Stressing Voltage regulators with HLT command

[gustavo]

In this thread some time ago I was asking about possible side effects of enabling HLT command detect on AMD platforms, one point that came to light was the stress imposed to the voltage regulators, it was commented by Peter....



<< the transition from and to power-saving Halt mode is the hardest test for the CPU voltage regulators. Actually, voltage regulator compliance testing is done by running a worst-case power consumption software while periodically HALTing the processor while watching by how far the voltage moves. >>



I am concerned to where do the Vcore regulators reside (at the cpu itself or at the maiboard or somewhere else), if they were to fail what would be the hardware piece that gets broken ?

Could somebody help me ?

Thanks Gustavo.-

 

[rimshaker]

I wouldn't worry about it. Given enough time, yes, it will probably fail or something.... just like for ANY electronics. But you have to look at it from a time perspective. By the time the regulator starts having problems due to prolonged use of the HLT feature, that's probly around a decade from now at least. Will you still be worried about that same m/b even 3-5 years from now?

 

[CTho9305]

the dont usually fail. a failure in the way meant by the article would most likely be that they allow the voltage to drop too far, and your system crashes. anyway, i've heard of only one "permament" failure, where all 11 capcitors around the CPU on a motherboard exploded :Q. no clue what caused it.

 

[gustavo]

The documentation for the Athlon XP says normal operating Vcore is 1.75V and ever above 30 Ampers while on low power states Vcore is 1.30V with just 1.54 Ampers !!

So I think the stress is really high and want to know were are the voltage regulators located. Does somebody know if they are at the mainboard or at the cpu ? This question is still ananswered.

Regarding that the problem would be the voltage to drop too far, what about severe oscillations ? Could they produce damage to the cpu ?

One further question would be why Intel platforms enable HLT command detect while AMD platforms dont.

Gustavo.-

 

[CTho9305]

<< So I think the stress is really high and want to know were are the voltage regulators located. Does somebody know if they are at the mainboard or at the cpu ? This question is still ananswered.

Regarding that the problem would be the voltage to drop too far, what about severe oscillations ? Could they produce damage to the cpu ?

One further question would be why Intel platforms enable HLT command detect while AMD platforms dont. >>



the voltage regulators are on the motherboard. usually the capacitors around the CPU, and probably some of the transistors. maybe a few more components. I have NEVER heard of damage to a CPU except in the case of the exploding motherboard, which is clearly rare.

by disabling HLT detect, you can get by with a cheaper voltage regulator that doesn't increase output fast enough to handle the sudden change. a voltage drop will not harm anything except your data (currently running programs, unsaved work, etc)

 

[pm]

First the voltage regulator is on the motherboard, near the CPU. There are several active components as well as passive components that make up the local CPU power supply system.

The problem with changing the power load more quickly that the regulator can handle isn't merely droop but the typical accompanying ringing/overshoot. If a power regulator is likely to droop from load changes, it's also probably going to ring - which is worse and can lead to damage. So it won't just move down, it also could/may move up.

 

[gustavo]

Thanks all you, I can?t help however making here one more question, enabling HLT command detect makes my 12Volts line go down to 11.80 at idle while normally it stays fluctuating inside the 12.4 - 12.7 Volts range. So enabling this bit makes my 12V line fluctuate up un down in a noisy wider and cripply way, you think this could be a likely source of hardware damage ?

pm: Why Bioses for Intel platforms enable the bit ? The motherboard manufacturers are the same, I mean the MSI mobos for Intel processors enable the bit, but the AMD ones dont. The voltage regulators quality should be roughly the same (same manufacturer) so the AMD processors are more demanding on the regulators ?

 

[pm]

gustavo, for that you need to talk to AMD - and I remember your post about the problems that you had with this - but only they know the answer. As far as Intel, HLT capability is an architectural feature of the chip. It is enabled because it's a feature that Intel's customers want. As far as why AMD disables it, I could guess, but I don't know.

 

[gustavo]

Ok pm, I guessed it was to be your answer, sorry I wrote that late yesterday night.

Let me please reformulate: we are talking about the quality of the voltage regulators required to deal with abrupt changes in power and voltage, maybe you know if the difference in price would be so high to build motherboards with low quality regulators.

Also I guess you could know the difference in voltage and amperage for intel processors between normal states and low power states, or at least where can I get that info, so with that data it would be my conclusion if the regulation stress is or not higher with AMD processors that with Intel ones.

Finally my collected data say that the Vcore does not oscillate more with HLT detect enabled than disabled but the 12Volts line does maybe you or some other expert here knows if any relationship is to be expected.

Thanks you, and apologize if I am bothering, Gustavo.-

 

[highwire]

gustovo -

Re: the 12 volt going lower under halt conditions.
Often, the main regulation of the power supply is to the 5 volt output and other voltages are basically ratio'd off that regulation. If the 5 volt demand drops due to the HALT, the PS reg will run lighter, likely causing the other voltages i.e. 12 V ( with constant loads ) to drop somewhat. One PS might be much better than another in this area.

Regarding hardware damage and reiterating what has been said, the likely result from "stress" of the CPU voltage source will be a system crash rather than component damage, though there is at least a small chance of that happening from the short transient voltage overshoots to the CPU.

This highly variable current demand from the CPU makes for more difficult electrical requirements. Who knows, maybe Intel addressed the problem by softening this demand change through some propietary magic, and AMD did not. A most elegant method would be to have the CPU itself monitor its core voltage and do "busy work" as necessary to control transient voltage excursions of the core.

Anand just made mention that the new AMD "Hammer" processors were not festooned with capacitors. Maybe, everyone is on to this now and is using the processor itself to cope with this power problem.

 

[gustavo]

Thanks Highwire, sounds reasonable, the strange thing is that the 5V line stays allmost put !!

Anyway regarding the severe oscilattions on the 12V line (11.8 to 12.7) you think some piece of hardware that feeds from it is at risk ?

And I understand from your comments that the stress is important to any voltage regulators, that is not a matter of using higher or lower quality ones but of smoothing the abrupt changes before they arrive the regulators of the mainboard ?

What a forum !!! What good people are all you, knowing a lot and sharing the knowledge.
Very pleased of having found you, thanks all.
Gustavo.-

 

[Peter]

Hey, what I said back then was about design verification, not longevity testing.

CPU voltage regulation quality does degrade over time because of capacitor aging. That problem never surfaced
ever again after the early Pentium days, when even ASUS had inadequate component quality in that area.
I recall having to scrap dozens of TP54 boards for exactly that reason.

gustavo, one last time: On AMD Athlon mainboards, the Disconnect-on-HLT feature often remains disabled because
the Thunderbird core has a couple of Disconnect related errata. Intel P6 architecture doesn't even have Disconnect,
so you can't say "that bit gets enabled" ... the CPU interface programmabilities of P6 and K7 chipsets are not comparable
at all.

regards, Peter

 

[gustavo]

Peter !!

I saw a couple of threads where people was missing you and reading your posts I can guess why people miss you, you are of really good help however I couldnt rate your expertise/assistance as I could with the others, you disabled that feature so I thank you by writing "Thanks Peter".

Could you provide me with any link regarding as how the way Intel processors handle disconect, I wanna know if there is or not a comparable stress to voltage regulators on their platforms.

Thanks again.
Gustavo.-

 

[Peter]

I said that already, Intel P6 architecture has no such feature as Disconnect. They internally snooze on HLT,
but the bus interface stays fully active. Voltage regulation testing however is done the same way ... the
variation in amperage just isn't as high because Intel P6 processors don't sleep as deeply as Athlons do,
draw less power while awake, and also do not reduce their core voltage level while sleeping.

regards, Peter

 

[gustavo]

Thanks Peter,

This is in itself a highly likely explanation/reason to disable HLT detect on AMD platforms aside of the one regarding the errata you gave me/us reference before.

I am no questioning the first explanation you gave but kept thinking of this other and it seems at least important to take into account (I am not an electronic expert however like all you are).

Thanks again
Gustavo.-

 

[Peter]

Yes, getting the CPU voltage regulation right for an Athlon is quite a bit harder than for a Pentium-III ...
but the same is true for Pentium 4. So if you're a mainboard designer and don't manage that you better
go back to shepherding or dishwashing

btw, for the conspiracy theorists: that cheating has actually happened before ...
back in the days when the first Cyrix 6x86 popped up which had a comparatively huge power draw
over the early Pentiums. Many mainboard makers didn't strengthen their voltage regulation but
rather have their BIOS engineers turn off Suspend-on-Halt in the Cyrix to avoid the large jumps in current
draw ... that sound familiar?

regards, Peter

 

[gustavo]

Peter:

You mean Pentium 4 do have a disconnect on Halt feature comparable with that the AMD´cpus ?

Have any link to see if there exists any voltage variation when Pentium 4 are disconnected than when full awake ?

Thanks Gustavo.-

 

[gustavo]

What is the mystery here ? Is that Intel CPU´s Documentation is not as freely available to the public as AMD´s is at their web page ?

PM: you work or worked for Intel don´t you ? Where could someone find that documentation supporting what Peter says about the disconnect protocol into the Pentium 4 architecture ?

Precisely I would want to get info about the voltage and amperage consumption difference between a Pentium 4 operating normally and a Pentium 4 Halted (disconnected).

 

[Peter]

gustavo, please stop extrapolating from what I said ... all I said was that in terms of voltage regulation,
P4 is at least as hard to manage as Athlon. If you want to learn about P4 bus protocol, then I guess
you'll have to license the technology from Intel first. This is not an open book.

regards, Peter

 

[gustavo]

Peter: I found the datasheet for the Pentium 4 is here ... ntel® Pentium® 4 Processor in the 478-pin Package at 1.40 GHz, 1.50 GHz, 1.60 GHz, 1.70 GHz, 1.80 GHz, 1.90 GHz, and 2 GHz Datasheet

At page 20 table 6 you can see that at low power states the voltage does not vary, and in the case of a 2 GHz Pentium 4 the amperage drops from 57.4 Ampers to 16.7 Ampers.

It means the stress is by far less than in the AMD´s processors, and I don?t want to upset you, neither extrapolate anything.
On the contrary you are helping me and any other interested member to understand this issue and I appreciate very much your help, is just that I am trying to discard that the reason for the HLT disabling on AMD platforms is the excessive voltage regulation stress (and the hardware damage risks it would imply).

I would like to discard this and be sure the only reason was the errata you helped me to find because the errata would not imply any hardware damage risk.

So now with this info on hand would you keep sustaining that the errata is the motive of the HLT disabling and no hardware damage risk is implied ?

Thanks for your understanding Gustavo.-

 

[gustavo]

sorry, duplicated post cannot delete it ...

 

[Peter]

See, the P4 goes up to almost 60 Amperes, while Athlon XP maxes out at 40. OK, so P4 doesn't switch voltages
dynamically, but it draws 50 percent more juice.

To summarize, on a well designed mainboard I still wouldn't enable the Disconnect-on-Halt protocol for
T-Bird Athlons because of the latter's errata. On a not-so-well-designed mainboard, there might be electrical reasons.

regards, Peter

 

[gustavo]

Thanks Peter,

But my last doubt:

Would you call an MSI K7T266 Pro 2 (the red one, VIA KT266A chipset based) a well designed mainboard looking at it from the voltage regulation perspective ?

 

[JHutch]

Gustavo,

It doesn't matter if the MSI is "well-designed." Peter still wouldn't enable the HLT ("Disconnect on Halt") in the Athlon MB, simply because the Athlon itself has some "issues" with the HLT command. The voltage regulation really isn't that much of an issue here. A motherboard with good components would be fine. It is the CPU itself that has problems.

The P4 examples are just to show that the Intel piece also has wide power fluctuations and manages. The difference being that the P4 doesn't have documented "errata" on its HLT command.

JHutch

 

[gustavo]

JHutch:

I understood it clearly, Peter would not enable the HLT ("Disconnect on Halt"). He was clear on this.

My point is finding if any hardware damage or shortening of lifetime is to be expected due to the voltage regulation stress.

Many Tbird systems are running to my knowledge with the HLT enabled without the hanging reported by AMD, so I thought you experts on so highly technical issues would help me determine if there is any possibility that the real reason for HLT detect disabling is the voltage regulation issue and the errata just a fake ...

You get my point ?