X2 Bug Warning: Thermal throttling bug / power saving mode use could fry your CPU!

[NickK]

It'll be interesting to see which motherboards
(a) have enough capacitor capacity and for how long they can provide enough current,
(b) have capacitors with a good enough speed (from 0-full current) to prevent a core power brownout,
(c) how warm the tracks get from the current draw
(d) How much current is provided by the MOSFETs (ie are they relying on the powersaving mode and short peaks over their maximum power delivery to be covered by the caps).
(e) number of phases - most AMD motherboards are 3 phase IIRC. Pentiums needed 4 or 5 depending on their current draw.

The reason for Intel's Tom's Hardware Intel failures could be that the current draw from the P4 dual core is way beyond the specification of the motherboard.

I wonder if soldering on larger, faster reacting caps will help prevent brownouts.

[ewitte]

If it happens, you can RMA.

Hopefully after availability is up or you'll have to wait a month

[STEvil]

the "brownouts" would more likely be induced by the SMD capacitors and the ESR resistance of the main vcore regulators.

[synergy]

Hmm I took a quick look at a few models of MSI, DFI, ASUS Socket 939
motherboards and all the ones I saw pictures of seemed to have three
toroidal or other form of transformers near the CPU, and therefore it
looked to me like they were using 3-phase Vcore power generation.
I'm sure there COULD be a wide variety of quality and capacity between
them even so, and some of them MAY be entirely adequate, who knows.

Here's an article by AOPEN about INTEL CPUs and their claims of
benefits of using 4-phase Vcore power regulation over 3-phase
so I suppose the analogy would hold for X2 motherboards too:
http://global.aopen.com.tw/tech/tech...owerEngine.htm


Re AMD motherboards -- I did find, however, that --
http://www.giga-byte.com/Motherboard...Socket+939.htm
GIGABYTE has several motherboards which they
list as having:
"Dual Power System Gold (DPS Gold)
Specially designed for AthlonTM64 platform, the Dual Power System delivers a total of 6-phase power circuit to provide more solid and durable power supply for the new generation AMD platform. Ensuring a more secure and stable platform than ever before."

That seemed rather encouraging since it seemed like they paid special
attention to Vcore power quality on several of these models, and that they
used more Vcore voltage regulation phases than it SEEMED many other
Athlon64 motherboards use. They even seem to have a cooled gold colored
VRM heatsink block on the "DPS Dual Power System" boards I've seen
pictures of.

Unfortunately their "CPU SUPPORT" list on a couple of these DPS
motherboards doesn't list Athlon X2, so I guess maybe they're slow
updating their WWW site compatibility list, hopefully they really
DO or WILL have BIOS / motherboard support for the X2s.

I have no idea what the ACTUAL quality of these boards is without more
research but it's something to look into.

[STEvil]

DFI's seem to have a problem with dieing unexpectedly, but have regularly been above 3ghz and up to 4ghz..

I assume the PSU may have as much to play in this as the motherboard also..

[xgman]

I assume if you don't enable thermal throttling or anything related you won't have this problem? I never enable throttling or spectrum or cool and quiet anyway.

[synergy]

"I assume if you don't enable thermal throttling or anything related you won't have this problem? I never enable throttling or spectrum or cool and quiet anyway."

XGMAN, well, yes, if you really *CAN* disable STPCLK usage in all the various
ways the BIOS / chipset / ACPI stuff may trigger it for you then, yes, you shouldn't
have this as a potential crash / glitch problem.

Your only danger then might be thermal "runaway" overload if disabling STPCLK / C2
state also has the side-effect of removing that protection --
which it may or may not; I don't have the full details on how the CPUs internal
protections work, or if the BIOS may have other recourses
to do high CPU temperature safety intervention other than STPCLK.

While you can turn off ACPI, CnQ, and maybe turn off or set very high temperature
thresholds to the CPU overheat remediation functions in the BIOS, I don't know if there
are other aspects of functionality that the BIOS doesn't usually give you control over
that also use this signal. It's probably best to ask your BIOS / Motherboard vendor
what they did about this AMD errata issue and see whether they feel their
motherboard is effected by the problem in the first place (I guess most are, and even
more so with overclocking).

[p0tempkin]

If you disable the STPCLK function, just keep an eye on your temps. I loaded my X2 on my DFI LP UT with throttling disabled, 2 instances of Prime95 Torture Test Small FFTs, and it maxed out around 65C. During everyday usage, the cores never get that high, so I'm not too worried. If something does happen, I'll use my spare X2 while the RMA goes through.

[synergy]

p0tempkin that's interesting --

Which X2 model did you use?

What actual clock speed were you running that 65C temperature dual Prime
test at (overclocked?)?

What was your cooling setup? AMD default CPU box included heatsink or
something different / better?

Thanks!

[STEvil]

and did your heatsink even feel like it was close to 50+c (ie: burning to the touch within 10 seconds).

[Tony]

Most of the new high end enthusiast boards should be able to cope, i know voltage regulation is huge on the DFI and the upcoming grouper design so we should be ok.

[joe2004]

I loaded my X2 on my DFI LP UT with throttling disabled, 2 instances of Prime95 Torture Test Small FFTs, and it maxed out around 65C.

Man that is way too high. What X2 is this in particular and what king of cooling you have?

[STEvil]

i would assume he's either overclocked pretty good (THG had load temps of about 55c on stock cooling, but its THG... so...) or his mobo sensor is a POS like 90% of them out there.

Heck, my winny says 60+c sometimes when at 2600mhz 1.63v loaded and a zalman7700 on top.. the base of the heatsink never passes 40c.

[joe2004]

Well, if you overclock X2-4200 to 2.6 GHz with 1.5V on the processor this CPU will likely generate 150+W. This is equivalent of running Venice 3000 at 2.8 GHz with 1.67V. There is not many heatsinks that can handle that. And you can forget about anything more than 1.5-1.55V on air cooling with X2 since the thermal design is already 33W above Venice to start with.

However 65C should not be allowed else you are looking at damages to your processor in a long run.

[joe2004]

What I can actually see, based on the current thermal design of processors the king of overclocking will remain - FX-55, FX-57 and San Diego alternatives, unless we get Vapochill "super" LS to handle 250-300W dissipation.
Also I think Yonah or desktop derivatives of Dothan will be big AMD killers next year if one can believe the current overclocking/thermal fable that is going on around here.

[Aries64]

'Also effects single core A64 "DH-E6" too.' -- It effects
BH-E4 (only X2's)
and JH-E6 (X2's and Dual Core Opterons).

"wonder if this could be related to p4 dualcores cooking mobos due to power transients when going into thermal protection modes (its the motherboards at fault really since power transients are caused by the vcore vregs not being able to cope with the power requirements of the CPU).
Besides, if going from full load to all stop is that dangerous, then so is going from all stop to full load." --

I agree, it seems likely.
Personally I am very disappointed at the penny-pinching,
quality-corner-cutting philosophy of consumer electronics design including
motherboards. As a PCB designer I've always been inclined to select
PCBs with full ground plane & power planes for each major voltage when
designing boards a lot simpler and lower speed than 2GHz + PC motherboards!
It really boggles my mind that they'd have such a push to try to get PCBs
down to 4 layers for Socket 939 motherboards with PCI, AGP, DDR,
etc. etc. that's all quite high speed and high pin count.
I very much suspect that it's more of a matter of *luck* than *engineering*
that they work as well as they do. Even though STPCLK / STPGNT type
power throttling is an easy to generate extreme test case of power load
fluctuation from ~ Zero power to full load, I suspect that many software
situations involving CPU / I/O bursts do about as much variation of loading
on the time scale of nanoseconds to microseconds. If the EMI design and
power filtering, regulation, capacitors, PSU, etc. isn't up to the task of
providing clean transitions from STPCLK to Full Run I suspect it's not REALLY
stable running common software programs under stress / bursty conditions
either, and that's probably a major reason for things crashing / flaking /
burning out on many systems.

What's needed is just better quality motherboards, 6+ layer PCBs, more
conservative design margins of capacitors / power regulators, thicker PCB
power traces, etc.

I agree as with what you have said as well, synergy. However, we can't place all the blame on the mobo manufacturers' for cost-cutting. In my opinion, most of the blame should be placed on the consumer.

Why? Because in my experience the public at large is a bunch of cheap bas$%ds. People are ALWAYS complaining about how much something costs. They want the best quality and best best performance but they don't want to pay for it. Need proof? Overclocking. Long-time AMD users are probably also at fault too.

Now you can't blame anyone for wanting to get the best price/performance ratio (who doesn't?). No one wants to pay more than they have to, or more than they feel is fair. But having worked in the high-performance aftermarket auto industry for many years (anyone ever heard of Neuspeed?), then PC and Mac computer hardware, and now having years of experience as a Realtor in California I have dealt with literally thousands of people. Some don't mind paying the premium that quality usually demands, but most will attempt to grind you down.

Rant over.

[xgman]

What I can actually see, based on the current thermal design of processors the king of overclocking will remain - FX-55, FX-57 and San Diego alternatives, unless we get Vapochill "super" LS to handle 250-300W dissipation.
Also I think Yonah or desktop derivatives of Dothan will be big AMD killers next year if one can believe the current overclocking/thermal fable that is going on around here.

I'm not so sure. The 4800+ I have cooled on water with IHS removed at 1.6v is still only in the high 30's temp wise at 2900mhz.

[STEvil]

Neuspeed.. I believe I remember a decal for that in NFS:U..

I dont think you'd see 200w until about 3500mhz @ 1.7v+.. A64's are very cool comparred to AXP/Barton.. cant imagine what its like going from a P4 to A64 temp wise

[computerpro3]

Neuspeed.. I believe I remember a decal for that in NFS:U..

I dont think you'd see 200w until about 3500mhz @ 1.7v+.. A64's are very cool comparred to AXP/Barton.. cant imagine what its like going from a P4 to A64 temp wise

Wouldn't be so sure, I overloaded my stock Mach1 with my 3500 claw @ 3ghz 1.8v. Had to get it chilly1 modded, which of course, I think got damaged in transit....

[synergy]

They (X2s) do around 110 Watts just at stock settings.

Power increases as the square of *both* Frequency AND Voltage,
so if the product of the factor of VOLTS INCREASE and
FREQUENCY INCREASE as ratios of their stock settings was the
square root of 2 (1.41), that'd GUARANTEE you'd be doing 220 Watts
average and a fair bit more on peak.

So in summary a 20% overclock and a 20% overvolt together is guaranteed
to exceed 200W on an X2 that does around 100W stock.

In reality when you're reaching the maximum ratings of a chip the
"point of diminishing returns" comes on even more quickly than that
theoretical consideration which applies to operating within stock specs.

As the IC heats up the timings will of course get worse, the resistance of
all the metal / n-type circuitry increases, leakage current from
the overvolting starts exponentially increasing, and things just rapidly go to hell.

To quote AnandTech on OCing a X2 4200 we can see the last little
bit of overvolting / overclocking resulted in very dramatic temperature
rises:
http://www.anandtech.com/cpuchipsets...spx?i=2452&p=7

"You can likely do even better than we have if you use more exotic cooling. We have seen many reports on the web of the 4200+ reaching 2.8GHz or even 3.0GHz. We have also seen reports of the 4800+ reaching even a bit higher, so even better overclocks may be available with a 4800+. Anand is reaching 2.8GHz on air with the 4800+ he has been testing.

In the end the 4200+ appears to be a good-performing dual-core CPU with quite a bit of overclocking headroom. We reached 2.7GHz with a PC Health reported CPU temperature of 61 degrees C at 1.55V. 240 was a breeze at 1.45V, exceptionally stable for days on end, with processor temps generally below 50C with our air cooling. It appears an easy task to reach the highest levels of Dual-Core performance with the cheapest 4200+ if you are willing to overclock a little - and the 4200+ is up to the task."

[ben805]

200w ain't too bad....100w per core

[bugeyes]

The 4800+ I have cooled on water with IHS removed at 1.6v is still only in the high 30's temp wise at 2900mhz.

did you take any photos of the naked core, and what tempreture difference can be obtained by non IHS use?

[Otter]

They (X2s) do around 110 Watts just at stock settings.

Fortunately, the X2 doesn't actually use that much power at the defaults. 110W is just AMD's thermal design power. Lost Circuits measured the power consumed by a 4800+ (2.4GHz Toledo, 1MB cache/core) with both cores fully loaded at 79 to 86 watts, depending on CPU temperature.
http://www.lostcircuits.com/cpu/amd_x2/12.shtml

Now, an overclocked X2 will still be a power-hungry beast. But OCing it to 2.9GHZ at 1.6v shouldn't generate 200W of heat.

[Gogar]

They (X2s) do around 110 Watts just at stock settings.

Power increases as the square of *both* Frequency AND Voltage

I always thought it increased linearly with the frequency and linearly with the Voltage, but since you also need a higher Voltage to get a higher frequency the power increase is square.

Though it does seem that you need more and more power to get it stable at very high clocks, my guess is that you're just fighting high temperatures (= high resistance) at that point.

IMO the power draw increase should float somewhere between to the second and third power depending on temps.

[synergy]

Just a quick correction to what I said before when I was tired. I said the
dynamic power consumption of an IC increased as the square of the frequency.

As Gogar pointed out, dynamic power actually has a linear relationship with frequency
though effects like "the skin effect" (increasing "resistance" of conductors with frequency), and the increasing loss factor of dielectrics, and other such things will in
practice cause a greater than linear power dissipation with increased frequency, but
I shouldn't have generalized it as being an exact square since that depends on the
materials and circuit involved.