Print 60 comment(s) - last by wetwareinterfa.. on Dec 14 at 12:46 AM

Two whitepapers published by AMD raise questions about its new thermal rating; an inconsistency AMD justifies with margin calculations

AMD's Average CPU Power rating scale underwent its first test of scrutiny yesterday when two nearly-identical whitepapers discussing ACP caught the attention of DailyTech and other technology forums. 

The inconsistencies between these two whitepapers included a table where the Thermal Design Power of AMD's new quad-core Opterons increased, without an increase in the Average CPU Power rating.

Several processor architectures ago, AMD and rival Intel used the same methods for calculating Thermal Design Power with regard to microprocessors.   From an engineering standpoint, the TDP represents the amount of power the cooling mechanism for the CPU must dissipate before failure.

AMD and Intel now differ with TDP calculations, and for different reasons.  Intel's current architecture, for example, allows the CPU to exceed the TDP rating for a small period of time before the processor throttles its frequency clock in order to reduce the temperature at the processor level.  AMD's current-generation processors do not practice this method, and thus AMD intentionally publishes conservative TDP ratings.

AMD's 2008 Phenom roadmap clearly illustrates the increase in TDP for upcoming K10 processors. Even though Phenom processors are desktop units, equivalently clocked Barcelona Opteron processors share almost all the same attributes and specifications.

AMD's Brent Kerby, author of both whitepapers, explains the inconsistency with timing and the nature of ACP itself.  "The measured value of ACP already included the changed TDP values," he explains. 

So are the ACP measurements represented in both documents wrong, or mutually exclusive? "No," says Kerby.  Even though the maximum thermal envelope has increased by as much as 21% between the two whitepapers, the algorithm for calculating the ACP is not entirely affected by this upperbound.

"When we published the first whitepaper, we had to anticipate TDP changes," adds Kerby.  The newest whitepaper, Kerby states, is relevant to both published TDPs.  The TDP references in the first document can be replaced with the TDP changes from the second, and in fact should have been.

Kerby could not comment on the specific margins built into the ACP rating, but mentions it's an integral part of the ACP process.  His whitepaper details the margin as follows:
The results across the suite of workloads are used to derive the ACP number. The ACP value for each processor power band is representative of the geometric mean for the entire suite of benchmark applications plus a margin based on AMD historical manufacturing experience.
However, Kerby could confirm that if the TDP were to increase again, the company's ACP values would need to be recalculated. 

Will AMD's thermal envelope increase again?  AMD won't say, but at least for now AMD's ACP/TDP mystery appears solved.

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

Change title to "WE LIED ABOUT AMD"
By TheJian on 12/10/2007 10:55:47 PM , Rating: 5
With Intel doing the same thing for ages this story shouldn't even have been printed and your LIE shouldn't have been told. Talk about irresponsible reporting. As long as we know BOTH TDP and ACP numbers who cares? Intel has been lying about theirs with avg numbers for years (I'm reminded of the stink of Dell Optiplex SFF machines with P4's above 2.4ghz...LOL. Presshots anyone?). Why is it a story when AMD does the same?
Read "The nuts and bolts of the quad-core Opterons" section.

As you can see they sound like they're saying it's about time AMD did this. They even say they like it but don't care because they'd test the chips themselves any way. Which is exactly what all review sites should do. They call it "justifiable". "This move may be controversial, but personally, I think it's probably justifiable given the power draw profiles we've seen from Opterons." Is there a problem with them wanting to be compared accurately to their rival? Most websites totally leave out the differences between Intel's TDP version and AMD's. What you get is AMD's are higher and must suck. I'm not saying they're better than Intel (I just bought a core2 myself a few months back thanks to 1.8's running at 3ghz on air so easily - and mine runs cool even without my koolance...LOL) but review sites have been doing this for ages. I don't blame them for wanting to fix this problem especially with their profits tanked. They should have done this 4 years ago. Instead their stupid marketing dept decided it was best to tell people the max TDP of any cpu that they might not ever produce in a particular product family. When they said 89w for A64's a few revs back it was for the whole family. Don't tell me you think a X2 3800+ is the same as a X2 5400+. Yet all of these were quoted at 89w. Look at Anand's chart here:

All AM2 chips show 89w. I've always thought they were stupid for allowing this. Review sites didn't help by making people believe they put off 89w. Usually forgetting to say it's so far off its ridiculous. AMD shot themselves in the foot by starting this MAX crap NOT by fixing it now. Why is it a LIE to tell the truth about their cpus? It's quite clear they LIED before by saying they were all 89w when they were really more like 40-89w. My old 3800x2 was cold with retail heatsink at 2.4ghz (400mhz oc per core, no heat). Their marketing dept should be fired. YES (and Ruiz too...Bring back Jerry!). But AMD correcting their marketing mistakes should not be attacked. Your previous story was the equivalent of me saying I saw you murder someone and I get you convicted with no proof. The following day (after the damage is done for anyone who doesn't come back to read a correction story) DNA proves you innocent and you get off. Don't I owe someone an apology here? You called them outright liars in the title. You even stated in the article that it may be a mistake in their docs. Isn't liar a bit conclusive knowing you might be completely off base? Is tomorrows article going to be called "The LIES START WITH DAILYTECH!"??? BTW core2 rocks and AMD is getting KILLED. Fanboys from both sides can just jump off a cliff :) That doesn't change the fact that AMD is having their chops busted for nothing here.

RE: Change title to "WE LIED ABOUT AMD"
By TheJian on 12/11/07, Rating: -1
RE: Change title to "WE LIED ABOUT AMD"
By JumpingJack on 12/11/2007 12:46:19 AM , Rating: 2
You're ranting.... they did not lie .. AMD published two specs in conflict, and the explanation still does not make sense.

RE: Change title to "WE LIED ABOUT AMD"
By bangmal on 12/11/07, Rating: -1
RE: Change title to "WE LIED ABOUT AMD"
By JumpingJack on 12/11/2007 1:52:18 AM , Rating: 3
I didn't call him a liar... I said he was ranting. What I pointed out was DT's original article was not a lie (as he was calling DT liars) ... DT pointed to a discrepancy in AMD's APC whitepaper, one version paired up one set of numbers, another version paired up a second set. TDP went up but APC stayed the same... this appears weird.

The follow up article, this one, is less than satisfactory explanation from AMD -- it is like that L2 latency argument for Brisbane.

Either a) their original spec was wrong or b) the APC is not using loads representative of the worst case, as TDP should encompass the max (AMD's responder calls this the upper bound).

5 apps with a geometric mean is not sufficient to capture all the variation, so it appears they are wiggling up a higher thermal envelop but keeping their marketing ACP number the same.

Look, a population of applications will drive a variety of thermal profiles as well as process variation from processor or to processor -- condensing this down into a 5 app geometric mean (which is not an average as you would typically understand average) is not sufficient. And if AMD adjusts the upper bound for what ever reason, then it is necessary to re-evaulate the ACP... but re-evaluation is especially if they are adjusting the upper bound to accompany higher operating voltages and consequently currents (which is how they arrive at Pmax/TDP per their own argument) in order to expand the clocking potential.

In short, I am not buying AMD's explanation above.

RE: Change title to "WE LIED ABOUT AMD"
By Viditor on 12/11/2007 2:31:10 AM , Rating: 3
the APC is not using loads representative of the worst case

I think that's where you went says in the whitepaper that APC represents typical useage, not worst case...

RE: Change title to "WE LIED ABOUT AMD"
By Viditor on 12/11/2007 2:34:33 AM , Rating: 3
It states under Test Conditions:
" Given the goal of representing typical power usage in
real world conditions
, environmental test conditions were
chosen to refl ect that aspect (room temp of 70ºF, server’s
fan heat sink used, closed case, etc.) The power for the
cores, memory controller, and HyperTransport™ links was
logged multiple times per second throughout the entire
duration of the workload tested, and the time-averaged
power consumption for that workload was calculated."

RE: Change title to "WE LIED ABOUT AMD"
By JumpingJack on 12/11/2007 2:44:20 AM , Rating: 2
They actually only used 5 workloads, quoted from the ACP WP:

These workloads were Transaction
Processing Performance Council (TPC-C), SPECcpu2006,
SPECjbb2005, and STREAM. The geometric mean of
measurements, taken during these workloads, is the ACP.

From these 5 apps, the measure the power consumed for each one as you described to get a single number... they then calculate a geometric mean which will be less than (or equal to -- unlikely) the arithmetic mean.

By tmouse on 12/11/2007 8:07:43 AM , Rating: 1
Is the distribution normal? (I do not know) If it is not then a geometric mean is the proper one to use as the arithmetric mean is meanlingless outside of a normal distribution.

RE: Change title to "WE LIED ABOUT AMD"
By JumpingJack on 12/11/2007 2:48:55 AM , Rating: 3
No I am not wrong, if the upper bound -- dictated by worst case forced AMD to change this spec and APC did not change with it, then the apps used to stress the process in assimilating ACP are not representative the entire spread or distribution of available workloads.

AMD, by matter of choice since this is their metric and not a standard, can choose what ever they like to create this value. So they chose 5 apps, all server related synthetics ... this is fine, but all things being equal, if the upper bound moved one would expect the geometric mean to shift unless they did not choose an appropriate subset of loads to capture the worst case.

RE: Change title to "WE LIED ABOUT AMD"
By Viditor on 12/11/2007 2:55:20 AM , Rating: 1
if the upper bound -- dictated by worst case forced AMD to change this spec and APC did not change with it, then the apps used to stress the process in assimilating ACP are not representative the entire spread or distribution of available workloads

You're comparing an upper bound theoretical limit (TDP) to a real world measurement (ACP). Movement of one does not necessitate movement of the other.
Again, you are assuming that ACP is meant to capture worst case instead of typical useage...

RE: Change title to "WE LIED ABOUT AMD"
By Viditor on 12/11/2007 3:00:33 AM , Rating: 2
Let me be clearer...

You're assuming that the theoretical geometric mean will be the same as the tested geometric mean...and that ain't always the case.

RE: Change title to "WE LIED ABOUT AMD"
By JumpingJack on 12/11/2007 3:29:29 AM , Rating: 3
Now this does not make any sense :) ... sorry, maybe you could be more clear :) ...

You seem fixated on the word theoretical... TDP is not theoretical, it is a spec indended to ensure that the cooling solution is sufficient, you choose it based on what you know or measure about the processor to ensure the processor will work.... AMD MEASURES to arrive at TDP, there is nothing theoretical about it. It's in their specs.

RE: Change title to "WE LIED ABOUT AMD"
By Viditor on 12/11/2007 4:57:37 AM , Rating: 2
AMD MEASURES to arrive at TDP, there is nothing theoretical about it

AMD doesn't run apps at that rate...I doubt there is any app short of a thermal virus that could do so.
As in most good theories, the premise is based on measured facts.
However, since the practice never actually takes place, it must remain theoretical.

Intel actually runs their apps and measures the outcome, so it is a measurement...

ACP is also actually run, so again it is a measurement...

RE: Change title to "WE LIED ABOUT AMD"
By JumpingJack on 12/11/2007 11:56:56 AM , Rating: 2
Well, I quoted AMD's spec... which states TDP is measured at these conditions... how they load it to arrive at that measure (prodding the CPU to max conditions) is ambiguous at best.

RE: Change title to "WE LIED ABOUT AMD"
By alanore on 12/11/2007 12:54:11 PM , Rating: 2
Since Rev E. AMD has placed the individual chips TDP on the chip, if you own a chip since Rev E. You can use TCaseMax to read your individual chips TDP, you'll find that its less that the TDP that your chips was designed for, and in some case well off it. ie if you have chips ranging from 2.0GHz to 2.8GHz all in the 89Watt envelop it is not very logical to think that the 2.8GHz chip will consume the same amount of power as the 2.0GHz chip.

All the TDP rating is the envelop in which the processor is. If you sample 50 phenom processors for the ACP, and then say, these all fall in the 95watt envelope but were going to scale up the clockspeed, knowing the upscaled clock speed will fall out of the envelop, so you then up the envelp to 120watts, the 50 processors you sampled aren't using any more power, its just you could make chips that do.

Second AMD is atleast calling theirs something else so it isn't confused with TDP, Intel just give its rating as the TDP even though its not. At this point it worth noting, because AMD chips have elements of the north bridge on them it increase there TDP but the systems TDP isn't increased.

The theoretical limit to the use of your brain is 100%, but the body just can't produce enough energy and oxygen to allow this to happen so it only operates at a fraction. Realistically the brain would only operate at higher percentages when something was hugely wrong. But say at 100% the energy it consumed was called the TDP, now if you listen to the TPD you would have to consume 8000 calories a day to provide it with energy. In reality we don't use all our brains so that intake of calories is overkill. Same with actual TPD the more core you have the further from the TDP you get, so there no point in the overkill.

By JumpingJack on 12/11/2007 10:57:43 PM , Rating: 2
I understand what you are saying, but this is off point.... you are talking about binning and where that falls on the power curve, this is different from measuring what the TDP should be to define the cut off for a processor operating at nominal within that bin.

By serajadeyn on 12/13/2007 2:51:10 PM , Rating: 2
Best Post in this thread, bar none.

Some of the marketing vs test result vs applied geometry calculations were muddying this issue far more than it was helping, and along comes your clear and easy to understand analogy. Thank you!

I think the point is that on the TDP vs ACP issue, there is no "best" specification over the other. They actually both serve to compliment one another. TDP expresses "this is how much thermal dissipation our chip can take before failure is expected" while ACP is "this is how much power the chip will dissipate as heat in the real world"

If taken this way, It would mean that AMD's processors are capable of higher thermal limits before failing; meaning it is more tolerant to thermal loads. I would regard this as a good thing personally.

All that being said, has anyone produced ACP figures on Intel's processors? It would be very interesting to see how they compare.

By mars777 on 12/12/2007 5:36:24 AM , Rating: 2
I tried to explain this in the article before:

If you were an avid overclocker you know that some CPU's were able to boot at higher clocks than others but could not achieve a so high STABLE frequency. So try to lower the clocks or pull up the voltage of these CPUs a bit and they work as expected. Their power consumption on 90% (ACP) load could very well be in the order of 1-3% similar as the normal version. But they would dissipate a lot more when charged at 100%+ (full 100% / not Intel) .

These are basically downclocked "overclocked" CPUs. Means that AMD has problems in production.

By JumpingJack on 12/11/2007 3:27:20 AM , Rating: 2
No, not really -- TDP is only meaningful if it defined or spec'ed to exceed the worst case or account for maximum thermal power to be dissipated. Why push it higher in the second revision of the whitepaper if ACP is truly representative of the population and has not changed?

Take it this way the geometric mean is calcualted by multiplying all 'n' values of the population and taking the n'th root i.e. geo mean = (X1*X2*X3...*Xn)^(1/n), the arithmetic mean is average = (X1+X2+X3...+Xn)/n. Both are ways to summarize a population of data, the geomean is always <= the arithmetic mean, however both respond to perurbations in the data. For example, take an arbitrary unit, call it boobles, say I sample one and get 2 boobles then I sampe again and get 3 boobles. The Geomean is (2*3)(1/2) = 2.44, the aveage is (2+3)/2=2.5 they do not mean the same thing, but lets say now instead of 3 boobles I do 4 boobles, so now my population is 2 and 4, here Geomean=2.8 and average is 3...

Now draw a bellshaped curve, and make it symmetric. Draw a line stright down from the top, this is the Arithmetic mean. To the far right, mark that as MAX, and somewhare to the left of the average put a line, this is the geometric mean. Here is a picture of what I mean:

Now, what AMD did was move the 'MAX' on the right up about 20%, but left the bellshape where it is... does this not beg the question -- why move the max up to begin with? Clearly ACP is not representing the entire body of possible power consumptions if this is true.

Also, as I explained above -- AMD's TDP is not a theoretical limit, it is measured ... and it is take from basically max supply voltage X max current (Iddmax), if anything the movement up in TDP by AMD would suggest they are trying to move up voltage and consequently current drawn in order to hit clock speed.... if this is true, then ACP should have changed.

Now, what AMD's response above suggests is that that 'margin' the quote in both papers is not consumed with this adjustment -- i.e. the revised WP is still wrong because there is really no margin left (what ever it was) when he states:
However, Kerby could confirm that if the TDP were to increase again, the company's ACP values would need to be recalculated.

By TheJian on 12/12/2007 4:55:28 AM , Rating: 2
"The TDP references in the first document can be replaced with the TDP changes from the second, and in fact should have been."

What part of 'I should have changed the doc' did you not understand?

I wouldn't have said much of anything if the article was titled something more along the lines of "Did AMD LIE?". Clearly they were mistaken and completely out of line. Stories like this just hurt a good company. I presented some facts from a pretty reputable hardware site and backed up my statements with other quotes. I don't call that ranting. Daily tech DID call AMD liars (conclusively!). No question about that and they hadn't even talked with them about it.

I hope you're going to be happy about next years cpu prices...ROFL. Better buy now, the cpus under $200 will probably be closed out by Intel in a year. Intel's top cpu is now around $1400 (desktops, they just stack on top now instead of dropping all and fitting top-end in at $995). When AMD was in the game in 2006 (early 06? :)) the top cpu was $995 from either side. I kind of like upgrading every 6 months. Stories like these will only tarnish AMD (unjustly with that title) even more than they already have themselves...I'm going to end up having to go back to the old schedule of every 2yrs here soon. I see the next good set of overclockers costing an extra $100 or so more.

We wouldn't even be talking about this if AMD had just put exactly what every CPU's TDP was years ago instead of naming a whole family X watts. They wouldn't have had to correct things today. They have been rating their chips FAR hotter than they were for ages. Look at the techreport article I posted previously...Even they think they are much cooler than rated (Kubicki quoted them so I used them too), so they didn't mind the new scheme. Did you even read my post or just attack me for pointing out DT's misrepresenting the facts? Apparently the guy should have put "not responsible for typos" at the bottom of the docs just like a fry's ad. :) Irresposible journalism at best to pump up hits or ???

By Screwballl on 12/11/2007 11:41:49 AM , Rating: 2
The TDP references in the first document can be replaced with the TDP changes from the second, and in fact should have been.

Since the information from #1 was faulty, it was updated and replaced with #2 so everyone should use #2... and since ACP already was based on #2, thats why it didn't change.

How hard is that to understand?

RE: Change title to "WE LIED ABOUT AMD"
By RogueLegend on 12/11/2007 1:34:59 AM , Rating: 2
You know, what ever happened to the fact that AMD was putting the memory controller on the die? Unless I've been asleep, Intel hasn't done this. Shouldn't this be taken into consideration? I know it doesn't change the way the cooling solutions are designed, but I would think we would need to start looking at system power consumption as a whole rather than just the TDP the processor alone.

And if Intel is allowing their processors to exceed TDP for short periods of time before the clock is throttled down- how is this affecting benchmarks, if at all?

Just questions, not attacking or defending anyone- I think these are valid questions. I own both Intel and AMD products.

By JumpingJack on 12/11/2007 2:10:34 AM , Rating: 3
The northbridge on or off the die is of no consequence, the thermal profile of what is in the socket is the key parametric as this is where the heat sink (and consequent design of the dissipation solution) sits.

On a platform level, if you want total power consumption for running your system, then the northbridge is important only if you are partitioning out the various components in order to assemble the system.

In other words you are arguing two different things, TDP is critical for system builders to know for designing the thermal solution ... process x needs a heat sink that can dissipate y Watts. If AMD has the IMC under the heat sink, makes no difference if Intel does or doesn't, the power of the chip must dissipate x Watts to stay within operating temperature.

The other parameter is how much will it cost to operate the thing... TDP is meaningless here, because TDP =/= consmption.

On you second question, Intel's specs are pretty clear -- Intel derives TDP based on worst-case, they choose worst case to avoid such scenarios, in otherwords, you will not see any affect on benchmarks because at full load using those applications they do not exceed worst case.

To get an idea of how Intel understands worst case, google an application called "Intel Thermal Analysis Tool". This is a leak application Intel supplied vendors with Pentium-m's and works with C2Ds, read up in the help file on their thermal profiling and the purpose, then (if you have a C2D) run it and meausre temps... then try to find any application that will run the processor harder, I doubt you will -- I haven't, Prime95, Everest, nothing stresses the CPU more than TAT.


Ohh, I also have both AMD and Intel :) ...

By nemrod on 12/11/2007 5:15:44 AM , Rating: 2
With Intel doing the same thing for ages this story shouldn't even have been printed and your LIE shouldn't have been told. Talk about irresponsible reporting. As long as we know BOTH TDP and ACP numbers who cares? Intel has been lying about theirs with avg numbers for years (I'm reminded of the stink of Dell Optiplex SFF machines with P4's above 2.4ghz...LOL. Presshots anyone?). Why is it a story when AMD does the same?

You believe that amd wouldn't have show that Intel lie if they could?


The natural question becomes “is this a new
industry standard measurement?” AMD believes the
methodology behind ACP should become the new
standard, as it includes both the processor cores and
the memory controller. Today, companies can mask the
true power consumption by fixating on the processor
itself and neglecting to mention the power consumption
of the external memory controller.

Enjoy agressive marketing:

But try to find where they said Intel TDP is a liar...

By Gul Westfale on 12/10/2007 8:37:21 PM , Rating: 5
why is something that should be so simple -telling people how much your CPU drinks, and thus its cooling requirements- so dufficult? of course, a regular consumer won't care; computer companies have warranties to cover failures, and boxed CPUs come with appropriate fans. enthusiasts always buy a fan that is massive overkill :), but the professional market will be confused by this.

if you need to buy several multi-socket servers, then how do you calculate how much power/cooling they need? and how do you compare that to similar machines from intel (and thus calculate possible power savings)?

it seems that by not telling people the actual, plain numbers AMD has shot itself in the foot again; and in this case that was not only avoidable but also completely pointless. why did they move away from their old TDP at all?

RE: why?
By Oregonian2 on 12/10/2007 8:49:29 PM , Rating: 5
Probably because providing "worst case" specs like most components have for most specs (timing specs for instance) provides numbers that aren't liked. Providing a number that is the highest power taken by the worst possible individual part under the worst possible operating conditions under the worst possible code stream (probably an invented not practically useful stream) is not something customers would like particularly when actual implementations would actually take a great deal less power under real conditions.

The problem is then how to characterize real conditions sufficiently to allow proper system design in the real world where squeezing watts and milli-cents is of utmost importance.

There then are different ideas on how to do it, and none are necessarily wrong or right, just different in intent and assumptions made.

Even if a single method of giving such information was decided upon, it's possible for it not to be a good thing because it then becomes artificial to the assumptions made in that method of providing a spec -- and those assumptions may not be appropriate for all users or over time (when the standard becomes fixed even as it becomes more useless). It's like providing a performance benchmark (actually that I think of it, its exactly that and everybody knows how accurate a measurement benchmarks are).

RE: why?
By feelingshorter on 12/10/2007 8:56:38 PM , Rating: 2
The general rule to marketing is that if you don't have anything good to say about it, don't say it. They are so behind when it comes to Intel that they have to do it like this. They are just trying to survive.

RE: why?
By Lifted on 12/10/2007 9:12:53 PM , Rating: 2
if you need to buy several multi-socket servers, then how do you calculate how much power/cooling they need? and how do you compare that to similar machines from intel (and thus calculate possible power savings)?

You would actually have to test each server you are interested in. If you are not in a position to have them sent to you for testing, you must rely on the specs given on the sites of HP, IBM, etc. HP had some spreadsheet you could download which let you configure a server and would calculate the power requirements.

I haven't been at it for a while, but APC had a tool on their website which calculated the power load for a rack based on the requirements of each server, switch, router, etc. It gave numbers pretty close to those available from HP, and included numbers from many other manufacturers that don't have the information as easily available as HP and the other Tier 1's.

RE: why?
By DallasTexas on 12/10/2007 10:05:50 PM , Rating: 4
In the past power dissipation algorithms were designed by engineers. Nowadays, the algorithms includes a marketing coefficient called "F prime" or Fudge factor.

The value of F' gets decided by a team of marketeers and then provided to an "engineer" to bury deep in the whitepaper. Not many buyers will argue with a "whitepaper"? Oooohh

RE: why?
By JumpingJack on 12/10/2007 10:43:29 PM , Rating: 2
This is funny.

RE: why?
By spluurfg on 12/11/2007 12:32:50 AM , Rating: 2
Something similar also happens regularly in finance, with regard to use of F-prime.

RE: why?
By caqde on 12/10/2007 11:03:12 PM , Rating: 2
Well the TDP at least for AMD is the max power usage although when you actually run a computer it is unlikely that you will actually reach this number. But there is another thing that effects newer processors like the phenom. While in use it is unlikely that all 2-4 (especially 4) cores will reach maximum usage all of the time sometimes they might never reach 100% for all four cores depending on the user and because the phenom can throttle the individual cores using Cool and Quiet even when a user is putting a large load on 1 or 2 processors the system will not reach it's TDP figure. Future processors will further skew these numbers with the newer power savings technologies that will be implemented in them.

RE: why?
By crystal clear on 12/11/2007 12:11:16 AM , Rating: 2
Just a point to note-

An Intel clarification-

Intel's comment: "The TDP is a specification that is primarily of interest to system builders (OEMs). It allows the OEM to design a system that can handle the heat generated by the processor based on this value. The unchanged TDP rating simplifies the process of designing and building new systems. To the end user, the value that is more interesting is power consumption on a system level, i.e., how much power the entire PC, server or laptop consumes."

If the OEMs dont get the correct/actual specifications then
they are in serious trouble.Why ?

It(TDP) allows the OEM to design a system that can handle the heat generated by the processor based on this value.

I know many on D.T. design & build their own rigs-customized to their needs/specifications.

I am sure there many out there who also do the same.

They do take these factors into considerations.

Moving Average
By InternetGeek on 12/10/2007 8:40:58 PM , Rating: 3
So basically what happened is that the geometric average was not affected by the TDP increase.

I think their average is quite flexible given that an increase of 20% in TDP does not actually move it. This could either make it a safe number or get you some shudders as a user: How well does your cooling solution handle the max thermal load after a long time?

So far their TDP has not caused any burnt out CPUs, but could there be any stability issues because the TDP did go up 20%?. That's 1/5 of the amount. For a TDP of 125W, that's 25W. Might not sound much, but apply 25W to a pot of water for 3 hours and it will boil. Long before the 3 hours are up.

RE: Moving Average
By JumpingJack on 12/10/2007 9:55:45 PM , Rating: 3
Actually, this doesn't make sense ...

One of two things are occuring here ...
1. Their original spec was wrong, in which case they are correcting the values (odd because the big deal with AMD was that they touted they would fit quad core into the same thermal envelop as thier 90 nm dual core).


2. The 5 applications they chose to determine ACP does not include worst case, which again would make the entire concept of ACP appear wrong.

The geometric mean will be affected if the population used to determine it changed. This is silly to think otherwise.

RE: Moving Average
By Viditor on 12/11/07, Rating: -1
RE: Moving Average
By JumpingJack on 12/11/2007 12:41:43 AM , Rating: 2
Yeah... right, please pull the Intel spec that states they produce a geometric mean on power by measuring on 5 applications to arrive at TDP.

Don't quote the article in this context, while he is correct... this is not how Intel arrives at TDP.

AMD uses a geometric mean... bogus, it artificially lowers the actual average that most average Joe's understand as average (which is arithmetic mean)... Geometric mean is always less (or if the population is tightly grouped equal to) the arithmetic mean, statistically speaking this means 1/2 the time you run your processor you will exceed this power... oddly, I have never had this issue on an Intel CPU.

AMD fudges this APC number down using ACP on only 5 apps, hardly representative of real world -- they are all synthetics .... that's ok, it is a start, but you are being lead down a primrose path.

RE: Moving Average
By Viditor on 12/11/2007 12:51:25 AM , Rating: 2
If you could get us a link to what you're talking about, that might help.

If you feel that TDP and ACP aren't comparable, you must also agree that TDP from Intel isn't comparable to TDP from AMD doesn't stop people from comparing them though.
So the question is, which is more representative of the truth...TDP or ACP?

Again, a link to Intel's documentation of their methodology on derivation of TDP would be a good start...

RE: Moving Average
By JumpingJack on 12/11/2007 1:29:42 AM , Rating: 3
Sure... no problem.... it gets complicated, the TDP definitions and meanings are most certainly not the same....

Intel TDP definition:
Thermal Design Power: a power dissipation target based on worst-case applications. Thermal solutions should be designed to dissipate the thermal design power.

So you can read through the spec, Intel measures TDP as:
The processor TDP is based on measurements of processor power consumption while running various high power applications. The data is used to determine those applications which are interesting from a power perspective. These applications are then evaluated in a controlled thermal environment to determine their sensitivity to activation of the thermal control circuit. This data is then used to derive the TDP targets published in the processor data sheat

The thesis is that AMD's ACP is Intel's TDP. This is categorically untrue, Intel does not derive their TDP from an 'average consumption' no where does Intel describe or spec an 'average' power value as the TDP, rather they evaluate worst case applications, derive the thermal profile and set the maximum power to be dissipated by the thermal solution for te processor to operate.

Intel's TDP definition is differnt from AMD's definition, AMD only considers 'relevant commercial software' as their load conditions.
page 10. They then use the max Vcc and Idd_max to buffer against thermal over-run (which you will find in the foot notes), mostly because AMD runs right up next to the spec limits of their processor capabilities.

For example, see the max supply voltage and max Idd for the FX-62 (now defunct), page 53. TDP is 125 Watts, VID_VDD and Iddmax for the Pmax state (i.e. no cool and quiet) is 1.4 Volts at 90.4 AMPs power=voltagexcurrent, = 1.4x90.4 = 126.6 Watts. Take for example, page 18 for the 3800, 4200 or 4400+, the high side of the voltage is 1.35 volts, current is 66.2 , they use power = 1.35 x 66.2 = 89 Watts (wowww).

AMD has to it this way because they run out the gambit of their spec values just to keep clock speed up. The thermal profile AMD chooses simply enables them to run their processors right up against spec if need be.

The fact that AMD covertly increased thier TDP for the quad core through these whitepapers is only indicative that they cannot achieve clockspeed at the voltages needed to meet their old thermals, so instead they just loosen the spec.


RE: Moving Average
By Viditor on 12/11/2007 2:28:03 AM , Rating: 2
The thesis is that AMD's ACP is Intel's TDP

I don't think that's it me it looks like a measurement of 3 different things.
In reversed order of power demand:

1. AMD's TDP is a measurement of theoretical max. Using every transistor at full capacity under worst case ambient temperatures. This is the highest estimate possible, and is never achieved in the real world.

2. Intel's TDP is a measurement of what they consider to be worst case applications under full load. This is an actual measurement, so it isn't as high as theoretical max, but it certainly is far higher than average useage.

3. AMD's ACP is an attempt to guage power under typical usage. It can be applied to AMD as well as Intel chips, and is more representative of what actual results will be...but it still isn't the same as Intel.

So no, there is still no comparing power usage between AMD and Intel without measuring each system (and that's not even mentioning the Northbridge power or FBDimms).

RE: Moving Average
By JumpingJack on 12/11/2007 3:08:16 AM , Rating: 4
Ok, I think we are reaching a consensus here...

point 1 - Actually, AMD measures TDP it is not theoretical but the measure it driving that power bin at the top of the supply voltage and current -- hence the reason IxV ~ TDP.
Thermal Design Power (TDP) is measured under the conditions of Tcase Max, IDD Max, and VDD=VID_VDD, and
include all power dissipated on-die from VDD, VDDIO, VLDT, VTT, and VDDA.
This is from one of their footnotes for TDP in their spec tables.

AMD's definition for TDP though is also in that same PDF:
Thermal Design Power. The thermal design power is the maximum power a processor can draw for a thermally significant period while running commercially useful software. The constraining conditions for TDP are specified in the notes in the thermal and power tables.

AMD's definition is a bit wierd to me ... what is the thermally significant period and what is commercially useful software? Nonetheless, it is clear from AMD's specs that they are near the max within that power bin, as the supply voltage at it's max end * Iddmax (max current the processor draws) ~ TDP. Nonetheless, I agree that AMD does run their TDP up to the max the processor can handle, and it is measured not using worst case apps for loading, but using 'commercially useful' software.

On point 2 - I can agree here, because Intel's processors have a much larger operating range for the supply voltage and Iddmax (Intel calls it Icc). Let's take for example a QX6800 -- page 18, max supply voltage is 1.5 volts, max current is 155 which would give a theoretical max of 232.5 Watts -- of course this is not where it is operated, but this is what their desing/process will allow per spec.

What is important is that Intel will does not assign an average power value to make TDP but the worst case such that the processor will work under all conditions provided it is cooled to specification. This is not the same as the max possible power for the CPU, which is Pmax and for the x6800 is 232.5 Watts.

On point 3 -- I am very skeptical of AMD's ACP measurement, it would at first pass appear to me as a means to assign a value for power to the CPU to fool the gullible. Let me put it another way, would you choose your heat sink based on ACP? Nope. It does give a guide on actual consumption which is not TDP, let me reinforce this TDP != consumption. But is it actually representative? Not really, 5 apps and, they calculate a geometric mean on top of that ... geometric mean is always <= arithmetic mean average (which is what most people understand).

Though it would be interesting to see what one would get for Intel chips within their power bins to see if the deltas are the same.


RE: Moving Average
By Viditor on 12/11/2007 5:10:52 AM , Rating: 2
I agree that we mostly have consensus...

As to ACP, remember that AMD isn't calling it anything like TDP.
Therefore, it seems to me to be what TDP isn't...namely a measurement for power usage during average useage.

RE: Moving Average
By JumpingJack on 12/11/2007 12:03:24 PM , Rating: 2
Ok... good ... so were are honing in... I am skeptical of AMD's APC for various reasons, most certianly we can agree that APC is not the number to use for designing the thermal solution.

You may or may not agree that APC is a good measure of actual consumption, I would not agree mostly because it comes across as a 'fudge' factor to me -- first they condense it down to a geometric mean but call it an 'average', a normal person might think they summed and divide, where as they actually multiplied and nth rooted. This always results in a lower summary mean than the arithmetic mean. As such, assuming a normal distribution, the probability of loading the processor the consume more than APC is higher than finding one that loads it lower than ACP.


RE: Moving Average
By JumpingJack on 12/11/2007 3:41:19 AM , Rating: 2
So no, there is still no comparing power usage between AMD and Intel without measuring each system (and that's not even mentioning the Northbridge power or FBDimms).

On the last point... very true... to compare consumption, the important measurment is the platform level all things accounted for... fixating on CPU power consumption as AMD has is nothing more than marketing 101. In that respect, TDP != consumption which is so often ignorantly and falsely assumed, TDP is important for designing thermal solutions because that is exactly what it is for ... a rate of disspating heat energy.

Honest approach.... is the way out.
By crystal clear on 12/10/2007 9:25:13 PM , Rating: 2
AMD should take the honest approach=

"yes we have problems with Barcelonas & Phenoms-we are working on them,till then NO products in this catergory to be released."

This honest approach will be much appreciated by all-as we all understand its not so easy & so simple.(Even Intel will agree)

Dont hide-dont fake-dont decieve...just be honest.

Shareholders/buyers/press will loose the confidence they have in AMD.....

Down at the market level the word going is-

Dont buy-Wait-let B2 stepping come out 1Q 08 approx or later-let it be revued independently & then decide.

If IBM releases products based on them then its fine-if IBM doesnt do so then there are still problems.
Thats the opinions at the martket level-IBM & SUN get first priority allotmrnts of CPUs.

Dont blame the R&D-Blame the management-they have to GO.

RE: Honest approach.... is the way out.
By TheJian on 12/10/07, Rating: -1
RE: Honest approach.... is the way out.
By crystal clear on 12/11/2007 1:47:48 AM , Rating: 2
AMD at this critical stage should be concerned about itself rather be bothered with What Intel did or does when it(Intel) has/had a problem.

Priority is going where the chips can go.

Priority should be as their CEO said recently-Profitability is our goal number 1

Priority should be what they said in 3rd Q results-

We to hope breakeven in the 4th Q

They can achieve these objectives by selling to Tier 1 Vendors/OEMs & not a comapatively small HPC market.

Your quote above sounds like a "clearance sale" rather than a new product launched into the market getting in those badly needed revenues+profit margins+marketshare.

Now I quote you-That's good business sense ,think marketshare/profitability/revenues etc.

I think the market level word is "I can't get any, so who cares until I can".

OEMs/Vendors usually the Tier 1 & 2 dont think like you.

These are people that bring in those huge revenues for AMD.

IBM & the rest have already stopped giving ETAs for servers that were set to use the quad-core Opteron .

Not deceit, hiding or faking. By the time we can actually hold ten in our hand on any given day they will have sorted this out.

No guarrantees on this-sounds like the usual AMD statements, like this one.

AMD's president and COO Dirk Meyer said in an October earnings call that its new server chips would be "widely available by the middle of this quarter" and that the company expected "to ship hundreds of thousands of quad-core processors this quarter into the server and desktop segments."

The Sunnyvale, Calif.-based chipmaker is still "on track to do hundreds of thousands of quad-cores" shipped in the fourth quarter, Hughes said. However, "on the first point, that has been revised a bit," he said, referring to Meyer's statement that the chip codenamed Barcelona would be "widely available" as of December.

Now can you trust them ???

By TheJian on 12/12/2007 5:25:40 AM , Rating: 2
I think its easier to tell one company with 1000's of your chips how to flash a bios than it is to deal with 1000 end users who struggle with finding the windows start button :) I'm guessing Tier1's didn't want these flawed chips (I wouldn't sell these to the public if I was Dell etc). So sell them where you can or make nothing on them. If sold to the public you'll only create a tech support nightmare (along with the bill to cover it). I hope that makes more sense now, that was all I meant by the comment about priority. It's a tough sell to us at home knowing they have a problem at even 2.4ghz. Intel's 2.4 can easily run at 3.4. Servers aren't interested in this point. They run at default and can be fixed in mass numbers without all the phone lines lighting up :)

I'd say it is a clearance sale of sorts. Sell something or go broke. It's not exactly out of norm anyway. They have been selling the first chips to servers before this. I may be upset that this means a further phenom delay but it's still in line with what they did on the last rev. Server...Desktop...Mobile. They don't have enough production (few 100K this year?) to satisfy any Tier1 or 2 vendor anyway. I had a PC business for 8 years before joining an IT dept so I kind of feel their pain with regards to being shafted by chip availability. I'm feeling it now over the 3870/8800GT. Have you tried to get one of these without paying a 25% premium over MSRP? Even they they say get on a waiting list (AMD's words...Nvidia has just been silent). OUCH.

I don't see having a production problem and then revising your numbers for end of year as deceit. Forcasting something like this is impossible. I don't think Intel shipped P3 1.13 or Camino boards thinking they'd get away with it. They recalled them because they found a flaw and had to deal with it. Much like AMD's production problems. I'm sure Bill Gates didn't plan on a BSOD on stage either (that still cracks me up...ROFL). It would have surprised me if they DIDN'T revise down barcy's shipment numbers. Better to revise down than recall 400,000 chips I think.

By crystal clear on 12/11/2007 12:22:03 AM , Rating: 1
Sorry typo-should read B3 stepping.

ACP Fundamentally Flawed
By Brian Clark on 12/10/2007 8:50:22 PM , Rating: 3
I find it difficult to accept AMD's explanation of ACP, and honestly find the concept to be flawed.

For one thing, power draw isn't a steady state function. It varies over time, and AMD is not exactly clear about where they measure the datapoints through which they compute the average.

Second, a geometric mean is all wrong for a useful power measurement, since they tend to discount measurements that are either very high or very low compared to the mean, and that's precisely the kind of datapoints that are most useful. You would rather know about the peak power draw for a processor, rather than the average, no matter if you were a system/thermal designer, or a customer who is worried about supplying power to their datacenter.

Although, if you were the latter, you would probably also be concerned about platform power, measured at the outlet. But let's forget about that for a moment, since that's another conversation altogether. The industry is in dire need for a platform power standard, but for right now, this article brings up an interesting topic about processor specific thermal specifications.

And on that topic, what would a person want to know about the thermal characteristics of a microprocessor?

As someone mentioned in the previous article, someone might be interested in how much power their datacenter might draw over time, in order to figure out electricity and cooling costs. But for one thing, the ACP specification is inadequate, since only the platform level power would be enough to calculate these things. And if you were also interested in electricity and cooling "requirements", then you would want to know about theoretical max power draw. Either way, ACP can't help you.

It is potentially a component of something larger, but what is AMD doing so that server vendors can get the average power of the rest of their components, and provide a platform level total? AMD seems interested in pushing ACP as a meaningful metric to end-users, but I would claim that end-users don't care, and would rather AMD supplied this number to the server vendors, who would combine it with the rest of the platform components to create something meaningful.

Although, if AMD did in fact do this, there is also the question of whether I should trust the results. AMD happened to choose a rather arbitrary set of application workloads to calculate their ACP number, and who are they to say that my workloads aren't more or less power hungry? Can AMD perhaps include a way of calculating error bars around their number, or perhaps derive a min/max value for ACP that allows me to estimate how my applications fall within this range?

It sounds like a good idea, but then... why not just give the max power, and I'll assume that my apps draw somewhat less than this. Hypothetically, we can call this max power the "Thermal Design Point", and rest assured that all real world applications that they've tested will not exceed this amount for a thermally relevant amount of time...?

Oh, wait. That's exactly what TDP is. Maybe someone can remind me again why ACP is supposed to be more relevant, in light of what I just wrote....

RE: ACP Fundamentally Flawed
By homerdog on 12/10/2007 9:23:03 PM , Rating: 2
Oh, wait. That's exactly what TDP is. Maybe someone can remind me again why ACP is supposed to be more relevant, in light of what I just wrote....

Because AMD doesn't want people to know that you can scramble eggs on their processors.

By NICOXIS on 12/10/2007 8:46:48 PM , Rating: 2
the pinoccio cartoon is really amusing and quite concluding LOL xD

By AmberClad on 12/10/2007 8:50:08 PM , Rating: 2
It's a pretty funny juxtaposition with the previous article's cartoon...

Pointless argument
By Justin Case on 12/11/2007 2:15:53 PM , Rating: 2
AMD's ACP value and Intel's (pseudo) TDP value are both completely useless.

For system design, what matters is the real maximum TDP (which AMD still reveals publicly, and Intel used to reveal to the major OEMs - I presume they still do, but smaller OEMs need to determine it on their own, or just leave a margin over the "public" figure and keep their fingers crossed that the system won't overheat and throttle down when doing something really heavy like 3D rendering, FEA, etc.).

For consumers and IT managers, what matters is the system's total power consumption (no sense in measuring the CPU alone when it needs - at least - a motherboard and RAM to work).

AMD's ACP is just a half-assed attempt to deliver a pseudo-TDP number (like Intel does) without being as blatantly dishonest as Intel (who can basically lie about whatever they want and never be called on it, because they have a lot of the "IT media" in their pocket).

So this is just another example of marketing-driven tech journalism (and [i]biased[/i] journalism, at that, but I guess that's to be expected, when advertising from one company pays people's salaries).

RE: Pointless argument
By TheJian on 12/12/2007 5:42:02 AM , Rating: 2
I couldn't have said it better myself. Though I think half-assed is a bit overboard. It's much closer to Intel's reality now. Why advertise cpus from 2.0-2.8ghz (2.6?) as 89w when the low models are nowhere near that? Who came up with that? All it got them was uneven press for 4 years.

But yeah the numbers are pretty useless for all the reasons you mentioned. System designers definitely get the proper numbers to design a box, and all an IT Dept needs is a KILL-A-WATT meter to find out what they need to know :) $20 at newegg last I checked (including shipping..heh). Test at the wall...Undeniable answers given.

So their original spec was wrong...
By JumpingJack on 12/10/2007 9:49:16 PM , Rating: 2
"AMD's Brent Kerby, author of both whitepapers, explains the inconsistency with timing and the nature of ACP itself. "The measured value of ACP already included the changed TDP values," he explains. "

So what is he saying?? That their original spec was wrong.

What happens to all those customers who already designed thier thermal solutions to the old TPD values?

Storm in a teacup
By psychobriggsy on 12/11/2007 6:08:42 AM , Rating: 2
Hmm, I remember that Intel have been giving TDP numbers that are what the CPU will use under normal high-load applications for years on end. AMD just used Volts * Amperes, and thus always looked higher (some exceptions like the 1.4GHz Thunderbird come to mind) which in the past two or three years has been quite bad for them when an 89W TDP CPU was actually using 50W. Of course AMD's "Family TDP" was a good reason for this - all cooling was made for 89W TDP so if you upgraded you could keep the same cooling. Especially good for servers too.

AMD recently switched to being similar to Intel's "Max Application Power Draw" type TDP if I recall correctly. Of course, they needed to do this as one way of not looking worse. They then used this in conjunction with a variety of family TDPs (65W, 89W, 105W, 45W ...) which is probably for the better.

Average TDP is good - if my CPU is going to spend most of its life at 1GHz then that should be factored in. Again, AMD drove power saving on the desktop by bringing their PowerNow! from mobile applications across, and Intel followed later on.

This whole storm in a teacup looks like a minor issue really - a result of AMD actually publishing data. Some websites though like to spot dishonesty, like those that said Asus deliberately broke the GPL with the EeePC recently, and they've been stirring the situation. That's just poor journalism.

By wetwareinterface on 12/14/2007 12:46:54 AM , Rating: 2
TDP the Intel way...

Find the highest average failure point in relation to heat and wattage usage for the cpu, then let the cpu scale higher (temporarily) in envelope before cutting it back.

TDP the AMD way...

Run several cpu's from the same family and stepping at several different frequencies, extrapolate the highest thermal and power requirements you expect to see in that cpu family's range at the highest forseeable clock values.
Stamp all cpu's from that family, regardless of actual operating frequency with this extreme utmost possible value for TDP.


Test several cpus of the same frequency and stepping using a workload metric and report back on the actual power and heat those particular cpus at that particular frequency dissipated under that load condition.

See how the new ACP measurement is actually closer to Intel's TDP measurement system now but still different?


both reflect individual cpu's

both do not accurately reflect highest possible heat and power requirements.

Intel's system gets closer to highest value you'll see

AMD's system is closer to real world conditions and power needs for "average" workloads than the AMD TDP rating which was always way too high

This is how the TDP of the AMD whitepaper can go up but the ACP stay the same. They simply extrapolated a higher end than they did originally. The ACP remained the same because that was run on actual cpus at a specific frequency. The TDP went up because they extrapolated a higher end part than they did in the first place which caused a higher TDP for the processor FAMILY.

"So, I think the same thing of the music industry. They can't say that they're losing money, you know what I'm saying. They just probably don't have the same surplus that they had." -- Wu-Tang Clan founder RZA
Related Articles
AMD Phenom 2008 Roadmap
December 5, 2007, 10:08 AM

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki