[AyeYo]

Ok, I lied. It's just voltage...

I was looking for a thread over on XS and came across some Bloomfield technical documents linked in a thread on anandtech. Not only do they blow the idea of absolute maximums being 24/7 safe out of the water and prove what I've been arguing all along, but they also give a new angle to something people don't look into enough before OC'ing.

Yes, the voltage safety line goes DOWN as amperage increases, quite significantly so.

Discuss...

Quote:

Refer to Table 2-8 and Figure 2-3 for the minimum, typical, and maximum VCC allowed for a given current. The processor should not be subjected to any VCC and ICC combination wherein VCC exceeds VCC_MAX for a given current

.

EDIT: Detailed discussion on this issue and the nature of CPU power delivery: http://www.thetechrepository.com/showthread.php?t=126 (link courtesy of *AcidBath*)

Bloomfield

Clarkdale

Lynnfield

 

[infected rat]

Nice post, it's not entirely surprising since voltage is just one part of the equation, we should really concern ourselves with both voltage and current delivered to the processer, in short the total power use.

I'm constantly dismayed on this forum when people say things like oh you're fine up to X Volts, or Y temperature because that's always just one part of a bigger picture. Sometimes people give very reasonable values for X and Y but all too often they are way out of spec. Generally I don't have the energy to get into endless debates about it though because sadly it will be an endless debate. People are obviously more than welcome to push their CPU's to the absolute limit and beyond but I suspect more than a few out there are not so aware that they may well be degrading their chip.

We have so litte overclock data to go on from the people who make the chips themselves, I suppose I like to err on the conservative side of things since I want my CPU to last 4-5 years, even though I'm likely to upgrade before that my older machine always drops back to become the backup/server so I want it to still work and not be degrading, even slightly.

 

[AyeYo]

Indeed.

The other beautiful part about these charts is that they support the idea that many of us here have had that idle Vcore is much less important than load Vcore. However, this was just a hunch/logic argument up until now and, when debating, we never had a solid defense. These charts prove that it's true because idle (actually below idle, 0 amps) is the only time it's safe for the CPU to see it's full VID. As you said, it's total power draw that really counts, not just some voltage figure with no qualifications.

 

[esp42089]

This adds a complication to monitoring your oc; how do you measure Ice to make sure you are not overshooting your max load line? I have no clue if I am drawing more than 140 amps while under load with 1.36 vcore. I suspect I am though. Any ideas?

 

[AyeYo]

Quote:

Originally Posted by esp42089 This adds a complication to monitoring your oc; how do you measure Ice to make sure you are not overshooting your max load line? I have no clue if I am drawing more than 140 amps while under load with 1.36 vcore. I suspect I am though. Any ideas?

I'm not even sure if that's possible. I suppose you could get a wattage meter and attempt to estimate where you're at amperage-wise.

The decrease seems to be totally linear: -0.004v for every 5 amp increase. If you can measure the wattage difference between stock max load and OC max load, you can calculate the amperage difference and tack it on top of the 140 amps the chart goes up to. Then just subtract an additional 0.004v for every 5 amps.

OR

Just stay a solid 0.2v, at load, away from max VID when overclocking 4+ ghz.

Then again, that might be too exaggerated.

Keep in mind, this is AT LOAD. If you have LLC disabled, then your voltage is going to drop at load anyway. So even though you may be close to or at VID in the BIOS, chances are, you're a good bit lower than that under full load. That's most of your safety margin right there.

 

[infected rat]

You can get a guide figure of the processor power draw from things like CPUID hardware monitor but I doubt they are massively accurate. Everest probably gives you the same info and I imagine there are others. As I say I'd use those as guide figures rather than exact but it's something at least. Max TDP of my 920 is 130W and I hit 147W in that monitor program under full LinX load with my 3.8 o/c...

 

[krameriffic]

We're enthusiasts, not electrical engineers or material scientists (for the most part). We don't have the expertise to authoritatively speak to most of this stuff, so we speak to it based on personal experience. That's 95% of what you're getting on here when you ask these questions.

 

[AyeYo]

Quote:

Originally Posted by krameriffic We're enthusiasts, not electrical engineers or material scientists (for the most part). We don't have the expertise to authoritatively speak to most of this stuff, so we speak to it based on personal experience. That's 95% of what you're getting on here when you ask these questions.

One of my local tuning shops has the slogan "knowledge is horsepower," and it's true.

The same applies to overclocking. Just because you aren't an engineer doesn't mean you throw your hands to the sky and just go off guess work. The more you know, the better off you are. You learn as much as you can along the way and always stay open to new experience/information. "Yo man I beenz runnin' 1.77v on my E8400 for sixteen years without an issue, yo!1!!!" shouldn't replace doing as much research as you can, so you can OC as well and as safely as possible. What really blows my mind though is that this board in particular, for whatever reason, tends to go with the homeboy and his 1.77v E8400, never bothering to question things further, while forums like XS and Anandtech are always striving to learn more and get a deeper understanding of how stuff works. That's probably why so many people here are hung up on misinformation, while the stuff those same people consider "too complicated" or "unnecessarily difficult" is common knowledge on other boards.

 

[Blameless]

Still amazes me that after almost three years on Intel's site, most i7 OCers haven't read the white papers.

Also, a similar load line pattern has applied for a long, long, time. I remember seeing a similar graph and chart in the white papers for the P4 Northwoods back in 2004. It's probably in much older papers too.

Quote:

Originally Posted by infected rat You can get a guide figure of the processor power draw from things like CPUID hardware monitor but I doubt they are massively accurate. Everest probably gives you the same info and I imagine there are others. As I say I'd use those as guide figures rather than exact but it's something at least. Max TDP of my 920 is 130W and I hit 147W in that monitor program under full LinX load with my 3.8 o/c...

That isn't an actual power consumption reading, and it's almost certain your CPU is consuming more than 147 watts at 3.8GHz.

Only if you have a voltage control IC that HW Monitor supports can you get a decent estimate of power consumption with it.

 

[Fooxz]

Wow, really cool find ayeyo.

But i do have a question too, when it says VID, does it mean the stock VID of a specific chip, or the maxium VID of that chip type? (ie. "1.4v" for the i3)

 

[ChickenInferno]

Current scales with voltage as power also increases. P=IV. Overclocking increases speed and therefore power draw as well as amperage inherently. Therefore, increasing voltage is helping to make the power more stable.

Look at the chart closely, the voltage scale is flipped from you would normal view. As current increases so does voltage. Though the chart is aimed down, the scales are still positive showing as voltage increases so does current. As I am being told by an electrical engineer, sometimes they flip charts like that to show the voltage difference such that final-minus is negative, but regardless it is still showing a fact that can't be argued P=IV and in this case P is also increasing.

You might want to consider reading about what is actually going on with vcore through a transistor gate channel here.

Edit: I left current completely out of TTATAV, becuase it is inherently linked to voltage and signal quality. Without the proper vcore, signal quality is poor and thus with wrong vcore you are also out of the current spec. Fix one fixes both.

 

[AyeYo]

Quote:

Originally Posted by ChickenInferno Current scales with voltage as power also increases. P=IV. Overclocking increases speed and therefore power draw as well as amperage inherently. Therefore, increasing voltage is helping to make the power more stable. Look at the chart closely, the voltage scale is flipped from you would normal view. As current increases so does voltage. Though the chart is aimed down, the scales are still positive showing as voltage increases so does current. As I am being told by an electrical engineer, sometimes they flip charts like that to show the voltage difference such that final-minus is negative, but regardless it is still showing a fact that can't be argued P=IV and in this case P is also increasing.

Just like your reading of Intel's statements, you aren't looking at the chart right. Voltage is not going up. The top of the Y-axis is VID MINUS 0.000v, i.e. VID. The bottom of the Y-axis is VID MINUS 0.175, i.e. 0.175 below VID.

This can be seen clearly by simply looking at the table in the second picture.

There is absolutely nothing ambiguous here. Voltage tolerance decreases with increasing current draw. For further clarification, see the quoted statement in the OP.

If it still isn't obvious for you, maybe the VTT chart will clarify things. VTT can actually handle a little MORE voltage at very low amps, so it makes it abundantly obvious that safe voltage is going DOWN with increasing amperage.
.

 

[AyeYo]

Quote:

Originally Posted by Fooxz Wow, really cool find ayeyo. But i do have a question too, when it says VID, does it mean the stock VID of a specific chip, or the maxium VID of that chip type? (ie. "1.4v" for the i3)

The chart is intended to be used in reference to the stock VID of the given chip. Intel always talks about VID in relation to a given stock chip because they aren't thinking about over-volting.

However, for OC'ing purposes, it makes sense to use the maximum VID as the reference point, as we know that's the maximum safe 24/7 voltage for the chip type (1.40v for the i3, as you gave for an example), MINUS whatever the table in the OP calls for at the amperage you're running the chip at.

 

[ChickenInferno]

Quote:

Originally Posted by AyeYo Just like your reading of Intel's statements, you aren't looking at the chart right. Voltage is not going up. The top of the Y-axis is VID MINUS 0.000v, i.e. VID. The bottom of the Y-axis is VID MINUS 0.175, i.e. 0.175 below VID.

Yes, I really did read the chart wrong. Myself and my friend Andrew were trying to figure out what the chart was really saying, because frankly it didn't look right. The personal attack however was unwarranted. There is a reason my thread became a sticky very fast; I stated things directly as they were and cited my information. Even then, there was alot learned over time and edits were made based upon more information over time.

Now to address the points.

Your chart shows VID at stock conditions. Everything you say about the chart is exactly correct. For a given stock speed and for a given range of voltage, there is a corresponding range of currents that is a narrow limit. This goes to signal quality and power quality. Without it, Intel has explicitly stated that we would be outside of the functional limits. Realistically, like I said earlier P=IV except here power can't be constant because the signal quality degrades with the lower voltage so current must be increased. What I was alluding to before was saying that the power was increasing heavily, which was only half right.

I agree with the chart that power must increase as vcore decreases because current must increase heavily to keep signal quality, but the assumption made is that this is a chart for all speeds and for all voltages. The charts and the tables provided never talk about anything but for a given speed with a set VID.

Another way to look at this chart is to say with everything at stock, I am going to undervolt my CPU but keep signal quality. In order to do this, amperage and power would have to increase. However with just a reality check on this, most people know that most motherboards don't provide the increased current and thus power goes down along with signal quality. That is why there is a fine line for undervolting. On a high quality motherboard, you expect to get the VID pretty close on so you can lower the vcore to the Vcc minimum and remain mostly stable at the same current.

I agree with the charts, but they honestly don't say anything about overclocking or using vcore above VID. One way to immediately see this is to just extend the charge to the left. It would then claim that you must use negative amperage which just isn't feasible.

Quote:

Originally Posted by AyeYo If it still isn't obvious for you, maybe the VTT chart will clarify things. VTT can actually handle a little MORE voltage at very low amps, so it makes it abundantly obvious that safe voltage is going DOWN with increasing amperage. .

Actually, think about why the charts are made this way. Both show the same trend: as voltage decreases amperage increases to compensate. Vcore is inherently undershoot almost always due to vdroop and vdrop. Intel doesn't support LLC or Vdroop control so to them a vcore overshoot shouldn't exist and thus the chart starts at a perfect VID-0.000. VTT is less controlled and overshoots happen quite frequently. Hence, same trend with different starting points.

Source Pg. 23-27

 

[AyeYo]

Quote:

Originally Posted by ChickenInferno Actually, think about why the charts are made this way. Both show the same trend: as voltage decreases amperage increases to compensate.

You still aren't seeing it and you're over looking the blatantly obvious and completely unambiguous statements relating to the chart.

The chart is NOT showing that amperage increases as voltage decreases. It is showing that as amperage increases, the limit of safe voltage decreases in order to keep the CPU within safe limits.

PLEASE, read this extremely clear statement that cannot possibly be taken in any way other than as intended:

Quote:

Refer to Table 2-8 and Figure 2-3 for the minimum, typical, and maximum VCC allowed for a given current. The processor should not be subjected to any VCC and ICC combination wherein VCC exceeds VCC_MAX for a given current

The chart is CLEARLY all about finding a safe VCC voltage FOR A GIVEN CURRENT. It is NOT stating that a certain amperage is required for each voltage below VID, that makes absolutely no sense at all. You're reading the chart backwards and completely missing its glaringly obvious point because you're trying to interpret it to fit what you've determined to be "correct".

Amperage is on the X-axis for a reason. Safe voltage (Y-axis) is a function of amperage (X-axis), not the other way around. This is 7th grade algebra. This is the same reason that amperage, and NOT voltage, is in the first column on the chart, because the safe voltage is a function of amperage, not the other way around.

 

[*AcidBath*]

Quote:

Originally Posted by AyeYo You still aren't seeing it and you're over looking the blatantly obvious and completely unambiguous statements relating to the chart. The chart is NOT showing that amperage increases as voltage decreases. It is showing that as amperage increases, the limit of safe voltage decreases in order to keep the CPU within safe limits. PLEASE, read this extremely clear statement that cannot possibly be taken in any way other than as intended: The chart is CLEARLY all about finding a safe VCC voltage FOR A GIVEN CURRENT. It is NOT stating that a certain amperage is required for each voltage below VID, that makes absolutely no sense at all. You're reading the chart backwards and completely missing its glaringly obvious point because you're trying to interpret it to fit what you've determined to be "correct". Amperage is on the X-axis for a reason. Safe voltage (Y-axis) is a function of amperage (X-axis), not the other way around. This is 7th grade algebra. This is the same reason that amperage, and NOT voltage, is in the first column on the chart, because the safe voltage is a function of amperage, not the other way around.

To say it again, this plot shows: for a given amperage, the commanded VID voltage (whatever it happens to be) should be reduced by the given amount and the voltage filtered such that when the amperage changes, voltage spikes should be bounded by the min/max voltages given at the new amperage. Nothing more needs to be said or inferred by this plot. And please note: there is nothing on this plot/chart related to a specific VID value (which is assumed to be within spec) because it could be all over the place (well, sort of), depending on the operational state of the processor. If you put this all together, this the dreaded vdroop spec! From this info you can see if you use LLC, you are removing an important safety net. But heck, who around here wants a safety net anyways...

 

[eddieck]

If you keep under VID and keep LLC disabled (so VDroop is enabled), are you clear?

 

[ChickenInferno]

Quote:

Originally Posted by AyeYo You still aren't seeing it and you're over looking the blatantly obvious and completely unambiguous statements relating to the chart. The chart is NOT showing that amperage increases as voltage decreases. It is showing that as amperage increases, the limit of safe voltage decreases in order to keep the CPU within safe limits.

I'll make this easier for you then. As x decreases, y increases. As voltage decreases, amperage increases. A motherboard gives a certain voltage when trying to achieve the VID almost all are exclusively lower with things like LLC off. So then at stock settings with a motherboard that does VID-0.0660v after vdrop for Vcc-Typical, you will know that your CPU is pulling aproximately 11amps to run within limits. This says nothing about overclocking and nothing about anything above stock VID. The chart doesn't even extend past VID.

Read the statements very carefully and always assume the least out of them. If you try reading into them, then ussually you find out it doesn't mean what you thought it did. All of Intel's docs and whitesheets are this way.

Quote:

Refer to Table 2-8 and Figure 2-3 for the minimum, typical, and maximum VCC allowed for a given current. The processor should not be subjected to any VCC and ICC combination wherein VCC exceeds VCC_MAX for a given current

For a given current, do not exceed Vcc_Max listed here. All other values are not mentioned. Nothing is said about values above VID. This is an electrical specification for what the socket will provide based upon these conditions. It is for telling the motherboard "given this vcore, give between this and this amperage only" This statement says nothing about saftey, overclocking, anything over VID, and other speeds.

Now let's read into your statement to the next level and see inherent logic flaws. Assuming (incorrectly) that anything above stock speed violates the amperage condition due to an increase in vcore, then what is an i7 950 in comparrision to an i7 920? They are engineered identically and are "the same under the hood" but with different multipliers. They also sometimes have the same or differing VIDs. For your statements to be correct, anything above stock VID would be out of spec and killing the CPU, but wait how do they get away with increasing the speed and voltage and calling it VID at the higher speed?

Quote:

Originally Posted by AyeYo the limit of safe voltage decreases in order to keep the CPU within safe limits.

There are two ways to examine any 2-D chart. One is to approach from the viewpoint of x, the other starts from y. The chart shows for as voltage decreases amperage increases. The other way to view it is that as amperage increases voltage decreases. You chose to look at it from amperage, and then read into it assuming a few things.

Things this chart does not say and is not implicitly stated anywhere.
a.) That this chart is common for all speeds or if it changes at higher speeds (as it must or else a 950 would be out of spec)
b.) Anything above voltages or amperages above VID
c.) If the amperage/voltage combinations at the end are even within signal quality. VID-0.112 pulls 140A! I highly doubt any motherboard would put out 140A at a low voltage for stock speed let alone that this would be stable. It might not be out of amperage spec, but signal quality might be another story.

Quote:

Originally Posted by AyeYo The chart is CLEARLY all about finding a safe VCC voltage FOR A GIVEN CURRENT. It is NOT stating that a certain amperage is required for each voltage below VID, that makes absolutely no sense at all. You're reading the chart backwards and completely missing its glaringly obvious point because you're trying to interpret it to fit what you've determined to be "correct".

See there you're at it again. It doesn't make sense to you because you're not reading it. For a given speed a specific voltage is needed in order to make the transistor gates run. Enough voltage and it's a 1. No voltage and it's a 0. Insufficent voltage (aka instability) 1 or 0. Motherboards ussually give lower than the specified voltage hence a possible instability. However, insufficent voltage+a slight amperage boost=enough voltage and stability. Look at the chart again, it shows that for motherboards of varying quality aka For VID minus some number, the amperage limits in order to run stable.

Quote:

Originally Posted by AyeYo Amperage is on the X-axis for a reason. Safe voltage (Y-axis) is a function of amperage (X-axis), not the other way around. This is 7th grade algebra. This is the same reason that amperage, and NOT voltage, is in the first column on the chart, because the safe voltage is a function of amperage, not the other way around.

Okay. Go back to college and get your money back. College Algebra based Intro Physics will teach you P=IV which I have quoted over and over again. The great thing about equations is that you can manipulate them and they still mean the same thing. I=P/V (This says voltage is a function of current). V=P/I (This says current is a function of voltage). Both statements are true and you arguing that they're not is asinine. Left vs. Right on a chart is x then y. Plot something in excel, sigmaplot, or origin and you'll quickly find out that when making any plot you almost always put X values in col.1 and y values in col.2. That was a ridiculious point and you get a ridiculiously easy answer to it. You can't read into these things and honestly you've done alot of that.

Quote:

Originally Posted by AyeYo You're reading the chart backwards and completely missing its glaringly obvious point because you're trying to interpret it to fit what you've determined to be "correct".

I wanted to come back to this point for the exact reason you called me out. You're trying to interpret a chart with a single sentance explaination and say alot more about things it never mentions. You wanted this to say something it doesn't openly state and so you've interpretted it and assumed things to make it seem like it's correct.

 

[AyeYo]

Quote:

Originally Posted by ChickenInferno *wall of text*

Just do me a favor and, instead of dropping a writing bomb, just address this very simple, very clear statement for me:

Quote:

Refer to Table 2-8 and Figure 2-3 for the minimum, typical, and maximum VCC allowed for a given current. The processor should not be subjected to any VCC and ICC combination wherein VCC exceeds VCC_MAX for a given current

Without sounding like an *******, you're just flat out wrong and anyone with an understanding of 3rd grade math and 1st grade reading comprehension can see it.

Before you reply, please review this quick tutorial on independent and dependent variables so you can understand why charts like this one can only be read one way: http://www.idea.org/page71.html

Another, completely unambiguous statement:

Quote:

Refer to Table 2-8 and Figure 2-3 for the minimum, typical, and maximum VCC allowed for a given current. The processor should not be subjected to any VCC and ICC combination wherein VCC exceeds VCC_MAX for a given current.

The fact of the matter is that the chart applies at ANY clock frequency. Current draw is current draw, it doesn't matter if you're drawing 100amps at 2.66ghz at full load or 100amps at 4.5ghz at 50% load, either way you're drawing 100amps and safe VCC voltage is decreased by a certain amount noted in the table.

Quote:

The chart doesn't even extend past VID.

The chart STARTS at VID. The top of the Y-axis is VID - 0.000, i.e. VID.

The CPU can only tolerate maximum VID at 0 amps, that's the whole point of the chart. When you start piling on the current draw, you start losing safe voltage margin.

Quote:

Look at the chart again, it shows that for motherboards of varying quality aka For VID minus some number, the amperage limits in order to run stable.

Honestly, man, what are you smoking? Quote me something in the datasheet that says anything coming within 19 ballparks of what you just claimed.

If that was the case, then the statement I keep quoting would read something more like: "Table 2.3 shows amperage necessary for a given voltage below VID. The processor should not be subjected to any VCC and ICC combination wherein ICC is lower than VCC minimum"

But it doesn't, no, it doesn't even say anything remotely close to that. In fact, it says the exact opposite.

 

[radaja]

wow i never thought i would see a voltage fight at OCN