Intel Ships 0x129 Microcode Update for 13th and 14th Generation Processors with Stability Issues

[vMax65]

This is the thing I don't get, I mean people concerned about VID being high. Isn't VID exactly what you say, so voltage CPU asks for, but one it gets being Vcore, so Vcore should be the one to concern? Btw what exact voltages are the ones to be concerned in the matter of this whole instability drama?

For me it has always been Vcore. You can also look at VROUT (which is probably a little more accurate). VID when everything is on auto is what is being requested but gets modified by a few parameters including LLC, Power states like EIST/C1E temps etc which when looking at vcore is the actual voltage being used. As soon as you start changing settings like fixed, adaptive, offsets etc, then VID becomes less relavent. For both Intel and AMD VID is just a value that the CPU pulls from a programmed table of Multiplier/Voltage, and the mobo uses this to determine what Auto voltage it should set.

Can be real confusing and I am not any sort of expert so if there are any CPU engineers on TechPowerUp please jump in to help...

 

[chrcoluk]

For me it has always been Vcore. You can also look at VROUT (which is probably a little more accurate). VID when everything is on auto is what is being requested but gets modified by a few parameters including LLC, Power states like EIST/C1E temps etc which when looking at vcore is the actual voltage being used. As soon as you start changing settings like fixed, adaptive, offsets etc, then VID becomes less relavent. For both Intel and AMD VID is just a value that the CPU pulls from a programmed table of Multiplier/Voltage, and the mobo uses this to determine what Auto voltage it should set.

Can be real confusing and I am not any sort of expert so if there are any CPU engineers on TechPowerUp please jump in to help...

My understanding is the same as you, the reason I showed VID in the screenshot was because Jay used VID in his video so was kind of apples to apples, and I knew that I was undervolting so was showing what my CPU is requesting, I did later post my actual vcore values though.

I can use vcore moving forward on TPU, not a problem for me, as I agree with what you said.

 

[vMax65]

This is the thing I don't get, I mean people concerned about VID being high. Isn't VID exactly what you say, so voltage CPU asks for, but one it gets being Vcore, so Vcore should be the one to concern? Btw what exact voltages are the ones to be concerned in the matter of this whole instability drama?

In terms of what to be concerned about, any voltages over 1.5v (okay even 1.45v for me would be too much! but some don't mind and like pushing to see what max they can get) would be a concern and I believe the issue was that up to 1.6v was being pushed under certain circumstances. For me limiting to 1.350v was a good middle ground (no loss in performance for me) and fingers crossed no issues and I monitor via HardwareInfo to make sure my voltage spikes do not go above 1.4v and since I have gone for an override/fixed voltage, this has been the case.

 

[ShrimpBrime]

In terms of what to be concerned about, any voltages over 1.5v (okay even 1.45v for me would be too much! but some don't mind and like pushing to see what max they can get) would be a concern and I believe the issue was that up to 1.6v was being pushed under certain circumstances. For me limiting to 1.350v was a good middle ground (no loss in performance for me) and fingers crossed no issues and I monitor via HardwareInfo to make sure my voltage spikes do not go above 1.4v and since I have gone for an override/fixed voltage, this has been the case.

Not an engineer, but you worry too much seeing a v-core reading.

The problem is you state certain loads. But never say which one.

Oh no 1 core utilized 1.5v to obtain a frequency of 5.6ghz through 6.0ghz depending on the cpu model. This does not mean 250w man.

The issue is from heavy loads like all cores and the wattage. So Volt x Amp = wattage.

1.35v limited while under all core load drooping to 1.288v. That's a big droop. Why not just set 1.288v with a higher LLC and keep a sustained v-core? You'd still be at nearly 250w peak consumption....

People really should look at the amps. Lower that. V-core isn't a big deal. Well it is to those that don't understand it I suppose.

OK - Sorry for double post.

This is TWO cores at 1.682v Cpu-z reading with High LLC so keeping fairly stead 1.68v observed during the benchmark as minimum.

This is only 60w...... Edit: (Only 1 core utilized for the benchmark.)

Degradation?

 

[Dr. Dro]

Yeah, it's possible. I think they pushed the 14 a few megahertz higher.

Are you monitoring your CPU vids in HWInfo during a few different games to see what the vids are doing? And did you do this before the flash to be able comparison of behaviour?

No need for further speculation - Buildzoid hooked his 14900K/Master X on the oscilloscope

Looks like there's a whole mess with the settings going on - and looks like my setup is affected too. Video showed me how to improve on the situation. Might have changes coming with newer BIOSes

 

[chrcoluk]

No need for further speculation - Buildzoid hooked his 14900K/Master X on the oscilloscope

Looks like there's a whole mess with the settings going on - and looks like my setup is affected too. Video showed me how to improve on the situation. Might have changes coming with newer BIOSes

Buildzoid seems to have got an improvement out of the updates, his CPU at the start of all these events couldnt even run cinebench now he is running it undervolted.

In the end the mess I think was just his cinebench being detected as a background app and limited to e-cores only by windows I think, he said it at the end.

 

[ShrimpBrime]

Buildzoid seems to have got an improvement out of the updates, his CPU at the start of all these events couldnt even run cinebench now he is running it undervolted.

In the end the mess I think was just his cinebench being detected as a background app and limited to e-cores only by windows I think, he said it at the end.

Have the same problem with Z790 Gaming wifi. Can't pass a benchmark to show a screenshot.

Turns out, cpu runs fine on my B660-G board with first release ME firmware. Which is sad cause I'd like to get back to overclocking.

The problem Intel has is that once ME updates, you can't roll it back. They fucked themselves with that.

 

[starfals]

Finally, intel is forced to go back to earth lol. THey have been pushing temps, speeds and voltages too hard to compete and win lol.

 

[phanbuey]

I just tested my unpatched 13700kf and max volts at stock with LLC offset was 1.317v at 5.4Ghz -- i think ill sit this one out since I like the bios version I'm on for ram subtimings.

Chip has been running great since release and at this point, and

 

[Minus Infinity]

No need for further speculation - Buildzoid hooked his 14900K/Master X on the oscilloscope

Looks like there's a whole mess with the settings going on - and looks like my setup is affected too. Video showed me how to improve on the situation. Might have changes coming with newer BIOSes

I was just wondering what builzoid had to say on the matter. He's the go to guy IMO.

I've not really owned an Intel cpu since 3570K days, so not really been into this stuff much, but I do watch buildzoid for MB/memory recommendations and have followed his videos since these issues first started popping up.

 

[iameatingjam]

I just tested my unpatched 13700kf and max volts at stock with LLC offset was 1.317v at 5.4Ghz -- i think ill sit this one out since I like the bios version I'm on for ram subtimings.

Chip has been running great since release and at this point, and
View attachment 358553

I am using stock clocks, not touched them, so 5.4ghz on the two best cores and 5.3 the rest, as I said I think I have a good chip.

The VIDs I pasted here are also using Intel spec'd AC/DC.

Here is pic of my configured undervolt, although actual undervolt seems a bit lower.

I swear I backed these off to 50mv for daily . Guess I didnt.

View attachment 358453

View attachment 358454

Your test is not surprising as I think the microcode is not making any adjustment with your current voltages.

Yeah I think you do have a good sample.

So I took the shackles off my cpu ( haven't decided if I'm going to keep it that way)

I'm getting a maximum of 1.43 vcore (as per hwinfo) after running the passmark cpu test (includes multi and single core). And... I even turned off turbo 3 off so there'd be no 5.6 cores. Put ac at 90 instead of the 110 it was at originally. And left the 253 limit on, so otherwise still 'intel defaults= high performance' which includes 307 iccmax so its not 100% unbound.

Really goes to show how the 14700k is just an overclocked 13700k, its not better binned or, at least in my case. Guess that was just me fooling myself when I made the purchase.

On one hand its like should I just go back down to 5ghz and turn off some ecores? Turn down the power limits.... Pretend its like a 13600k/13700k and hope it lives a long life? Its not like that extra 10% of performance really matters for daily tasks. On the other its like, because I damn well paid for it whats why, and rma has been extended so if something goes wrong, maybe I'll get a better sample?


I just don't know ='(


Can always go back to 104 and do offsets but that corrupted my os last time I tried so I don't think I have much wiggle room (even though I did in January..... not a comforting thought) then again thats also when I found a couple pieces of something in my socket so that could have been what was causing issues. Also don't want to lose these new ucode protections...

On the bright side, I can confirm the vid limit (1.4) in my bios works as before turning it off all my pcores could not hit 5.5 unless they were basically idle. (Actually for some reason, vid was showing not going higher than 1.37... but thats software reporting for you I guess)

Perhaps setting a vid limit of 1.5 would be enough to allow me to hit 5.5 consistently enough while also offering a bit more protection than what intel is offering. Does intels vid limit vary by sku or is it just 1.55 across the board? I would really like to know.

Too much choices, giving me a headache. Anyway sorry for venting.

 

[chrcoluk]

You could try emulating a 13700k and see what voltage you end up with either a all core 5.3/5.4 or 5.4 on 2 best cores and 5.3 for rest, that has me curious how well the voltages are in that configuration. If you curious my e-cores cap at 4.2ghz, I assume they got boosted on 14th gen as well. Also dont know how open you are to undervolting, the risk is instability if your chip cant handle it but there should be no risk to the health of the chip as you lowering voltage load on it.

I do have my power limits conservative and even lowered my tjmax from 100C to 95C might drop it later to 90C. My power limits are 175w for short and 125w for long. I ran a quick and dirty CPUZ bench yesterday which hit the 175w limit, but still outscored the 12900k reference on it.

I just tested my unpatched 13700kf and max volts at stock with LLC offset was 1.317v at 5.4Ghz -- i think ill sit this one out since I like the bios version I'm on for ram subtimings.

Chip has been running great since release and at this point, and
View attachment 358553

Yeah similar situation for me.

 

[iameatingjam]

You could try emulating a 13700k and see what voltage you end up with either a all core 5.3/5.4 or 5.4 on 2 best cores and 5.3 for rest, that has me curious how well the voltages are in that configuration. If you curious my e-cores cap at 4.2ghz, I assume they got boosted on 14th gen as well. Also dont know how open you are to undervolting, the risk is instability if your chip cant handle it but there should be no risk to the health of the chip as you lowering voltage load on it.

I do have my power limits conservative and even lowered my tjmax from 100C to 95C might drop it later to 90C. My power limits are 175w for short and 125w for long. I ran a quick and dirty CPUZ bench yesterday which hit the 175w limit, but still outscored the 12900k reference on it.

Yeah I was kinda curious about that too. Only thing is... I'm not sure how I can turn turbo 3 back on, but make those two core boosts go to 54 instead of 56 in the bios. Maybe throttlestop could help me with that. Anyway, I'll look into it. Would be very interesting if the results turned out to be the same.

So, if you haven't updated, does that mean your ac/dc loadline is at 110? (since the default just changed for me, not sure if thats a gigabyte or intel thing) Should be in hwinfo>summary>Motherboard under 'ai ac loadline' and ''ai ad loadline.' As of now my other loadlines are all on auto. Just to get it as close as possible.

 

[chrcoluk]

Ahh I see you have a B760, so you probably cant do it, is possible on Z chipsets.

How did you get it down to 5ghz then?

 

[iameatingjam]

Ahh I see you have a B760, so you probably cant do it, is possible on Z chipsets.

How did you get it down to 5ghz then?

I can still access the multiplier. It just wont take if I put it higher than the default boosts. So underclocking the all cores is no problem. Just not sure how to do it with the two 5.4 cores, separate from the rest of the 53x cores is what I mean. I'm sure I can figure it out though. Just... right now I need some game time. Been reading text and looking at numbers for too long. I'll get back to you though. I am interested in the result.

 

[Lesha]

Does anyone else think that "incorrect voltage requests to the processor that are causing elevated operating voltage" is just an excuse for pushing the voltage to be able to compete with AMD?

 

[Caring1]

ASUS' BIOS release for Z790 Creator is broken in some way

with Intel profile I get around 13k score in R23 (13700K) and it should've been around 29-30k

No, that's how it is supposed to work, Asus made the error by giving it "all the power" and people assumed that was the norm.

 

[Dr. Dro]

No, that's how it is supposed to work, Asus made the error by giving it "all the power" and people assumed that was the norm.

13k in R23 is not normal for a 13700K. It's less than half what it should be.

 

[chrcoluk]

I can still access the multiplier. It just wont take if I put it higher than the default boosts. So underclocking the all cores is no problem. Just not sure how to do it with the two 5.4 cores, separate from the rest of the 53x cores is what I mean. I'm sure I can figure it out though. Just... right now I need some game time. Been reading text and looking at numbers for too long. I'll get back to you though. I am interested in the result.

Ok then try all core 5.3.

I had a look at my notes for AC/DC, so basically I equalised them as intel now recommends, 1.10/1.10 is set in bios, and this bumped my voltages up a bit so thats why my undervolt increased to -75mv. I think it might be ok to reduce them, but they both should be same value.

 

[zmeul]

No, that's how it is supposed to work, Asus made the error by giving it "all the power" and people assumed that was the norm.

highly doubtful as other reviewers with MSI, GB mobo have 29-30k score

 

[phanbuey]

highly doubtful as other reviewers with MSI, GB mobo have 29-30k score

This happens when you have voltage protection enabled and undervolt too far. For instance, if I run my chip at 1.3v flat, it's actually stable but I get like 12K cinebench score, and 8K MT CPU-z score instead of the 27K (@190W) and 12.6k i usually get.

That's been the case since basically day 1 bios -- they released a bios later where you could turn off the undervolt protection, and that let the chip crash but rip at full performance with an undervolt. I think the undervolt protection just underclocks the chip if the voltage fed is below the vid table and it's just a flat 50% underclock.

 

[iameatingjam]

Ok then try all core 5.3.

I had a look at my notes for AC/DC, so basically I equalised them as intel now recommends, 1.10/1.10 is set in bios, and this bumped my voltages up a bit so thats why my undervolt increased to -75mv. I think it might be ok to reduce them, but they both should be same value.

Alright, that sounds close enough. So I moved ac/dc up to 110. Reduced ecores to 8. pcores 53x and ecores 42x. pl1 125 pl2 175 (is that right? I like to keep them the same personally). The only difference is I can't apply that offset (not without doing things I don't want to do anyway)... And I wasn't sure what you usually do for load testing but I saw you mentioned cpu-z so I tried that and this is what I got: (should I have pressed stress cpu instead? I'm not really that familiar with this one).

Max vcore of 1.35. See that seems more normal to me. If a -75mv offset were stable on here ( don't know, don't want to try, since it involves ucode 104 which I've had bad luck with - twice! That thing is cursed). That would put us roughly around the same, right? If I were to jump to conclusions, it would almost seem like 13700k and 14700k aren't even binned that differently. I guess I assumed that was the whole point of the new gen. No new silicon, but move the bins up a tier.

Anyway yeah thats why I do have to use the ac/dc loadline as other than frequency its pretty much my only tool to affect vcore while staying up to date. But I can only get so far with it. 90 seems to have the same performance as 110 at least as far as I can see. Somebody on reddit suggested 60 but my performance nosedived. I guess some people increase llc to make up for this but w/e. Intel's whole system is just too complicated as far as I'm concerned. I'm just gonna run intel defaults with ac/dc at 90 and turbo 3 off, vid limit of 1.5 and try to forget about it.
.
.
.
.
.
.
.
Off topic but last time I ran this test on my 14700k, the score was 14700, I'm not even kidding... I think I still have the pic.....

lol!

 

[Moofachuka]

This needs a re review... 13 and 14 gen

 

[Caring1]

highly doubtful as other reviewers with MSI, GB mobo have 29-30k score

Yes, they all allowed too much power to be given to the CPU until now.
There are plenty of videos online as proof they exceeded the 253W Intel specification.

 

[chrcoluk]

Alright, that sounds close enough. So I moved ac/dc up to 110. Reduced ecores to 8. pcores 53x and ecores 42x. pl1 125 pl2 175 (is that right? I like to keep them the same personally). The only difference is I can't apply that offset (not without doing things I don't want to do anyway)... And I wasn't sure what you usually do for load testing but I saw you mentioned cpu-z so I tried that and this is what I got: (should I have pressed stress cpu instead? I'm not really that familiar with this one).


View attachment 358601

Max vcore of 1.35. See that seems more normal to me. If a -75mv offset were stable on here ( don't know, don't want to try, since it involves ucode 104 which I've had bad luck with - twice! That thing is cursed). That would put us roughly around the same, right? If I were to jump to conclusions, it would almost seem like 13700k and 14700k aren't even binned that differently. I guess I assumed that was the whole point of the new gen. No new silicon, but move the bins up a tier.

Anyway yeah thats why I do have to use the ac/dc loadline as other than frequency its pretty much my only tool to affect vcore and stay up to date. But I can only get so far with it. 90 seems to have the same performance as 110 at least as far as I can see. Somebody on reddit suggested 60 but my performance nosedived. I guess some people increase llc to make up for this but w/e. Intel's whole system is just too complicated as far as I'm concerned. I'm just gonna run intel defaults with ac/dc at 90 and turbo 3 off, vid limit of 1.5 and try to forget about it.
.
.
.
.
.
.
.
Off topic but last time I ran this test on my 14700k, the score was 14700, I'm not even kidding... I think I still have the pic.....

View attachment 358607
lol!

Your pretend 13700k score probably about right as you dont have two 5.4 clocking scores. Although that isnt a stress test.

Everyone has their own way of stress testing, I prefer to not go overboard so for me, its usually a few runs of each cinebench, asusbench 30 minutes, and 5 minute run of each xtu stress test (standard, avx, avx2, memory), and also the xtu bench, as I have in the past had bench cause instability. Also memory testing but no need in your case as you only adjusted the cpu params. Then just using the machine, as transient loads always have the potential to throw something up. Keep an eye on WHEA counter as well, hwinfo does log the counter for conveniance. I think with the loadline you set, I dont see why the chip would be unstable though as you still on the stock v/f curve.

Now intel have confirmed wattage isnt the root cause but its voltage, I may change my pl1/pl2 back to 175/175. Although I dont anticipate any workload I would run that would need that wattage, as I have only ever exceeded 125w in a benchmark or stress application. 175 is about right for my cooler/ambient/case airflow combination.

I seen your post in the other thread as well, and i do agree with the other guys, your chip is 5 year warranty now, and even at stock you are well within Intel's guidelines. But if you are uncomfortable above 1.4v, it looks like this 13700k config might be an acceptable middle ground for you.