• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Asus Z790 14900ks throttling question.

Joined
Mar 7, 2007
Messages
3,994 (0.60/day)
Location
Maryland
System Name HAL
Processor Core i9 14900ks @5.9-6.3
Motherboard Z790 Dark Hero
Cooling Bitspower Summit SE & (2) 360 Corsair XR7 Rads push/pull
Memory 2x 32GB (64GB) Gskill trident 6000 CL30
Video Card(s) RTX 4090 Gigagbyte gaming OC @ +200/1300
Storage (M2's) 2x Samsung 980 pro 2TB, 1xWD Black 2TB, 1x SK Hynix Platinum P41 2TB
Display(s) 65" LG OLED 120HZ
Case Lian Li dyanmic Evo11 with distro plate
Audio Device(s) Klipsh 7.1 through Sony DH790 EARC.
Power Supply Thermaltake 1350
Software Microsoft Windows 11 x64
Hello ladies and gentlemen.
I hope the day is finding you all well. I just have a question. I’m overclocking my 14900KS on a custom loop. I’ve got it to the point to where it’s just barely hitting 90°C and I’m keeping two of my cores at 63 and the rest at 59. I’m pretty happy with that.

I’d like to stop it from thermal throttling at 90C though. How can I do that?
Do I have to change C states or do I have to manually go into TVB and adjust things independently?
Or can I set TVB to disabled? I just want to completely disable CPU and Ring thermal throttling at 90 C lol.
I’ve got most of the stuff I need set up in bios, but this one thing is making me nutty.

Setting the CPU temp max to 105° didn’t stop it and I never get close to that in testing anyway.

Thanks and happy Friday everyone!
UPDATE:
So I found a step toward the solution? I just used Intel XTU, enabled TVB and disabled the steps and set it to 100° C.
However in Hardware monitor I can see it is still downclocking my CPU..
So uh… Yea… Idk..
Maybe a step in the right direction.
Perhaps TVB manual adjustments in bios is the way to go?
 
Last edited:
When you're not sure (we are not sure) what settings are changed, which are not, clear cmos and start over.

At stock, it should be hitting 100c. You need to change nothing. Enable xmp. Use balanced mode in windows. Do not enable XTU to apply the profile when windows loads, keep the profile save in bios once you have it set up and working well.
 
For 13/14th I think it's easier to gauge what temps should be at X wattage. For example the 13900K I use is all-core 5.5 / 4.3 GHz. It loads at around 320watts in the 90c range for the custom loop I'm using. Less voltage, lower wattage and lower temps is the goal. But you have to fight it not being stable if the voltage is too low.
 
For 13/14th I think it's easier to gauge what temps should be at X wattage. For example the 13900K I use is all-core 5.5 / 4.3 GHz. It loads at around 320watts in the 90c range for the custom loop I'm using. Less voltage, lower wattage and lower temps is the goal. But you have to fight it not being stable if the voltage is too low.
Not sure about stable, but effective clocks will droop with UV. Too Much UV, then yes it will become unstable.
For effective performance increase, power must increase. Especially for those that have adequate cooling.
Effective frequencies must be observed for P and E cores as well as Cache. Typically if v-core is too low and Cache frequency is too high, you may get memory errors, even possibly blue screens.

OP's issue is that TVB sets throttle point at 90c. If there isn't a setting in the TVB settings to raise or lower it, that's because it's a pre-set profile and you get what TVB gives you. Essentially auto-overclocking based on temps and power availability. I think he'd be better off leaving TVB auto/disabled, and doing the tweaking manually.

What we don't have, but kinda need to help is some screen shots of during and after benchmarks so we can see the working effective clocks and voltage. Even if he winds up with lower clocks, we still have plenty of tuning room in the memory department too. He's running G.Skill CL30 (A-Die??) at CL28. (from system specs list under avatar, not sure if it's the same rig) I am going to recommend increasing the frequency to closer to 6800mhz, even if need to run CL34. Then test and tweak further once the frequency is established.
 
I’d like to stop it from thermal throttling at 90C though. How can I do that?
Drop your voltage and clocks, in that order. Because of the recent problems with 13th & 14th gen voltage issues, you should really be aiming to keep your max temps at or below 85C, keeping your voltage low and wattage within spec. You could improve your cooling but that is very involved and can get pricey.
 
Drop your voltage and clocks, in that order. Because of the recent problems with 13th & 14th gen voltage issues, you should really be aiming to keep your max temps at or below 85C, keeping your voltage low and wattage within spec. You could improve your cooling but that is very involved and can get pricey.
The root cause was fixed in the most recent bios revision according to Intel. I only expect to hit those temps during stress testing
Not sure about stable, but effective clocks will droop with UV. Too Much UV, then yes it will become unstable.
For effective performance increase, power must increase. Especially for those that have adequate cooling.
Effective frequencies must be observed for P and E cores as well as Cache. Typically if v-core is too low and Cache frequency is too high, you may get memory errors, even possibly blue screens.

OP's issue is that TVB sets throttle point at 90c. If there isn't a setting in the TVB settings to raise or lower it, that's because it's a pre-set profile and you get what TVB gives you. Essentially auto-overclocking based on temps and power availability. I think he'd be better off leaving TVB auto/disabled, and doing the tweaking manually.

What we don't have, but kinda need to help is some screen shots of during and after benchmarks so we can see the working effective clocks and voltage. Even if he winds up with lower clocks, we still have plenty of tuning room in the memory department too. He's running G.Skill CL30 (A-Die??) at CL28. (from system specs list under avatar, not sure if it's the same rig) I am going to recommend increasing the frequency to closer to 6800mhz, even if need to run CL34. Then test and tweak further once the frequency is established.
Thank you, that is very helpful.
I’ve managed to get a 5.9-6ghz OC on everything. Originally my clocks were dropping down to like 5.5ghz under high stress and that was unexpected. I’d still like to get that all core 6 though lol.
Im messing around with some new memory as well. We will see.
But thank you all for the help.
 
The root cause was fixed in the most recent bios revision according to Intel. I only expect to hit those temps during stress testing

Thank you, that is very helpful.
I’ve managed to get a 5.9-6ghz OC on everything. Originally my clocks were dropping down to like 5.5ghz under high stress and that was unexpected. I’d still like to get that all core 6 though lol.
Im messing around with some new memory as well. We will see.
But thank you all for the help.
Welcome! Always a pleasure!
6Ghz is hit or miss depending on the score. Binning of course. Probably would want to see an SP rating of 112SP or higher. I think this rating is only displayed on Top ROG boards however.
I think the memory you have could be a fast kit, but 64GB is rougher than 32GB on the memory controller. Give a little system Agent voltage a bump, I'd start at 1.35v and work my way up from there.
 
Welcome! Always a pleasure!
6Ghz is hit or miss depending on the score. Binning of course. Probably would want to see an SP rating of 112SP or higher. I think this rating is only displayed on Top ROG boards however.
I think the memory you have could be a fast kit, but 64GB is rougher than 32GB on the memory controller. Give a little system Agent voltage a bump, I'd start at 1.35v and work my way up from there.
I’m actually finding that to be pretty good advice. Bumping up the SA voltage seems very necessary.
Right now, I’ve got my voltages at:
Memory controller 1.55
Dram VDD and VDDQ at 1.55
SA at 1.35

I’m on the Z790 Dark Hero.

I tried 1.5 on the memory voltages but it crashed. Running more tests right now but temps seem sustainable on air.
I’ll lower it once I get stable and see where it ends up. It could be too high but I’m not so sure.

I know most posts say 1.45 on the memory controller but I don’t see anything definitive on it.
And it definitely needed more than 1.5 on the DRAM at 6800 with stock timings for the 2x32 (64gb) Gskill 6000mhz 30-40-40-96 1.4v kit.
Trying not to loosen my timings at all and see where it can get.

Running Aida64, memtest (desktop version) google chrome and signal rgb at the same time just to give it some random variance. So far, so good. Going to think of some other things to throw at it because we all know, no one test proves much of anything. Lol
Maybe I’ll toss some furmark in there too.
I’d love some suggestions for things I could run simultaneously to really try and trip this thing up.
Thanks for the help buddy.
 
Last edited:
I’m actually finding that to be pretty good advice. Bumping up the SA voltage seems very necessary.
Right now, I’ve got my voltages at:
Memory controller 1.55
Dram VDD and VDDQ at 1.55
SA at 1.35

I’m on the Z790 Dark Hero.

I tried 1.5 on the memory voltages but it crashed. Running more tests right now but temps seem sustainable on air.
I’ll lower it once I get stable and see where it ends up. It could be too high but I’m not so sure.

I know most posts say 1.45 on the memory controller but I don’t see anything definitive on it.
And it definitely needed more than 1.5 on the DRAM at 6800 with stock timings for the 2x32 (64gb) Gskill 6000mhz 30-40-40-96 1.4v kit.
Trying not to loosen my timings at all and see where it can get.

Running Aida64, memtest (desktop version) google chrome and signal rgb at the same time just to give it some random variance. So far, so good. Going to think of some other things to throw at it because we all know, no one test proves much of anything. Lol
Maybe I’ll toss some furmark in there too.
I’d love some suggestions for things I could run simultaneously to really try and trip this thing up.
Thanks for the help buddy.
1.55 is not great, but not bad. Just as long as your cooling is decent. I usually don't go past 1.50v myself. But then our memory is very different lol.

So the latency doesn't always get worse with a looser cas latency because you make it up in raw frequency.
you can view that performance here with the new Calculator @W1zzard made us! https://www.techpowerup.com/dram-latency/

So 6000mt/s CL 30 would be 10ns, the same as running 6800mt/s at CL 34 is 10ns. The difference is you are gaining bandwidth and that's where DDR5 really shines at.
Looser CL also means you don't have to beat the shit out if the Vdimm.

So you don't need to benchmark (test) anything, you can see the latency plain as day.
Performance wise in programs may be a different story, you would make gains raising tREFI for example, other tweaks as well.

EDIT IN:
Programs to test with-
Y-cruncher 1b and 2.5b - great for stability, tweak sensitive
PiFast - this one loves low latency!
PiMod 1m 32m - Fast and stable, great for measuring memory metrics
7-Zip, not my personal favorite, responds well to memory tweaking.
And of course any 3D benchmark would work, but differences will be minimal. These are mostly GPU intensive benchmarks.
 
Last edited:
1.55 is not great, but not bad. Just as long as your cooling is decent. I usually don't go past 1.50v myself. But then our memory is very different lol.

So the latency doesn't always get worse with a looser cas latency because you make it up in raw frequency.
you can view that performance here with the new Calculator @W1zzard made us! https://www.techpowerup.com/dram-latency/

So 6000mt/s CL 30 would be 10ns, the same as running 6800mt/s at CL 34 is 10ns. The difference is you are gaining bandwidth and that's where DDR5 really shines at.
Looser CL also means you don't have to beat the shit out if the Vdimm.

So you don't need to benchmark (test) anything, you can see the latency plain as day.
Performance wise in programs may be a different story, you would make gains raising tREFI for example, other tweaks as well.

EDIT IN:
Programs to test with-
Y-cruncher 1b and 2.5b - great for stability, tweak sensitive
PiFast - this one loves low latency!
PiMod 1m 32m - Fast and stable, great for measuring memory metrics
7-Zip, not my personal favorite, responds well to memory tweaking.
And of course any 3D benchmark would work, but differences will be minimal. These are mostly GPU intensive benchmarks.
lol thank you. You are very right that I am in fact beating the shit out of my VDIM

Thats a really cool tool wizard made, I’ll definitely be sure to check that out.
Thank you for the suggestions on additional testing tools and for the clarification on latency and the effects of frequency vs timings.
That’s incredibly helpful.
Much appreciated.
 
1.55 is not great, but not bad. Just as long as your cooling is decent. I usually don't go past 1.50v myself. But then our memory is very different lol.
I'm surprised it hasn't burnt out yet. 155v is crazy for the IMC and unnecessary. 6800 MT/s Dual-rank may be close to the upper limit, I didn't dare try above 1.45v to get 7000 MT/s stable.

For some reason EVGA likes to allow super high voltages and it won't turn "red" warning until 1.6. everyone else does it at 1.4-1.45v.

I would lower it soon or you'll have a dead IMC sooner than later.
 
I'm surprised it hasn't burnt out yet. 155v is crazy for the IMC and unnecessary. 6800 MT/s Dual-rank may be close to the upper limit, I didn't dare try above 1.45v to get 7000 MT/s stable.

For some reason EVGA likes to allow super high voltages and it won't turn "red" warning until 1.6. everyone else does it at 1.4-1.45v.

I would lower it soon or you'll have a dead IMC sooner than later.
For Vdimm have to unlock "high vdimm" allowed to go past 1.45v, but it doesn't go yellow till 1.55v.
System Agent is also in the yellow on Asus boards at 1.55v. I think it goes yellow, magenta, then red.
Yeah, I'd try stay 1.50v or lower IMC voltage. usually starts getting hot there any ways.
Not sure about smoking the controller though, but at the peak before fast/er degradation on ambient cooling.
 
Memory controller 1.55
you'd be toasting the chip in just a week of use..and just for 6000MT's?

6000MT's can be done with just 1.2v IMC Voltage (or even less), even Auto settings from Asus should work out of the box for VDD (dimm) 6000MT's CL30 should be doable around 1.3v
 
Not sure about smoking the controller though, but at the peak before fast/er degradation on ambient cooling.
The LN2 guys or those who aren't planning on using the CPU for years don't have to worry :). Really I'm just speculating based on previous generations.

Anyone want to donate a CPU to science. Run at 1.6v 24/7 and let us know when it dies

I also think there is a diminishing return once you pass 1.45v.
 
I also think there is a diminishing return once you pass 1.45v.
To a point I agree. You at least need to consider a major cooling upgrade at that point, and not those shitty ram cooler things either.....lol Thats VDIMM. As for IMC volts, unless you are looking to hit 9000+, there is absolutely no need for that level of power driven to it.
 
The LN2 guys or those who aren't planning on using the CPU for years don't have to worry :). Really I'm just speculating based on previous generations.

Anyone want to donate a CPU to science. Run at 1.6v 24/7 and let us know when it dies

I also think there is a diminishing return once you pass 1.45v.
1.60v for....what? ... v-core? I thought OP and I was talking about Vdimm and SA here. LN2 guys?

1.55v IMC is high, but I was focusing on System Agent, not IMC. I think the conversation got cross-fired.
Also, myself, repeating said 1.55v is probably top and would start a fast/er degrade for IMC?

Am I missing something or you poking a stick at me for a reason?
 
The SA at 1.35 is the least of his voltage issues imho, high, but dude needs a different kit to do what he wants, or run xmp and be happy with much lower volts...imc/CPU heat.

There could be a happy medium in there at 6400, but ya, volts are a tad crazy for 6800c30.
 
1.60v for....what? ... v-core? I thought OP and I was talking about Vdimm and SA here. LN2 guys?

1.55v IMC is high, but I was focusing on System Agent, not IMC. I think the conversation got cross-fired.
Also, myself, repeating said 1.55v is probably top and would start a fast/er degrade for IMC?

Am I missing something or you poking a stick at me for a reason?

SA voltage on RPL is the IMC voltage. -- what voltage is the "IMC voltage" you're referring to? IMC is part of the SA along with the uncore and pcie controllers etc.
 
SA voltage on RPL is the IMC voltage. -- what voltage is the "IMC voltage" you're referring to?
Ask OP brother. I'll quote him for you here.
Dram VDD/Q 1.55v
System Agent 1.35v

HWinfo must have gave him a seperate IMC reading? 1.55v

Memory controller 1.55
Dram VDD and VDDQ at 1.55
SA at 1.35
 
Ask OP brother. I'll quote him for you here.
Dram VDD/Q 1.55v
System Agent 1.35v

HWinfo must have gave him a seperate IMC reading? 1.55v
Ohhh yeah sometimes it takes one of the VCCIO voltages as "imc" but no - it's the 1.35v sa. Mine doesn't even have that category:
1730312436662.png
 
Ohhh yeah sometimes it takes one of the VDDIO voltages as "imc" but no - it's the 1.35v sa.
That's because in the past the VCCIO is the IMC and the VCCSA is SA and on his board, is likely to have separated values and inputs (Z790 Dark Hero)
 
There are seprated SA and Memory Controller volts on my Z790 APEX Encore. Labeled as such.
 
There are seprated SA and Memory Controller volts on my Z790 APEX Encore. Labeled as such.
When I am using 14100F (right now actually) a lot of settings disappear lol.
System agent remains, can change it all I want, stays put 0.960v
Even settings like locking the BCLK disappear.
It's weird, like turns the board into a base model OEM or something.

So I'm running 6000mt/s CL34 1.40v-dimm, System Agent 0.960v and no readout for IMC which is probably at 1.0v (0.960v) but get to guess about it. I'm also not running 64gb 2x32's either. I'd wager be at a lower frequency. 3 days now.

Screenshot 2024-10-30 133556.png
 
That's because in the past the VCCIO is the IMC and the VCCSA is SA and on his board, is likely to have separated values and inputs (Z790 Dark Hero)

My Encore does that as well - it must be an Asus thing.

Also shows MC voltage on the summary screen.

MCV.png
 
I’ll probably just end up settling on like 6600 or 6400 or something. The voltages are definitely too high, but I was curious to see what it could do. I won’t let it stay like that for a daily though.

PS: Hey uhh, lemme get that encore real quick. I PROMISE I’ll give it back…
 
Back
Top