Why YOU need LLC!!

[borisvodofsky]

Scenario, 1.33volts "MINIMUM" required @ 4.8Ghz, anything above is stable.

I must target 1.33volts minimum condition, because if it drops below this in ANY situation, the CPU will be unstable.

Voltage Feed to CPU at various loads

The numbers below have nothing to do with Vdroop, these are the voltage ranges that are CONSISTENTLY fed to the CPU during the task.

LLC off:

Prime 95 small-ftt: 1.320v-1.328v
Intel Burn test, max memory 6.8GB: 1.328v-1.336v

HALO (Pc-game): 1.368v-1.376v
MPC-HC 1080p Decoding 27Gb MKV Lav+Madvr:1.368v-1.376v
^^ extremely high voltages for low load tasks, 20-40% load,


LLC (High):

Prime 95 small-ftt: 1.32v-1.33v
Intel Burn test, max memory 6.8GB: 1.344v-1.352v

HALO (Pc-game): 1.344-1.352
MPC-HC 1080p Decoding 27Gb MKV Lav+Madvr:1.336-1.344
^^ much better.


As you can see, Daily tasks (Halo and Video playback), without LLC will use an absurd amount of voltage under intel's built in VID power scheme.

Again, I emphasize, these numbers have nothing to do with vdroop!



Tested on Asus z68 pro gen3, 3xxx bios.

Unproven allegations:

-Voltage overshoots
-Stresses memory controller

 

[Kenmitch]

Strange how you draw the line at acceptable voltages vs extremely high voltage. Most people would consider 0.02v's a little to nothing.

 

[JAG87]

Listen, I'm telling you for the last time:

Whatever your motherboard's droop level is, you are getting the exact same oscillation whether LLC is on or off. You just don't see it.

If your motherboard droops 0.05V under load (i.e. LLC off 1.37V during light load, 1.32V during P95), you're getting the exact same fluctuation when entering load with LLC on. Your just forcing the VRM to recover quickly to 1.35V (which actually makes your VRM less power efficient). So when you start P95 or IBT from 1.35V at idle, you're actually dipping down 1.30V for a very brief moment, and then recovering to 1.35V quickly. And when you exit load, you're now overshooting to 1.4V, and then going back down to 1.35V.

This doesn't equal stability, nor power efficiency.

You're not stable (even thought it might appear so), because you have no control over your voltage regulation, and whatever wattage your saving on your CPU by running a few mV less, you're burning it in the VRM.

End of story.

 

[borisvodofsky]

Listen, I'm telling you for the last time:

Whatever your motherboard's droop level is, you are getting the exact same oscillation whether LLC is on or off. You just don't see it.

If your motherboard droops 0.05V under load (i.e. LLC off 1.37V during light load, 1.32V during P95), you're getting the exact same fluctuation when entering load with LLC on. Your just forcing the VRM to recover quickly to 1.35V (which actually makes your VRM less power efficient). So when you start P95 or IBT from 1.35V at idle, you're actually dipping down 1.30V for a very brief moment, and then recovering to 1.35V quickly. And when you exit load, you're now overshooting to 1.4V, and then going back down to 1.35V.

This doesn't equal stability, nor power efficiency.

You're not stable (even thought it might appear so), because you have no control over your voltage regulation, and whatever wattage your saving on your CPU by running a few mV less, you're burning it in the VRM.

End of story.

You are so lost, my friend.

I am NOT talking about voltages Moving up and down.

Those numbers I posted, are STATIC numbers that do not move, as long as the task it's doing continues. They are NOT vdroop numbers.

What LLC does for me has nothing to do with Vdroop. It has everything to do with LIMITING the maximum voltage at "Low load tasks, approx 50%"

 

[borisvodofsky]

Strange how you draw the line at acceptable voltages vs extremely high voltage. Most people would consider 0.02v's a little to nothing.

The thermal difference between 1.35v and 1.376v is 5* celcius :thumbsup:

 

[JAG87]

You are so lost, my friend.

I am NOT talking about voltages Moving up and down.

Those numbers I posted, are STATIC numbers that do not move, as long as the task it's doing continues. They are NOT vdroop numbers.

What LLC does for me has nothing to do with Vdroop. It has everything to do with LIMITING the maximum voltage at "Low load tasks, approx 50%"

It doesn't limit anything because you obviously don't understand how it works.

How much are you raising your vcore when you turn LLC off to maintain stability?

If you get 1.352 with LLC high, and 1.376 with LLC off, clearly you're not using the same vcore.

 

[borisvodofsky]

It doesn't limit anything because you obviously don't understand how it works.

It is you that have no idea how this works. You are confusing Vdroop with eist.

I am NOT talking about Vdroop. This is not about Vdroop, you have a droopy addiction.

Without LLC

The Motherboard feeds 1.376 CONSTANT volts to my CPU while playing halo/decode HD video.

IT also feeds 1.33 CONSTANT volts to my CPU while prime/ Intel burn test


WITH LLC

Prime / IBT goes higher 1.352v

HOWEVER

HALO/ Decode HD is also 1.352v, down from 1.376

 

[borisvodofsky]

It doesn't limit anything because you obviously don't understand how it works.

How much are you raising your vcore when you turn LLC off to maintain stability?

If you get 1.352 with LLC high, and 1.376 with LLC off, clearly you're not using the same vcore.

Yes I am using different voltages, because I need to maintain the 1.33 minimum.

I know for fact that prime 95 small ftt full load requires 1.33 volts, If the voltage drops below this threshold, the system becomes unstable.

However to meet this setting, I need + 0.03v offset in bios
This amounts to prime 95 voltage of 1.33, Simultaneously, 1.376 at light loads


To limit that range, I must use LLC

Under LLC, Prime small ftt will still consume 1.33v, However, low load power is cut to 1.54v.


I am trying to fill 2 conditions, 1.33 minimum, and at the same time achieving the lowest light load voltage, which requires LLC.


I am starting to suspect you've been out of the loop a long time, or has never even used a new iteration motherboard with LLC.

If all you've done is read the Anandtech article you've been completely mislead.

Because that article offers a completely wrong interpretation on the purpose of LLC. Of which is NOT to combat V-droop. While it Does do this, the purpose is actually to LOWER the maximum voltage during light loads.

 

[Coup27]

I get the impression this thread was so one person could continue his discussion with one other person in another thread.

 

[borisvodofsky]

I get the impression this thread was so one person could continue his discussion with one other person in another thread.

No I'm just trying to help people like you.

 

[Annisman*]

I have very little technical knowledge of this topic, but I remember very distinctly that LLC was a MUST when using my old X38 P5E, and trying to succesfully OC my E8400.

 

[Coup27]

No I'm just trying to help people like you.

Mate I have no idea what LLC is and if you caught my rig specs I have no interest in overclocking either! Apart from setting a few personal preferences my mobo UEFI is largely ASUS defaults so whatever ASUS recommends I am happy with!

 

[Coup27]

I just rebooted and my LLC is set to auto, with options of auto, enabled and disabled. Not sure which camp that puts me in.

 

[Yuriman]

My real-world experience with LLC vs no LLC:

With LLC on, I need a higher voltage (as reported by CPU-Z) to get stable at a given clock. I have a lower offset, but overall the voltage in my chip is higher.

With LLC, full load minimum stable voltage: 1.280v
Without: 1.248v

It seems it's when changing loads and clocks that I get crashes with LLC on, and it's at full load when voltage is lowest that I would crash with LLC off.

Edit: I don't presume to know which is better for my CPU, but LLC off results in lower power consumption.

 

[fastamdman]

It also depends on whether or not eist is being used.

 

[reb0rn]

I have eist + c6 and all saving on for oc from 4.6Ghz same for 5Ghz, and that is the lowest power consumption
with LLC on hight it was a mess to make it stable with tasks that use just one CPU core (I have asus p67)

Duno for P77 chipset or for future, for my point LLC still need tweaking from mobos maker/intel

 

[Puppies04]

Unproven allegations: -Voltage overshoots -Stresses memory controller

Care to explain how the voltage doesn't overshoot, this is a simple fact of physics. If you choose not to believe it that is your problem but I suggest you go and find a bit more proof other than the 1 source you linked in the other LLC thread before you go insinuating that something that has to happen doesn't.

 

[cmdrdredd]

I have eist + c6 and all saving on for oc from 4.6Ghz same for 5Ghz, and that is the lowest power consumption
with LLC on hight it was a mess to make it stable with tasks that use just one CPU core (I have asus p67)

Duno for P77 chipset or for future, for my point LLC still need tweaking from mobos maker/intel

C3 and C6 etc are what causes stability issues. Voltage is jumping all over. With just C1e and speed step, there is not as much movement in between load levels. So generally you see people recommend to turn those off.

That's my understanding of it and I know it's not technical or anything, for that someone else would have to chime in.

 

[JAG87]

Without LLC

The Motherboard feeds 1.376 CONSTANT volts to my CPU while playing halo/decode HD video.

That's because you're using a higher VID when you disable LLC. Why don't you actually spill out what you set your vcore in the bios with LLC on and off. I bet it's something like this:

~1.35 with LLC on
~1.38 with LLC off (because you wouldn't sustain 1.328V under P95 otherwise)

Of course voltage is going to be higher in light loads. It's not LLC doind anything, it's you.

IT also feeds 1.33 CONSTANT volts to my CPU while prime/ Intel burn test

So would 1.384V with LLC off. And it would guarantee that you never feed less than 1.328V under load, and never feed it more than 1.384V when exiting load.

With LLC on, you have zero guarantees. With 1.35V set in bios, you're undershooting to ~1.30V briefly (because ~0.05V is what your motherboard manufacturer calibrated the droop to be, and different boards can have different amount of droop) and because your VRM is quickly rising up to 1.328V again (and wasting more power when doing so), when you exit load you're briefly overshooting to ~1.38V.

And if you use the Extreme setting, you'd be even worse off. You would maintain 1.35V under load, undershoot briefly to ~1.30V entering load, and overshoot briefly to ~1.40V exiting load.

This is how it is. You don't have some magical VRM that maintains linear voltage when current demands goes from 10 amps to 100 amps. What you choose to believe is up to you, but don't spread BS.

 

[borisvodofsky]

My real-world experience with LLC vs no LLC:

With LLC on, I need a higher voltage (as reported by CPU-Z) to get stable at a given clock. I have a lower offset, but overall the voltage in my chip is higher.

With LLC, full load minimum stable voltage: 1.280v
Without: 1.248v

It seems it's when changing loads and clocks that I get crashes with LLC on, and it's at full load when voltage is lowest that I would crash with LLC off.

Edit: I don't presume to know which is better for my CPU, but LLC off results in lower power consumption.

It may be true that Your FULL load will be stable at a lower voltage.

But your @ 50% load, it will pump a much higher voltage. This is the problem.

Your voltage requirements for stable DOES NOT change whether you have LLC on or Off, your cpu NEEDS X-voltage to be stable. It's a single LINE to cross.

 

[borisvodofsky]

That's because you're using a higher VID when you disable LLC. Why don't you actually spill out what you set your vcore in the bios with LLC on and off. I bet it's something like this:

~1.35 with LLC on
~1.38 with LLC off (because you wouldn't sustain 1.328V under P95 otherwise)

Of course voltage is going to be higher in light loads. It's not LLC doind anything, it's you.





So would 1.384V with LLC off. And it would guarantee that you never feed less than 1.328V under load, and never feed it more than 1.384V when exiting load.

With LLC on, you have zero guarantees. With 1.35V set in bios, you're undershooting to ~1.30V briefly (because ~0.05V is what your motherboard manufacturer calibrated the droop to be, and different boards can have different amount of droop) and because your VRM is quickly rising up to 1.328V again (and wasting more power when doing so), when you exit load you're briefly overshooting to ~1.38V.

And if you use the Extreme setting, you'd be even worse off. You would maintain 1.35V under load, undershoot briefly to ~1.30V entering load, and overshoot briefly to ~1.40V exiting load.

This is how it is. You don't have some magical VRM that maintains linear voltage when current demands goes from 10 amps to 100 amps. What you choose to believe is up to you, but don't spread BS.

You have no clue what you're talking about. It is you who's spreading mis-information.

I "target" 1.33 minimum. Using whatever offset voltage to guarantee that number.

The minimum at full load without memory controller engaged, happens in Prime 95 small FTT.

So no matter what, I have to hit 1.33, otherwise it'll be unstable at low loads 4.8ghz.

NOW, from there I have 2 option,

Use LLC or Don't use LLC



If i use LLC, I need a -0.025v offset to get the stable 1.33v in prime small ftt

my IBT/Halo/Movie voltage is a "CONSTANT" 1.35, this has no entering or exiting, attached. As the movie is playing, as halo is running, as IBT is burning, I GET 1.35v under CPUz


If i do NOT use LLC, I need a +0.06v offset to get stable 1.33v in prime small ftt

my IBT/Halo/Movie voltage is a "CONSTANT" 1.37v, again, this has nothing to do with entering or exiting load. AS THESE TASKS are running, I SEE 1.37v in CPUz.


MY ONLY CHOICE is to use LLC, because otherwise, My two voltage brackets are 1.33 and 1.37, the 1.37 is during simple movie playback, which is way to high.

However, if I use LLC, it feeds a constant 1.35v to my CPU during low load 4.8ghz.

 

[Yuriman]

boris, I tested and you're right, kindof. Loading only a single core brings my vcore up to a peak of 1.288v which is 0.008v higher than my lowest stable voltage with LLC. However when loading 2 or more cores my voltage drops below the lowest that's stable with LLC and it remains stable. It looks like if I'm only using 1 core I'm best off with LLC on, but if more cores are in use then I'm feeding my chip more voltage than it needs.

Because I'm using Intel's utility to overclock, changing my offset doesn't affect any clocks but my highest so that's a non-issue.

 

[JAG87]

You have no clue what you're talking about. It is you who's spreading mis-information.

I "target" 1.33 minimum. Using whatever offset voltage to guarantee that number.

The minimum at full load without memory controller engaged, happens in Prime 95 small FTT.

So no matter what, I have to hit 1.33, otherwise it'll be unstable at low loads 4.8ghz.

NOW, from there I have 2 option,

Use LLC or Don't use LLC



If i use LLC, I need a -0.25v offset to get the stable 1.33v in prime small ftt

my IBT/Halo/Movie voltage is a "CONSTANT" 1.35, this has no entering or exiting, attached. As the movie is playing, as halo is running, as IBT is burning, I GET 1.35v under CPUz


If i do NOT use LLC, I need a +0.6v offset to get stable 1.33v in prime small ftt

my IBT/Halo/Movie voltage is a "CONSTANT" 1.37v, again, this has nothing to do with entering or exiting load. AS THESE TASKS are running, I SEE 1.37v in CPUz.


MY ONLY CHOICE is to use LLC, because otherwise, My two voltage brackets are 1.33 and 1.37, the 1.37 is during simple movie playback, which is way to high.

However, if I use LLC, it feeds a constant 1.35v to my CPU during low load 4.8ghz.

Okay, you're right. Go ahead with LLC.

You clearly know more than Intel engineers about their processors.

 

[Yuriman]

My question is this: why am I stable with LLC on at 1.280v and no lower, but with LLC disabled I'm still stable with all 4 cores loaded when vcore dips down to 1.248v? Granted with LLC disabled, a single core will go as high as 1.288v, but that's such a small amount higher that I think I'll run without LLC.

 

[borisvodofsky]

My question is this: why am I stable with LLC on at 1.280v and no lower, but with LLC disabled I'm still stable with all 4 cores loaded when vcore dips down to 1.248v? Granted with LLC disabled, a single core will go as high as 1.288v, but that's such a small amount higher that I think I'll run without LLC.

The extremely high load voltage while at "low loads" is intel's way of prepping for an EVEN HIGHER LOAD. because when you go from low load to high load, you may overshoot your VID, So instead of overshooting VID, they design it so that it gets high on the VID scale, THEN DROP down to the necessary FULL load voltage. This way, It neither overshoots nor undershoots the necessary Full load VID.

This is counter to what we need in our overclocked cases, BECAUSE our Overclocked setting require postive VID offset that is SIGNIFICANTLY higher than the Intel preset VID.

So during low loads, our constant voltage at the overclocked frequency will be much higher than necessary. Because the processor is gearing up to DROP the voltage


This is exactly where the misinterpretation occured in the anandtech article.

LLC affects vdroop. But the goal of it is to tweak the max voltage to the CPU to be more IN LINE with what it needs, rather than giving it a HUGE excess.

As I mentioned, the Difference between 1.35 and 1.37 is 5* Celcius


Can you post the following voltages, I can help you fine tune.

When you test IBT max mem, CPUz won't refresh the voltages quickly, you must watch it for at least 1 full cycle, about 2 mins. Post the HIGHEST one it reads.

Your offset vid setting with LLC
Prime 95 small ftt
IBT max memory
HIGH bitrate video playback

NEXT

Your offset vid setting without LLC
Prime 95 small ftt
IBT max memory
HIGH bitrate video playback