[725241]

I updated my gigabyte bios to latest and when i hit cb23 it rams 300 watts into my 13900k. Before it was 42200 score with 275 watts or
So. Now it scores 41100 at same cpu speeds at 240 watts but thats not stable cause ycrncher 2.5 locks up pretty much right away. I tried adding + offset but seems to be locked at same volts 1.15. Man i should have skipped these dumb bios updates. Old bios worked normal. At least memory overclock is still same stable setup so far.
I bet that cep thingy is enabled. Looks like
Clock stretching going on forsure. Just thought about that now.

 

[yzonker]

Only Intel has any chance of knowing what the true voltage limit is. And nobody should take the 1.55v limit as the limit because Intel is in damage control mode, so they are toeing the line between nerfing performance and reducing degradation.

Probably hoping the changes made will stem the tide of degraded CPUs enough to fall out of the media spotlight. Let the warranty cover the rest.

I'm sure Intel is badly wanting to release Arrow Lake to also help take attention away from Raptor Lake.

I'm hoping AL is more successful as I'm already getting excited about the possibility of at or near 10k MT/s DDR5. I think AMD has missed the boat for CUDIMM which makes that platform less appealing for sure (if true).

 

[CarSalesman]

No one knows. Whatever voltage won't shrink the CPU at whatever temp you're at. Under 1.20v load is obviously ideal.

Oh, the constitution! 🤣

 

[fray_bentos]

Same as BIOS voltage, it is gaming-only 8P+8E setup so it ran 6.0 1.42 when in game and then downclock to 800MHz and downvolt when idling. I had TVB setup so it drops to 59 at 70C and 55 at 80C but it would basically never happen since it cannot really go above 250W with only 16 threads, and only occasional sharder compilation could cause that kind of power. It wasn't hammered with heavy loads at all, it was downclocking and downvolting properly at idle and 1.42 is not some kind of an insane voltage to begin with, unless maybe it is run with unlocked power with 32 threads and it draws 450W, but I don't do that. Based on my previous experiences and compared to all my previous CPUs this one was ran the least aggressively both in terms of settings and loads and yet it already behaves as if it was 5-6 years old barely 2 years in.

Except 1.42 V is now established as an insane voltage for RPL.

 

[Krzych04650]

1.42v load with all cores, even with gaming loads, is far too high for 10nm.
This is already questionable on 14nm (people didn't like >1.35v load on 14nm), but you're doing it on 10nm, on a stepping that is even less resilient (13900k will degrade more at the same voltage than 14900k since the silicon is not as refined), so it's no surprise it degraded.

I already tested 1.51v bios set with LLC3 (1.1 mohm), giving 1.46v pure idle, with no load at all for a week on a 13900k, and that caused degradation (vmin increased drastically in R23), and you were close to that at your loads.

Even something more reasonable like 1.42v idle (bios set) with LLC6 or 5 and down to 1.30v load is already asking for trouble on a 13900k, and a 14900k won't like it either.

Well if that is the case then there is no wonder why all this degradatation drama exists given how many of those chips come with 1.4+ stock VF point for that dual core boost. And it is still relatively early with only 13th gen hitting 2 years old, it will probably only get worse.

I thought it should be reasonable to assume that if Intel itself is using those kind of voltages stock then those CPUs should be able to take it, especially for light loads, but I guess not

Will see what ARL does and then decide what to do.

Except 1.42 V is now established as an insane voltage for RPL.

Now is slightly too late. Go back in time and tell that to Intel from 2-3 years ago.

 

[fray_bentos]

Well if that is the case then there is no wonder why all this degradatation drama exists given how many of those chips come with 1.4+ stock VF point for that dual core boost. And it is still relatively early with only 13th gen hitting 2 years old, it will probably only get worse.

I thought it should be reasonable to assume that if Intel itself is using those kind of voltages stock then those CPUs should be able to take it, especially for light loads, but I guess not

Will see what ARL does and then decide what to do.



Now is slightly too late. Go back in time and tell that to Intel from 2-3 years ago.

Not really "too late". I made a very conservative decision almost 2 years ago not to exceed 1.28V under load based on early degradation reports on this very site (at voltages as "low" as 1.35 V). "Stock" voltages for these CPUs were clearly massively excessive given the undervolt headroom (and overclock headroom on 13600K). When I learned I could run 5.4 GHz clocks at 120 mV less load voltage than 5.7 GHz, it was an easy decision to settle on 5.4 GHz @ 1.16 V heavy load (vs. 5.7 GHz @ 1.28 V). The result is that my Vcore never crosses 1.22 V even in fleeting low loads. I have no worries about degradation.

 

[CarSalesman]

Not really "too late". I decided almost 2 years ago not to exceed 1.28V under load based on early degradation reports on this very site. "Stock" voltages for these CPUs were clearly massively excessive given the undervolt headroom (and overclock headroom on 13600K). When I learned I could run 5.4 GHz clocks at 60 mV less load voltage than 5.7 GHz, I decided to settle on 5.4 GHz @ 1.16 V heavy load. The result is that my Vcore never crosses 1.22 V even in fleeting low loads. I have no worries about degradation.

1.28V load voltage is already a metric f*ckton of voltage. No board would be able to continually supply that load voltage because of temp limits being reached on the CPU unless you are custom loop with a chiller or direct die or both, anyways.

I've kind of done the same thing with my binned KS. I disabled all idle and power saving features. 1.28V idle; 1.15 load voltage. I'll still bench aggressively on cold mornings but I don't want to daily crazy profiles anymore.

 

[Betroz]

Except 1.42 V is now established as an insane voltage for RPL.

As I wrote in another post, if that is true then Intel is completely morons for thinking 1.55v is fine for 2 core boost. It is not only with 2 core boost a CPU can ask for over 1.4v either. My 14900KF has VID requests for over 1.4v even for stock 5.7 Ghz when I use stock LLC (Level 3 on Asus). Even at LLC4 it does that. I have to play with the VF curve to change it, which should not be necessary for stock.

Not really "too late". I made a very conservative decision almost 2 years ago not to exceed 1.28V under load based on early degradation reports on this very site (at voltages as "low" as 1.35 V).

One explanation is that the CPUs that degraded at <1.35v were of a faulty batch, maybe even among those with that oxidation problem. Who knows. My point here is that there are forum members here who run their CPUs at higher voltages that have not degraded. All we can do is speculate right.

 

[fray_bentos]

1.28V load voltage is already a metric f*ckton of voltage. No board would be able to continually supply that load voltage because of temp limits being reached on the CPU unless you are custom loop with a chiller or direct die or both, anyways.

I've kind of done the same thing with my binned KS. I disabled all idle and power saving features. 1.28V idle; 1.15 load voltage. I'll still bench aggressively on cold mornings but I don't want to daily crazy profiles anymore.

I was talking about 13600K at 1.28 V not an i7 or i9 where indeed, much more current would flow at such a voltage.

 

[SSBrain]

As I wrote in another post, if that is true then Intel is completely morons for thinking 1.55v is fine for 2 core boost. It is not only with 2 core boost a CPU can ask for over 1.4v either. My 14900KF has VID requests for over 1.4v even for stock 5.7 Ghz when I use stock LLC (Level 3 on Asus). Even at LLC4 it does that. I have to play with the VF curve to change it, which should not be necessary for stock.

In theory the system integrator / builder / etc. would adjust the AC loadline so that load voltage wouldn't exceed that defined in the CPU's VF curve (which should be easy to retrieve on ASUS boards, although I don't have direct experience with them).

But indeed it's difficult to believe that just idling (under what conditions/BIOS settings, though?) around 1.4V would cause the CPU to degrade. The CPUs should be capable of indefinitely sustaining prolonged workloads with 1 to a few cores as long as temperatures are under control. Surely having the CPU doing useful work would be worse than idling? (although if C-states are enabled and the current limit is in place, potentially damaging transient current spikes would never happen).

 

[viki_zaiko]

hmm, even at lite load mode 1 with LLC 7 +2pcore +1ecores is stable, I guess I will be using that
It's best to keep lite load 1 with the biggest number of LLCs that you are stable.
Maximum voltage is 1.28v after test like cinebench15 extreme, gaming for 2 hours 3 different games

IA Domain Loadline (AC/DC): 0.010 / 0.010 mOhm

 

[Betroz]

In theory the system integrator / builder / etc. would adjust the AC loadline so that load voltage wouldn't exceed that defined in the CPU's VF curve

Lowering ACLL is not enough with a droopy LLC to keep VID requests low (like stock LLC3 on Asus). Then again having a droopy LLC is benefical for getting high 2 core boosts.

 

[SSBrain]

Try checking out if decreasing Ring frequency eliminates VID request peaks under those circumstances. On my 12th gen Intel CPU I find it's the main reason why on few-cores workloads load voltages appear to be unreasonably high. At lower Ring frequencies the voltage behavior/AC LL operate more or less as expected.

Here are two HWInfo screenshots of my CPU operating with the same 2-core workload with Prime95 and Ring at the default dynamically-changing maximum (4.6 GHz), then manually configured to the all-core maximum (3.6 GHz—on 12th gen CPUs it used to get decreased considerably when E-cores were engaged), both with the droopiest LLC and a moderately high AC LL bringing voltages to (supposedly) the processor's VF curve levels under load.

Ring=Dynamic/4.6 GHz ⇒ Vcore reaches 1.39V

Ring=Fixed/3.6 GHz ⇒ Vcore tops at about 1.315V

 

[fray_bentos]

In theory the system integrator / builder / etc. would adjust the AC loadline so that load voltage wouldn't exceed that defined in the CPU's VF curve (which should be easy to retrieve on ASUS boards, although I don't have direct experience with them).

But indeed it's difficult to believe that just idling (under what conditions/BIOS settings, though?) around 1.4V would cause the CPU to degrade. The CPUs should be capable of indefinitely sustaining prolonged workloads with 1 to a few cores as long as temperatures are under control. Surely having the CPU doing useful work would be worse than idling? (although if C-states are enabled and the current limit is in place, potentially damaging transient current spikes would never happen).

It's not about "what is difficult to believe"; @Falkentyne already did the experiment to show idle voltages around 1.4 V leads to degradation. Intel themselves recommend ensuring c states and adaptive voltage in their new safe recommendations.

 

[Betroz]

It's not about "what is difficult to believe"; @Falkentyne already did the experiment to show idle voltages around 1.4 V leads to degradation. Intel themselves recommend ensuring c states and adaptive voltage in their new safe recommendations.

He had 1.5 idle voltages, not 1.4.

Take a look at my 14900KF VF curve. If I were to follow your guidance to safe voltage, then 5.1 pcores would be the limit (assuming no undervolting/tweaking).
Old screenshot and not my current settings btw.

 

[SSBrain]

It's not about "what is difficult to believe"; Falkentyne already did the experiment to show idle voltages around 1.4 V leads to degradation.

Am I not allowed to be at very least skeptical of the circumstances that led to that conclusion? Unlike single-core boost over 1.5V (which one might argue could be slowly causing cumulatively damaging current/voltage spikes over time), 1.4V would be easily reached and sustained under typical few-cores workloads. Are we to believe that Intel never saw evidence of degradation with that in their internal testing before releasing their CPUs?

Intel themselves recommend ensuring c states and adaptive voltage in their new safe recommendations.

Intel always recommended enabling all C-states for "long-term reliability" in their public datasheets.

And operation outside the processor's VF curve has also always been considered out-of-specification, which implies that fixed-voltage operation would in turn be as well.

Besides, performance limiting (power, thermal, etc) doesn't work correctly with fixed voltages, and you don't get things like TVB Voltage Optimizations either.

 

[Betroz]

Are we to believe that Intel never saw evidence of degradation with that in their internal testing before releasing their CPUs?

Not only that, but apparently they screwed up the VF curve completely aswell, or so the story goes.

 

[yzonker]

Am I not allowed to be at very least skeptical of the circumstances that led to that conclusion? Unlike single-core boost over 1.5V (which one might argue could be slowly causing cumulatively damaging current/voltage spikes over time), 1.4V would be easily reached and sustained under typical few-cores workloads. Are we to believe that Intel never saw evidence of degradation with that in their internal testing before releasing their CPUs?

You're assuming Intel's manufacturing process remained stable during production. That might or might not be a good assumption, although they already admitted to one issue with the 13th gen oxidation. If it didn't, that would explain why they missed this so badly. I'm still of the opinion that there is a wide range of durability with these CPUs. It's the only thing I can think of that would explain such a wide range of results we've seen.

I ran my 13900k at voltages almost identical to what @Krzych04650 was running for a few months. Zero degradation. Mostly gaming, same as now with my 14900k. But somehow my 14900k degraded at less voltage. Power levels seen by both CPUs was similar too (mostly low load from gaming, a small amount of 300w+). I literally have some evidence first hand, albeit anecdotal, that what I'm saying is correct.

 

[6u4rdi4n]

Oh well. At least getting a replacement doesn't seem too troublesome. A couple of emails back and forth, old CPU picked up early afternoon on Thursday, new CPU shipped out on Friday and is estimated to arrive on Monday due to the weekend.

 

[SSBrain]

@yzonker
I think it's plausible that some batches might have had unexpected manufacturing issues that led to premature degradation at otherwise normal voltages, or even that Intel intentionally pushed single-core boost clocks/voltages to levels where under some circumstances degradation might have occurred faster than they expected.

I'm only calling into question the general idea that idling (?) at ~1.4V would make the CPU degrade in a week or so and that this voltage level should be declared as unsafe regardless of the circumstances. If this is always happening, then the maximum safe voltage to make the CPU last years would have to be 1.25V or even less—regardless of the load—and frequencies and current limits be downgraded to 12th gen levels.

I'm more inclined to think that that CPU in particular had manufacturing issues and that probably the testing conditions didn't help, if that was under Windows with C-states off and no limits (something that people keep ignoring despite Intel's recommendations).