5000mhz the whole time

It's been a whole night of light gaming & browsing as well so far since that last post - no errors logged or funky behaviour. All is good so far. Since the -0.030mv offset to Cache Svid Voltage brought back the R23 load voltage to 1.225vCore as was needed to pass repack decompression (originally 1.217 was fine for R23), which also fixed GB6 randomly erroring out, I don't think making the offset any bigger will do anything since it seems right now, the Core Voltage is the strongest vid request out of all.


I just wish I knew how to lower my max idle vid. None of my voltages in any way shape or form come close to the voltage it peaks up to when idling or afk in light load situations. Spin up a light task like Super Pi 1M bouncing across single cores? Easy, 1.5-1.525v as-set in the VF#10 1.474+0.050 offset as expected. Sitting there on desktop? See fluctuations up to 1.56, 1.57, 1.58 on the VID requests.

The SVID Adaptive Voltage hasn't seemed to do anything (I imagine it's just getting ignored) - 1.5, 1.49, 1.48, no diff. It SEEMS like when I raised "ACLL" up to fix load voltage, it also upped that max idle range.

 

Yeah during R23, gaming, shader compiling, etc. I see it sitting flat at 5000mhz the whole time

It's been a whole night of light gaming & browsing as well so far since that last post - no errors logged or funky behaviour. All is good so far. Since the -0.030mv offset to Cache Svid Voltage brought back the R23 load voltage to 1.225vCore as was needed to pass repack decompression (originally 1.217 was fine for R23), which also fixed GB6 randomly erroring out, I don't think making the offset any bigger will do anything since it seems right now, the Core Voltage is the strongest vid request out of all.


I just wish I knew how to lower my max idle vid. None of my voltages in any way shape or form come close to the voltage it peaks up to when idling or afk in light load situations. Spin up a light task like Super Pi 1M bouncing across single cores? Easy, 1.5-1.525v as-set in the VF#10 1.474+0.050 offset as expected. Sitting there on desktop? See fluctuations up to 1.56, 1.57, 1.58 on the VID requests.

The SVID Adaptive Voltage hasn't seemed to do anything (I imagine it's just getting ignored) - 1.5, 1.49, 1.48, no diff. It SEEMS like when I raised "ACLL" up to fix load voltage, it also upped that max idle range.

Yea ACLL will bring everything up.

 

Yeah during R23, gaming, shader compiling, etc. I see it sitting flat at 5000mhz the whole time

It's been a whole night of light gaming & browsing as well so far since that last post - no errors logged or funky behaviour. All is good so far. Since the -0.030mv offset to Cache Svid Voltage brought back the R23 load voltage to 1.225vCore as was needed to pass repack decompression (originally 1.217 was fine for R23), which also fixed GB6 randomly erroring out, I don't think making the offset any bigger will do anything since it seems right now, the Core Voltage is the strongest vid request out of all.


I just wish I knew how to lower my max idle vid. None of my voltages in any way shape or form come close to the voltage it peaks up to when idling or afk in light load situations. Spin up a light task like Super Pi 1M bouncing across single cores? Easy, 1.5-1.525v as-set in the VF#10 1.474+0.050 offset as expected. Sitting there on desktop? See fluctuations up to 1.56, 1.57, 1.58 on the VID requests.

The SVID Adaptive Voltage hasn't seemed to do anything (I imagine it's just getting ignored) - 1.5, 1.49, 1.48, no diff. It SEEMS like when I raised "ACLL" up to fix load voltage, it also upped that max idle range.

If you use less droopy llc it will idle at a lower voltage.

 

Okay so it's the ACLL doing it! I had a hunch, but my understanding in Roberto's guide was I could lower max idle by setting the svid adaptive lower (even if VF 10 / a point before the OC Ratio VF 11 was a higher voltage), up until the innate VID limit (ie 1.474 for me @ 60). It never seemed to do anything... so I guess it isn't doing anything.

I have to be on LLC-4 as LLC 5 I could not reduce ACLL lower than 0.01 anymore to counter the load voltage that the offset on VF-10 would produce (if I left VF 7-8 on no or very little adjusted offset).

But it leaves me with a question, I can also control the full load voltage by manipulating offsets on VF 7 & VF 8 (5400mhz and 5800mhz respectively) and have it interpolate my 5600 full load voltage to be at 1.225v. In that situation... What is the difference or point there of ACLL?

So if I wish to lower my max idle vid to something more reasonable (again no idea why it's spiking up so high during idle, I don't like it though) or closer to my vf10+offset level, I need to reduce ACLL below my current "0.21 / 1.01" setup - down to say 0.15. Then my load voltage goes down, so I could increase VF 7's (5400) offset from +0.010 to like +0.025. Is there a downside to that? It seems that both tools serve the same function, to me.

 

Okay so it's the ACLL doing it! I had a hunch, but my understanding in Roberto's guide was I could lower max idle by setting the svid adaptive lower (even if VF 10 / a point before the OC Ratio VF 11 was a higher voltage), up until the innate VID limit (ie 1.474 for me @ 60). It never seemed to do anything... so I guess it isn't doing anything.

I have to be on LLC-4 as LLC 5 I could not reduce ACLL lower than 0.01 anymore to counter the load voltage that the offset on VF-10 would produce (if I left VF 7-8 on no or very little adjusted offset).

But it leaves me with a question, I can also control the full load voltage by manipulating offsets on VF 7 & VF 8 (5400mhz and 5800mhz respectively) and have it interpolate my 5600 full load voltage to be at 1.225v. In that situation... What is the difference or point there of ACLL?

So if I wish to lower my max idle vid to something more reasonable (again no idea why it's spiking up so high during idle, I don't like it though) or closer to my vf10+offset level, I need to reduce ACLL below my current "0.21 / 1.01" setup - down to say 0.15. Then my load voltage goes down, so I could increase VF 7's (5400) offset from +0.010 to like +0.025. Is there a downside to that? It seems that both tools serve the same function, to me.

There are multiple ways to change voltage that give similar results. Sometimes combining them can help, which you can definitely do. You can combine ACLL, VF point offsets, and a global offset to try to get the curve where you want it.

Do you have TVB voltage optimizations enabled? That will reduce your low load voltage a lot as it drops vCore based on temperature. So you'll get about the same full load voltage that runs at higher temps, but it'll drop idle/low load quite a lot.

 

There are multiple ways to change voltage that give similar results. Sometimes combining them can help, which you can definitely do. You can combine ACLL, VF point offsets, and a global offset to try to get the curve where you want it.

Do you have TVB voltage optimizations enabled? That will reduce your low load voltage a lot as it drops vCore based on temperature. So you'll get about the same full load voltage that runs at higher temps, but it'll drop idle/low load quite a lot.

OK, I'll try reduce my ACLL and then up VF7 (the pt for freq. 5400, there is none for exactly 5600mhz) to compensate. So this should then reduce the very high idle voltage )but will need some tweaking to find what gives me 1.225v under load again) and will do the same thing as ACLL for my load voltage. I guess 54-55-57 mhz ratios get given more voltage than they need, but their VID is so low it doesn't matter too much I guess. I wonder what the consequence or benefit of purely ACLL tuning is then? If one sets it to 0.01, then just fixes their load voltage via VF 7-8, it is effectively the same thing, to me. I've always just done AC/DC & then tuned VF 10 (which also raises load voltage, requiring less ACLL...).

I currently have VF 8 (5800) set to +0.010 but I believe it's unnecessary since I did it initially when fixing decompression WHEA's thinking it was one of the OCTVB freqs. failing (it wasn't).

Also, no TVB Optimizations. I disabled both it and something called "Enhanced TVB" when trying to fix why my OCTVB wasn't working as intended (max voltage was only ever 1.45v as well, no matter what was set). TVB Opt wasn't the issue in the end, but I read it can cause instability when ambients change a lot.

 

OK, I'll try reduce my ACLL and then up VF7 (the pt for freq. 5400, there is none for exactly 5600mhz) to compensate. So this should then reduce the very high idle voltage )but will need some tweaking to find what gives me 1.225v under load again) and will do the same thing as ACLL for my load voltage. I guess 54-55-57 mhz ratios get given more voltage than they need, but their VID is so low it doesn't matter too much I guess.

I currently have VF 8 (5800) set to +0.010 but I believe it's unnecessary since I did it initially when fixing decompression WHEA's thinking it was one of the OCTVB freqs. failing (but it wasn't as all problems solved there after upping ACLL enough points).

Also, no TVB Optimizations. I don't know if I should re enable it, I disabled both it and something called "Enhanced TVB" when trying to fix why my OCTVB wasn't working as intended (max voltage was only ever 1.45v as well, no matter what was set). TVB Opt wasn't the issue, but I read it can cause instability when ambients change a lot.

Yea, I've seen that comment too, but in reality it shouldn't unless it's changing the voltage too much (which it could be). You really want voltage to adjust with temp, particularly if temps go up.

 

Yea, I've seen that comment too, but in reality it shouldn't unless it's changing the voltage too much (which it could be). You really want voltage to adjust with temp, particularly if temps go up.

Alright, I'll try simply turn it back on, no other acll/vf/etc changes and see how it goes... I worry about it because I Just like to be in control of exactly what voltages are happening (which is why this high idle voltage is annoying me lol) and I worry that - and I know 1.225v 100% load is stable - it will make the voltage go up to 1.232, or 1.243 etc when it doesn't need to.

 

Alright, I'll try simply turn it back on, no other acll/vf/etc changes and see how it goes... I worry about it because I Just like to be in control of exactly what voltages are happening (which is why this high idle voltage is annoying me lol) and I worry that - and I know 1.225v 100% load is stable - it will make the voltage go up to 1.232, or 1.243 etc when it doesn't need to.

What happens if you leave all voltage controls on auto and only enable all tvb controls?

 

What happens if you leave all voltage controls on auto and only enable all tvb controls?

if i set all vf points to auto and re-enable tvb opt and enhanced tvb settings the system is unstable

 

Yea, I've seen that comment too, but in reality it shouldn't unless it's changing the voltage too much (which it could be). You really want voltage to adjust with temp, particularly if temps go up.

Also FWIW: the maximum idle states voltage it peaks up to is 1.55-1.58. Super Pi 1M / actual light load for the 61x4 range, is correctly running at 1.5-1.525 during. Then as it hops between cores or goes back to idle etc it gets those spikes again.

When GAMING, for example something moderately heavy like Ratchet & Clank Rift Apart - it's basically 1.33-1.39v bouncing around the entire time. Gaming is mostly in that 1.3-1.4v range, in fact. Every now and then i'll see a core go to 61/60 and bring the voltage up to 1.4X or 1.5~ for a split second. Temps wise it's primarily 40-55c in most of the games I play, with spikes to 65-75 range (OCTVB limits) in those light load moments. IE: 2 hours of Black Desert Online (an MMO) = average 40-44c, max on cores from 68c to 79c. Amps max 135, avg 49. Really low across the board there, so I think even with this high idle voltage everything else is "okay/safe".

 

Also FWIW: the maximum idle states voltage it peaks up to is 1.55-1.58. Super Pi 1M / actual light load for the 61x4 range, is correctly running at 1.5-1.525 during. Then as it hops between cores or goes back to idle etc it gets those spikes again.

When GAMING, for example something moderately heavy like Ratchet & Clank Rift Apart - it's basically 1.33-1.39v bouncing around the entire time. Gaming is mostly in that 1.3-1.4v range, in fact. Every now and then i'll see a core go to 61/60 and bring the voltage up to 1.4X or 1.5~ for a split second. Temps wise it's primarily 40-55c in most of the games I play, with spikes to 65-75 range (OCTVB limits) in those light load moments. IE: 2 hours of Black Desert Online (an MMO) = average 40-44c, max on cores from 68c to 79c. Amps max 135, avg 49. Really low across the board there, so I think even with this high idle voltage everything else is "okay/safe".

Well FWIW, a lot of the 14900ks' hit voltages at least close to that. On the latest bios, mine sits in the bios at 1.51v. Not saying that's good necessarily, but somebody at Intel/Asus must think so.

 

All this work and tuning is exactly why i don't bother with by core settings and dealing with how ACLL affects V/F points (ACLL is NOT predicted current anymore, this stopped when Z490 came out, yet all the people on the chiphell chinese forums still thinks that's how ACLL works, and I can't register there to tell them in English that they're wrong). And I do not think that 1.51v at idle or very low loads on one or two cores is safe for those cores. All this tuning and klingon pain sticks testing is just too much work, when I can spend that time losing at chess over and over.

 

All this work and tuning is exactly why i don't bother with by core settings and dealing with how ACLL affects V/F points (ACLL is NOT predicted current anymore, this stopped when Z490 came out, yet all the people on the chiphell chinese forums still thinks that's how ACLL works, and I can't register there to tell them in English that they're wrong). And I do not think that 1.51v at idle or very low loads on one or two cores is safe for those cores. All this tuning and klingon pain sticks testing is just too much work, when I can spend that time losing at chess over and over.

I'd love to be able to do a real per core overclock like the MSI board that Ichirou has, that's what I actually imagined it being in the first place. I think this OCTVB stuff is so much fluff. I'd rather pure per core, with the thermal downbinning part of octvb. But as far as I know... by core usage is all I have for Asus. Could be wrong though.

It's simple enough now I have the hang of it and learned a lot, the only foe is just lowering voltages slowly on offsets and tackling the increased idle voltage that it seems a higher acll has given me. In which case it seems working off exclusively vf offsets would have been better. I've used your 1520-(amps*dcll) formula for staying away from load degradation a lot and its been a good help to keep things in check on that front.

 

I'd love to be able to do a real per core overclock like the MSI board that Ichirou has, that's what I actually imagined it being in the first place. I think this OCTVB stuff is so much fluff. I'd rather pure per core, with the thermal downbinning part of octvb. But as far as I know... by core usage is all I have for Asus. Could be wrong though.

It's simple enough now I have the hang of it and learned a lot, the only foe is just lowering voltages slowly on offsets and tackling the increased idle voltage that it seems a higher acll has given me. In which case it seems working off exclusively vf offsets would have been better. I've used your 1520-(amps*dcll) formula for staying away from load degradation a lot and its been a good help to keep things in check on that front.

At some point last year, MSI updated their BIOSes to support ring clock ranges, which is basically the same as ASUS's. But no Cache SVID Voltage tweak though (if it even works as intended and isn't just some visual VID change). But the old BIOS I'm currently on works better for memory overclocking, so I haven't updated it.

I can run 52x ring just fine anyway at around 1.35V or so, so there's no real reason to have to clock it down.

 

All this work and tuning is exactly why i don't bother with by core settings and dealing with how ACLL affects V/F points (ACLL is NOT predicted current anymore, this stopped when Z490 came out, yet all the people on the chiphell chinese forums still thinks that's how ACLL works, and I can't register there to tell them in English that they're wrong). And I do not think that 1.51v at idle or very low loads on one or two cores is safe for those cores. All this tuning and klingon pain sticks testing is just too much work, when I can spend that time losing at chess over and over.

I put a ticket in with Intel about my 14900ks. After they suggested the idea of setting SVID behavior to Intel Failsafe, I sent screenshots of it running at absurd vCore and asked if it was safe for the lifetime of the CPU. They have been completely silent since then. Coming up on 2 weeks I think. I sent another message prodding them, but nothing.

 

At some point last year, MSI updated their BIOSes to support ring clock ranges, which is basically the same as ASUS's. But no Cache SVID Voltage tweak though (if it even works as intended and isn't just some visual VID change). But the old BIOS I'm currently on works better for memory overclocking, so I haven't updated it.

I can run 52x ring just fine anyway at around 1.35V or so, so there's no real reason to have to clock it down.

@Ichirou For me, the Cache SVID Voltage Offset did work. It seems like the -0.030mv offset to it (roughly equal to what I noticed my load voltages go up from setting 50x ring from 48x ring with no other changes), leaving the actual voltage on auto, counter-acted the increased load voltage that 50x ring's VID requested, back to my normal voltage. And fortunately enough for me it is still stable like this, even as I am tweaking down AC_LL/VF-10 offset and shifting load voltage sourced via VF7 offset instead. It feels good to navigate that issue with that setting though! Before I never really knew what Cache SVID Voltage setting did.

Performance wise... 46->50 Ring.. I don't know... Maybe like 10~ more average points on Single Core GB6 lol (from 3310-3342 best range, to like 3320-3346 more commonly) and a more stable 23900-24100 multi core. Beyond 50x is probably not viable given that I needed to set this offset already though. I have Ring Down Bin enabled too, and lowest Ring to "Auto" as well, because I found that, back on 46x, my lowest voltage wouldn't be lower than 1.18~. Now I go down to 0.67~ lowest as it properly clocks down.

@yzonker I've seen stock VID screenshots of people's 14900K/KS's VF Curves and many of their 6200mhz boosts are already 1.5-1.53v natively. Then there's also that Intel rep saying 1720mv is their absolute max limit for 13th (Voltage of I9-13900k.) but this is of course idle and I don't think anyone would want to come anywhere near 1.6v+ idle!

 

.... is correctly running at 1.5-1.525 ...

There is nothing "correct" about these chips ever seeing such high voltages.

"I've seen stock VID screenshots of people's 14900K/KS's VF Curves and many of their 6200mhz boosts are already 1.5-1.53v natively. "

It's not safe, because the stock mobo settings aren't safe in the first place, e.g.
https://videocardz.com/newz/falcon-...has-a-guide-for-mitigating-stability-issues-with-intel-raptor-lake-core-i9-cpus

https://videocardz.com/newz/nvidia-...mmends-intel-raptor-lake-cpu-users-to-consult-intel-in-case-of-stability-issues

Best option is to set your max all-core frequency at a voltage you are happy with and avoid any high voltage spikes (and minimise your voltage as much as you can for that particular frequency); that is if you want your chip to last. Boosting for nanoseconds at superhigh clocks isn't meaningfully worth it since any serious load (i.e. when you need your CPU) will bring you down the all-core frequency anyway. The aim with 13th and 14th gen is to keep your voltage and power as low as possible.