MSI BIOS Update to address Intel Raptor Lake Instability - enforce Intel standard power limits?

[jlkoras12df02b8]

Will MSI be issuing a BIOS update yo address the ongoing Raptor Lake instability issues. ASUS has already issued a beta update allowing for a enforcement to standard Raptor Lake power limits... Will MSI be following suit? I am currently sitting with a Z790 Godlike with Intel 14900K and experiencing issues regardless of what I do to tweak the BIOS. While some tweaks make situation better it never seems to resolve without drastic power reductions. Can we just get a BIOS with enforced standards and I'll determine my overclocks, is update in the works? If there is already a current tread on this then I apologize and ask moderator to move.

 

[FlyingScot]

That leaves one more question I just remembered: In your collect experience, do MSI motherboards tend to over report or under report Vcore to apps like HWINFO? And is it generally recommended to stick with the default VCC Sense rather than the optional Socket Sense?

 

[polishsirhawk]

Lite Load is the same as SVID Behaviour, the caveat is, MSI gives you lots of numerical presets and Asus just gives a handful of options that increase AC Loadline A LOT. Then, for fine tune both companies give you the same granular option as well. It impacts VID constantly, since you can't disable it, and always stick to VCC Sense too.

 

[FlyingScot]

Lite Load is the same as SVID Behaviour, the caveat is, MSI gives you lots of numerical presets and Asus just gives a handful of options that increase AC Loadline A LOT. Then, for fine tune both companies give you the same granular option as well. It impacts VID constantly, since you can't disable it, and always stick to VCC Sense too.

Interesting.

From following BuildZoid and Gamers Nexus testing, it does (or at least did) appear that MSI motherboards generally over reported Vcore a little, which I would imagine would be far more desirable than the other way round, like with some of the motherboard models from other manufacturers. Under reporting can lead to possible damage from users thinking their 1.4v, etc, were still fine, but not realizing that they had no headroom left for transient spikes as the real voltage was much higher, especially when running aggressive LLC settings. I was looking to see if you guys had any experience with this.

 

[citay]

MSI defaults to level 8, which is 0.8 mOhm, so pretty much works with every CPU and is in a good spot in terms of making sure no CPUs crash under Intel's guidance.

Depends on the board and potentially also the CPU. I've mostly seen default values between Mode 8 to 13. But yeah, MSI err on the side of caution here, that's why - if the circumstances allow for it - there is a huge potential for improvement in terms of lower power consumption (and thus resulting heat) from tuning this setting.

- CPU quality
- VRM impedance (LLC level)
- CPU SKU

Yes, the CPU quality would probably be the most important factor, and of course it is "per CPU model" (SKU). Within the same model, there is a certain VID range (low VID = high quality and vice versa, as discussed), but the VID range is generally at the highest level for an i9, at a lower level for an i7, and again at much lower level for an i5. While for a 14900K, the VID range might be between 1.4 and 1.5V, for a 14600K it might be between 1.2 and 1.4V. That doesn't mean that the 14600K generally has much higher CPU quality, it just means that it's more in the comfort zone for the silicon as an overall package, it's not pushed nearly as far in regards to frequencies etc., so it's satisfied with lower voltages. Still, between the individual 14600K samples, there's a large difference in CPU quality, just at their own specific VID range.

This goes back to what i wrote here, in that thread, he couldn't lower CPU Lite Load for his 14900K at all in the end. Those CPU models are just pushed very hard from factory to reach those clock speeds, and what's more, all the best 14900K are getting binned to become a 14900KS. So the remaining 14900K, you will have less chance to encounter a sample that you can undervolt by a lot, because that would be the higher silicon quality that would qualify for a -KS which can be sold at a higher price (3.5% aka 200 MHz higher peak turbo clocks and sold for 20-25% higher price). Of course the silicon is screaming for mercy then, the most these CPUs should be pushed (if Intel cared for efficiency) should probably be mid-5-GHz or so. But they can only compete with AMD on the high-end level by using too high clock speeds which carry some downsides in high power draw, the resulting high heat and low efficiency, and even instability at stock sometimes.

Gigabyte's rushed bios is hilarious in a way, since they're applying 1.7 mOhm AC LL, which according to Intel's datasheet is meant for 35W CPUs, the rest should only see a maximum of 1.1 mOhm. So GB is violating Intel's guidance.

Yes, that's the gist of the translated article, and also what Buildzoid noted in his video (watched at bunch of it now at 1.25x speed). While at the previous defaults, they had the AC Loadline setting too low, leading to instability at stock. So unlike MSI erring on the side of caution, Gigabyte were perhaps overconfident in how low this AC Loadline could be set. Realizing this however, they have now backpedaled way too far in the newly rushed out beta BIOS. Not only in these settings (AC / DC Loadline way too high at 170), they have also misintepreted the power limits, which from 12th gen onwards are always PL1 = PL2. Gigabyte however set different limits and way too low (PL1 LONG 125W / PL2 SHORT 188W / ICCMAX 249A) instead of something like 253W 253W 307A. With a Long Power Limit of 125W, an i9 is slowed down significantly for fully multithreaded load. So at the moment, nobody knows what they are smoking in their BIOS dev department.

Here's hoping that Intel, as promised for this month, will bring out a comprehensive statement with more detailed numbers for certain things, instead of vague references to "Intel® specified operating conditions".

Case in point is what happens to VCCSA and IO when you implement XMP. I literally spent months researching that one in an effort to answer the basic question of what do they do, what is safe, and what is stable? I found a starting point and then I tested from there. I'm sure I speak for all of us newcomers when I say it's crazy how much information and misinformation is out there, with very little to help you pick through it. Again, thanks goes to Citay for some of his great posts on the topic. I would follow BuildZoid's advice, but he does so love to play with fire when it comes to VCCSA, etc. And he once admitted that he has a wall of dead CPUs that he degraded. Yikes!

Yes, XMP and what the board/BIOS sets once you enable it for RAM kits of different speeds, that's another area where the board makers have completely free reign. And here they only care for stability, as in, less people complaining to their support team about problems. They don't care that they're setting the voltages excessively high, they don't care that they're potentially degrading IMCs when enabling XMP for an enthusiast-grade RAM kit makes the IMC-related voltages shoot up sky-high, they only care that challenging RAM configurations have a better chance of running. This became evident when the AM5 X3D CPUs blew up from excessive SoC voltage.

About Buildzoid, you don't go there if you are looking for sane voltages, i posted about it here. His channel is not called "Actually Hardcore Overclocking" for nothing. But also, when he says now that Gigabyte are using settings in their new beta BIOS which are resulting in crazy high voltages sometimes, that has an extra weight to it. If he calls a voltage too high, you know it's too high.

About Load Line Calibration, as i said, this is something i need to test more myself when i have the time and patience for it. With my lowly i5-13500, i'm far away from any problems like people with a 14900K can experience, so i don't have a pressing need to experiment a lot, i only optimized CPU Lite Load and some other settings and i was satisfied with the results. But there can probably be some further improvements with LLC.

 

[octor]

CPU Lite Load Normal Mode is literally just a bunch of AC and DC LL presets (I think even each step is 0.1 mOhm/10?), where in the old bioses one was tied to the other, which was just stupid. MSI defaults to level 8, which is 0.8 mOhm, so pretty much works with every CPU and is in a good spot in terms of making sure no CPUs crash under Intel's guidance.

I think you got your values wrong, at least for 13th Gen MZI Z boards (Z690, Z790) paired with 13th Gen CPUs. I haven't tested 14th Gen but I guess impedance values should be pretty similar.
So Lite Load Mode 9 sets AC/DC to 0.5 mOhm / 0.8 mOhm. Explicit Mode 12 (different from Auto shown as Mode 12) sets 0.8 / 0.8 mOhm.
Mode 7 which I use sets it to 0.35 mOhm / 0.8 mOhm etc.
There was a complete listing/mapping of all modes, but Mode 8 doesn't map to 0.8 mOhm and I doubt it ever did for 13th gen and onwards.
All of these values can be inspected in HwInfo64's CPU page (not sensors, main UI) once set in the BIOS.

 

[polishsirhawk]

So either newer mobos or newer bioses have presets changed, on Z690-A 1.24U2 bios it was just such crude stepping, on Z790-P no-idea-which-bios and 13900k it was also that way (which I wouldn't be surprised, since Z790-P~=Z690-A).

Still, DC should be strictly tied to LLC, which I think it still is not on MSI boards.

 

[dr.abdulhakim.ahme158802e3]

I have a 13900K on MSI Z790 Carbon WiFi motherboard using both MSI GPU 4090 and MSI CPU water cooling. The CPU Power and Current specs are 253W and 307A respectively.

My question: To overcome the high power usage,
Can't we just set/change the "Long/Short Duration Power Limit" to 253W and the "CPU Current Limit" to 307A as shown in the image below to allow for the default Lite Load Control to work?
If not, what is the detrimental result if we do this?

 

[polishsirhawk]

LLC is short for Loadline Calibration, it exists to fight VRM's impedance and on auto it defaults to a predefined, singular value (it's in the advanced voltage tab), you can't affect it the tiniest bit through power/current limiting. Weaker LLC = bigger vdroop, thus smaller power draw but ONLY if you've been actually current limited. If you've been power limited, then the bigger vdroop won't do anything, since the CPU will be able to pull more amps and the resulting power draw will remain the same.

 

[octor]

Still, DC should be strictly tied to LLC, which I think it still is not on MSI boards.

Yes, I also understand they need to be equal in order to represent/calculate power and VID numbers correctly.
At least on my board, I've observed the default LLC should be higher impedance than the 0.8 mOhm Lite Load Modes 1-12 Have. Probably around 1.02 mOhm.
How I got there? If my understanding is correct (my sources are the hardware forums, obviously), a fairly practical method to determine aproximate LLC impedance is to adjust DC Loadline so that under heavy all core load (used Cinebench R23 for that) you get P-core's VID matching the VCore. Then you have found it. In my case that was DC Loadline set to 1.02 mOhm. The default one, 0.8 mOhm, under the same heavy load, always gets me VID a bit higher than VCore even under load, but by a very small margin (like VID of one random P-core is 1.194 V, VCore shows 1.17 V).
But at the end of the day I settled on using the simple Normal Lite Load since these deviations in power are small.

My question: To overcome the high power usage,
Can't we just set/change the "Long/Short Duration Power Limit" to 253W and the "CPU Current Limit" to 307A as shown in the image below to allow for the default LLC to work?
If not, what is the detrimental result if we do this?

You can certainly do that and call it a day, if you're satisfied with your cooling and thermals, and like to run your CPU at stock performance. Nothing detrimental about it.
I don't do overclocking so I'm an advocate of stock PL and IccMax (Current Limit), also didn't fiddle with the default LLC, which has big Vdroop.
I think the Mode 9 default uses fairly sane values of 50/80 (0.5/0.8 mOhm), but this depends on motherboard model, BIOS version and CPU model and can be checked outside BIOS. Anyway, the AC Loadline is the one influencing voltage hence power and heat and gets changed by the modes.
However, if you want a bit more thermal improvement, lowering Lite Load Mode will give you that, but as @citay noted, you will reach a stability wall somewhere down the line. So it depends how geeky you are about it. On my CPU/mobo I decreased from the default Mode 9 to Mode 7 and called it a day. The switch lowered my temps by 3-4 degrees under heavy load, associated with 8W lower power consumption. I would say how much you should care about this depends on your cooling capabilities. I saw there was some room for improvement in my case, but maybe your cooling is better. At stock config with no overclocking and official limits, performance will be the same.

 

[polishsirhawk]

How I got there? If my understanding is correct (my sources are the hardware forums, obviously), a fairly practical method to determine aproximate LLC impedance is to adjust DC Loadline so that under heavy all core load (used Cinebench R23 for that) you get P-core's VID matching the VCore. Then you have found it.

9/10 points. The necessary load is any 100% all core workload, resulting in a 100% CPU utilization, so a memtest is good for that too, so the CPU would not be hammered.

 

[octor]

9/10 points. The necessary load is any 100% all core workload, resulting in a 100% CPU utilization, so a memtest is good for that too, so the CPU would not be hammered.

Maybe, since an approximation is wanted, but you will see that the distance between VID and VCore is not identical regardless of the all core load. For example between Cinebench R23 and Prime 95. The higher the power consumption, the lower the VCore, and the greater the VID-VCore diff, by comparing between these two.

 

[polishsirhawk]

Maybe, since an approximation is wanted, but you will see that the distance between VID and VCore is not identical regardless of the all core load. For example between Cinebench R23 and Prime 95. The higher the power consumption, the lower the VCore, and the greater the VID-VCore diff, by comparing between these two.

No, under proper DC LL and 100% all core load the VID to vcore remains almost exactly the same, the caveat is not using averaged reading and having per core VID. Vcore reading can differ from VID when the system isn't under 100% all core workload, but usually that's the case only under idle state or low load.

 

[dr.abdulhakim.ahme158802e3]

LLC is short for Loadline Calibration, it exists to fight VRM's impedance and on auto it defaults to a predefined, singular value (it's in the advanced voltage tab), you can't affect it the tiniest bit through power/current limiting. Weaker LLC = bigger vdroop, thus smaller power draw but ONLY if you've been actually current limited. If you've been power limited, then the bigger vdroop won't do anything, since the CPU will be able to pull more amps and the resulting power draw will remain the same.

Thank you so much for your reply. I meant LLC = Lite Load Control. The question still remains: Do those settings limit the CPU Power/Current Limits so they don't exceed the 13900K specs? If not, what is the detrimental result if we do this?

 

[dr.abdulhakim.ahme158802e3]

Yes, I also understand they need to be equal in order to represent/calculate power and VID numbers correctly.
At least on my board, I've observed the default LLC should be higher impedance than the 0.8 mOhm Lite Load Modes 1-12 Have. Probably around 1.02 mOhm.
How I got there? If my understanding is correct (my sources are the hardware forums, obviously), a fairly practical method to determine aproximate LLC impedance is to adjust DC Loadline so that under heavy all core load (used Cinebench R23 for that) you get P-core's VID matching the VCore. Then you have found it. In my case that was DC Loadline set to 1.02 mOhm. The default one, 0.8 mOhm, under the same heavy load, always gets me VID a bit higher than VCore even under load, but by a very small margin (like VID of one random P-core is 1.194 V, VCore shows 1.17 V).
But at the end of the day I settled on using the simple Normal Lite Load since these deviations in power are small.


You can certainly do that and call it a day, if you're satisfied with your cooling and thermals, and like to run your CPU at stock performance. Nothing detrimental about it.
I don't do overclocking so I'm an advocate of stock PL and IccMax (Current Limit), also didn't fiddle with the default LLC, which has big Vdroop.
I think the Mode 9 default uses fairly sane values of 50/80 (0.5/0.8 mOhm), but this depends on motherboard model, BIOS version and CPU model and can be checked outside BIOS. Anyway, the AC Loadline is the one influencing voltage hence power and heat and gets changed by the modes.
However, if you want a bit more thermal improvement, lowering Lite Load Mode will give you that, but as @citay noted, you will reach a stability wall somewhere down the line. So it depends how geeky you are about it. On my CPU/mobo I decreased from the default Mode 9 to Mode 7 and called it a day. The switch lowered my temps by 3-4 degrees under heavy load, associated with 8W lower power consumption. I would say how much you should care about this depends on your cooling capabilities. I saw there was some room for improvement in my case, but maybe your cooling is better. At stock config with no overclocking and official limits, performance will be the same.

Thank you so much for your reply. I have both the 13900K and 4090 GPU Water Cooled. I'm just afraid of going over the power limits as 4096W is more than 15 times 253W CPU power limit. I don't want to end up burning my CPU just because the motherboards specs far exceed the CPU limits. Any thoughts?

 

[polishsirhawk]

Thank you so much for your reply. I meant LLC = Lite Load Control. The question still remains: Do those settings limit the CPU Power/Current Limits so they don't exceed the 13900K specs? If not, what is the detrimental result if we do this?

AC LL is an impedance correction, that's used for VID formula, so all it affects is voltage fed to the CPU. The detrimental result by setting it too high is unnecesary overvolt.

 

[citay]

Intel has just released some new guidance for the default BIOS values, see the post here.

 

[citay]

And now there's an official blog post from MSI Taiwan as well as first BIOS updates with MSI's solution.

Original blog post in Chinese: https://tw.msi.com/blog/improving-s...l-core-processors-with-intel-default-settings

Google-translated to English: https://tw-msi-com.translate.goog/b...l=auto&_x_tr_tl=en&_x_tr_hl=en&_x_tr_pto=wapp

First beta BIOS updates for some Z790 boards (example from the Z790 Carbon WIFI):

Let's look at what they have done at MSI. Rather logically and predictably, they have chosen the cooler selection prompt (which should appear on the very first boot or after a BIOS update) as a way to set the recommended values, if one wishes so. I have criticized that cooler selection prompt (which is really just the power limit prompt) in the past, because the three options were not chosen very wisely. The "Tower Air Cooler" selection was easily too much for even the best tower coolers, in case of a CPU model like a 14th gen i7 or i9 which really want to go to 288W of power draw and beyond. Then the "Water Cooler" selection was just everything maxed out.

But let's see what the revised prompt in the new BIOS does now.

"Boxed cooler" now becomes "Intel Default Settings", which implements the "Performance Profile" as previously explained in this post. For a 13th/14th gen i9 (-K/-KF), this sets a PL2 (Short Duration Power Limit) of 253W, a PL1 (Long Duration Power Limit) of 125W, and Current Limit IccMax of 307A. And presumably, these values would change to 253W PL1/PL2 and 307A IccMax with a 13900KS/14900KS. In other words, exactly what Intel recommends for the Performance Profile. Additionally, they implemented a warning text, which also appears as a pop-up prompt when selecting Tower or Water cooler in the OC section of the BIOS later. So far, so good.

Well, "Intel Default Settings" might not be that good for some owners of a 13900K/14900K, since they will experience a significant loss of performance under sustained fully multithreaded workloads. Because after around a minute of full load, the CPU will be limited to 125W of power draw, while we all know that a 14900K would gladly sit at 2.5x to 3x that power draw, if given the chance. At 125W Long Duration power limit, it will have roughly 74% of the performance compared to running with the power limits fully maxed out. Of course, it also has only a third of the power draw, so the efficiency is much better. But still, this 125W limit after a minute of full load is rather harsh for an i9, which in most cases should be used with a more powerful cooler that can deal with at least 200W, if not 253W of heat indefinitely.

And once we look at the two other options, we again notice a wasted potential. MSI could've easily renamed the three options:
1) Intel Default Settings (Performance Profile)
2) Intel Default Settings (Extreme Profile)
3) Unlimited

In other words, have the second option set the Extreme Profile of 253W PL1/PL2 and 400A IccMax for a normal i9, and 320W PL1/PL2 and 400A for an i9 -KS version. Then only have the "Unlimited" setting max out everything for those with crazy powerful cooling or overclockers.

But instead, they are keeping the previous Tower Cooling option, setting 288W PL1/PL2 (again, too much for any tower cooler, this can trouble even some 360mm AIOs), with the current limit completely maxed out, which Intel don't even allow in the Extreme Profile for a 14900KS ("Never exceed 400A")! And yeah, the Water Cooler option is the unlimited option.

I would say, this solution is not ideal, or at least it still has room for improvement.

Another question mark is about the AC/DC Loadline settings (with MSI, they're controlled by the "CPU Lite Load" options). Intel mentions them in their recommendations, but they basically leave it up to the board vendors to set suitable default values (which can also be different according to each board model's VRM capabilities). There can be problems when setting these values too high (excessive voltages and power draw) and too low (instability), and the ideal values also depend on the silicon quality of each individual CPU. So it's not easy to set values which work well for all CPUs without going too much in either direction (too much voltage for most CPUs vs. some CPUs becoming unstable from too low voltage).

 

[octor]

Another question mark is about the AC/DC Loadline settings (with MSI, they're controlled by the "CPU Lite Load" options). Intel mentions them in their recommendations, but they basically leave it up to the board vendors to set suitable default values (which can also be different according to each board model's VRM capabilities). There can be problems when setting these values too high (excessive voltages and power draw) and too low (instability), and the ideal values also depend on the silicon quality of each individual CPU. So it's not easy to set values which work well for all CPUs without going too much in either direction (too much voltage for most CPUs vs. some CPUs becoming unstable from too low voltage).

Exactly my curiosity as well!
Can't wait too see it although I think I'm a bit hesitant to jump to the latest when it shows up. Are you on latest these days, Citay, or which Z790 BIOS do you currently recommend?

I noticed a tendency of improvement when I updated to the September 2023 BIOS, which had as default the Lite Load Auto Mode 9 with AC/DC Loadline of 0.5/0.8, reasonable values. Previously, defaults were 1.1/1.1 (Jan 2023 BIOS) with a lot of heat and voltage. I further dropped it to Mode 7 0.35/0.8 for my CPU and never looked back. Never noticed any stability issue either, although I haven't tested it with Unreal Engine shader compilation, which seems to be the ultimate stability test nowadays.

Also I observe Intel are now recommending CEP to be active, which never was by default on my MSI board. I played around a bit and it only brings heat (14W more in CB23 at same clocks) and a minor performance decrease to 23K from around 23.6K (2.5% decrease). Unless I see any instability I find little reason to enable it just to comply with recommendations. What are your thoughts on this? Do you think not having it enabled could somehow put CPU in danger by way of more than normal current draw or otherwise?

 

[RemusM]

Another question mark is about the AC/DC Loadline settings (with MSI, they're controlled by the "CPU Lite Load" options). Intel mentions them in their recommendations, but they basically leave it up to the board vendors to set suitable default values (which can also be different according to each board model's VRM capabilities). There can be problems when setting these values too high (excessive voltages and power draw) and too low (instability), and the ideal values also depend on the silicon quality of each individual CPU. So it's not easy to set values which work well for all CPUs without going too much in either direction (too much voltage for most CPUs vs. some CPUs becoming unstable from too low voltage).

This is true for all the default settings.
And the only solution to cover all the CPU samples with the default values is to use higher than "normal" values.
It's as simple as that (and many people still don't get it).
Those open to "tinkering" are free to tune up their system.
But MSI, ASUS, Gigabyte and all the other manufacturers need to do something for the rest.
And the rest represents more than 90% of the users.

 

[polishsirhawk]

Manufacturers ignored the existence of vdroop by not implementing LLC bump with their unlimited power plan profiles, that welcome users upon first bios visit and look how it went for them :^)