AMD TDP definition

[mrmt]

It seems that AMD disagrees with some here. TDP = Power consumption:


http://support.amd.com/us/Processor_TechDocs/43374.pdf

TDP. 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.

TDP is measured under the conditions of all cores operating at CPU COF, Tcase Max, and VDD at the voltage requested by the processor. TDP includes all power dissipated on-die from VDD, VDDNB, VDDIO, VLDT, VTT and VDDA.

 

[ShintaiDK]

And the trick is to not release any specs, so you can run the CPU out of spec like the FX series.

Commercially useful software is also another little neat trick to avoid holdign a proper specification.

 

[inf64]

What is the purpose of the thread exactly? We have topics discussing this and similar specs of AMD CPUs. Unless you had in mind instigating a flame war...

Oh and you forgot to underline and bold the most important parts above:
"while running commercially useful software."
"TDP includes all power dissipated on-die from VDD, VDDNB, VDDIO, VLDT, VTT and VDDA. "

Since you reported this thread; we're of the opinion that it's useful, disagree that it's flamebait (though there's an obvious need to keep an eye on it) and will be keeping this thread open
-ViRGE


EDIT by me:

Wow you are so not biased and helpful. Thank you so much.
-inf64


EDIT by me:
Mod callouts in threads because you disagree with mod actions are not acceptable. If you disagree with mod actions, post here. Virge was responding to your threadcrapping - something you do a lot. It's not your job to determine if a thread is flame bait, it's ours, and Virge made that decision.
Administrator allisolm

 

[monstercameron]

It seems that AMD disagrees with some here. TDP = Power consumption:


http://support.amd.com/us/Processor_TechDocs/43374.pdf

TDP. 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.

TDP is measured under the conditions of all cores operating at CPU COF, Tcase Max, and VDD at the voltage requested by the processor. TDP includes all power dissipated on-die from VDD, VDDNB, VDDIO, VLDT, VTT and VDDA.

i guess that means furmark s out of the question.
so let me get this straight, an 8W apu uses 8W normally or does the igpu and ifch inflate the power draw above its rated tpd?

 

[mrmt]

i guess that means furmark s out of the question.
so let me get this straight, an 8W apu uses 8W normally or does the igpu and ifch inflate the power draw above its rated tpd?

Nope, 8W would be for everything.

 

[Atreidin]

"...for a thermally significant period while running commercially useful software"

There is so much wiggle room in that definition that it is practically meaningless, and pretty much refutes the point you are trying to make. I'm not sure what arguments you are referring to, but I only recall people writing about how the definition is more complex than people want it to be, and that AMD is not violating their definition. Which it is, and they aren't. So good for you for looking it up and proving them right.

And the trick is to not release any specs, so you can run the CPU out of spec like the FX series.

How do you run a CPU out of spec when you don't have any specs?

 

[MisterMac]

Most people that aren't fanboi of either side - already knew this.

90% of us already know how the game work for AMD & Intel and other chip companies.


...This is kinda flamebait Virge.



Re: "This is kinda flamebait Virge".

The proper place for this is Mod Discussions, as I explain in my post, #35.

Please refrain from this behavior. I hope you can understand why this is a floodgate that simply cannot be opened.

Moderator jvroig

 

[inf64]

...This is kinda flamebait Virge.

He disagrees... What a surprise.
But glad someone else had balls to say it beside me.

 

[BrightCandle]

The specification for TDP is often misused. Its purpose is for the design of the cooling system that the CPU uses, that is it needs to be able to dissipate X watts of power. The CPU can however rely on the fact that the heatsink is enormous and thus changes temperature slowly in comparison to the CPU. Thus the CPU can for short periods (depending on the cooler, its size and peak dissipation) draw more than TDP if there is thermal room to do so. I think the definition of TDP is clear enough to serve its purpose.

But where people often go wrong is equating TDP to electrical usage. Because we can already see that the electrical power consumed can and does exceed TDP on a regular basis for small periods of time. Thus there are different electrical specs than thermal design specs for the CPU. TDP is not about the maximum power draw or the peak thermal output, its about the average peak thermal output with the assumption that the cooler can absorb the short peaks that exceed this value (where short is measured in a few milliseconds according to my bios) so long as the average is meeting the design.

I think the problem once was the power vampire programs that can push a modern part into a thermally problematic case. As far as I know these circumstances don't happen anymore, that boost, thermal monitoring and dynamic clockspeeds have put an end to this situation on CPUs and GPUs. Is that not the case?

AMDs definition does look problematic, that it potentially ignores a power vampire program and could exceed its cooling and kill itself. I don't know if that is possible but the spec implies it is.

 

[monstercameron]

The specification for TDP is often misused. Its purpose is for the design of the cooling system that the CPU uses, that is it needs to be able to dissipate X watts of power. The CPU can however rely on the fact that the heatsink is enormous and thus changes temperature slowly in comparison to the CPU. Thus the CPU can for short periods (depending on the cooler, its size and peak dissipation) draw more than TDP if there is thermal room to do so. I think the definition of TDP is clear enough to serve its purpose.

But where people often go wrong is equating TDP to electrical usage. Because we can already see that the electrical power consumed can and does exceed TDP on a regular basis for small periods of time. Thus there are different electrical specs than thermal design specs for the CPU. TDP is not about the maximum power draw or the peak thermal output, its about the average peak thermal output with the assumption that the cooler can absorb the short peaks that exceed this value (where short is measured in a few milliseconds according to my bios) so long as the average is meeting the design.

I think the problem once was the power vampire programs that can push a modern part into a thermally problematic case. As far as I know these circumstances don't happen anymore, that boost, thermal monitoring and dynamic clockspeeds have put an end to this situation on CPUs and GPUs. Is that not the case?

AMDs definition does look problematic, that it potentially ignores a power vampire program and could exceed its cooling and kill itself. I don't know if that is possible but the spec implies it is.

that might be the case my netbook over heats all the time when trying you watch flash video[especially if cpu decoded].

 

[mrmt]

But where people often go wrong is equating TDP to electrical usage. Because we can already see that the electrical power consumed can and does exceed TDP on a regular basis for small periods of time. Thus there are different electrical specs than thermal design specs for the CPU. TDP is not about the maximum power draw or the peak thermal output, its about the average peak thermal output with the assumption that the cooler can absorb the short peaks that exceed this value (where short is measured in a few milliseconds according to my bios) so long as the average is meeting the design.

I think the crux of the issue is that AMD TDP definition doesn't left too much room for doubt when it equates average power consumption and average thermal dissipation power (TDP). Spikes might happen as long as they aren't long enough to be deemed thermally significant.

It would be something like that:

Average Power Consumption = TDP
Average Power Consumption =! Maximum power consumption

With maximum power consumption not happening long enough to significantly change the average power consumption and consequently TDP.

 

[monstercameron]

using amds metrics would that mean that...bit of a jump here...arm soc tdps are closer to actual power draw?

 

[USER8000]

Most people that aren't fanboi of either side - already knew this.

90% of us already know how the game work for AMD & Intel and other chip companies.


...This is kinda flamebait Virge.

Agreed. It seems OTT of a mod to be editing someones answer just because they PMed some concerns about a thread. If it is not a concern,just ignore it.

Does that mean,everytime someone complains or reports a post or thread,the mods are going to edit the complainers/reporters post,saying they complained??

Edit!!

Wasn't there an older thread about this anyway

You don't have a clue what you're talking about so you should just not have posted.
Admin allisolm

 

[ShintaiDK]

using amds metrics would that mean that...bit of a jump here...arm soc tdps are closer to actual power draw?

You cant use AMDs TDP to extrapolate to ARM in any way. It might be the same, it might be difefrent, it might be ACP type, it might be SDP type.

 

[SiliconWars]

I don't follow every troll thread religiously so can somebody explain to me in a few words what the relevancy of this thread is, or is it simply an information broadcast on AMD's definition of TDP? (something I can find myself through google).

 

[inf64]

I don't follow every troll thread religiously so can somebody explain to me in a few words what the relevancy of this thread is, or is it simply an information broadcast on AMD's definition of TDP? (something I can find myself through google).

Don't ask. It's useful somehow. According to some "users" (who edit posts posting PM material in edits- just amazing).

Three modcallouts and a threadcrap all in the same thread. Not good.
Admin allisolm

 

[Gikaseixas]

Just find it wrong to make PMs a public thing. That was very inappropriate deserves an apology.


The proper place for this is Mod Discussions, as I explain in my post, #35.

Please refrain from this behavior. I hope you can understand why this is a floodgate that simply cannot be opened.

Also, see post #36 for an explanation to your concern.

Moderator jvroig

 

[Imouto]

He disagrees... What a surprise.
But glad someone else had balls to say it beside me.

Well, the police state in the VC&G forums ain't any better.


The proper place for this is Mod Discussions, as I explain in my post, #35.

This is counted as a mod callout - you are effectively calling out not just one person, but the entire VC&G mod crew.

Please refrain from this behavior. I hope you can understand why this is a floodgate that simply cannot be opened.

Moderator jvroig

 

[Cerb]

It seems that AMD disagrees with some here.

Um, OK. Can you shed light on what others are not in agreement about? I watch a few forums and posters, but generally just at many of the common flame/troll threads (thought some have been enjoyable, such as the PS4 thread in VC&G ).

So what do the, "some here," think, that's supposed to be an important point?

Intel tried to more or less make their TDP be like their new SDP, and backtracked, when some P4s could be found using much more power running real software (TBH, I think a userspace-settable SDP that the processor would adhere to would be awesome, and it looks like we may not be far from having that). While the physical layer is much more rudimentary by AMD, both them and Intel now have the CPUs tracking their workloads, so that they don't rely just on the temperature sensor to figure out where they are at in relation to thermal spec, so they can overshoot it a little for bursts of high load, followed by lower load periods (which are quite common for interactive use cases). None of it is so simple as "turning all of the transistors on," these days.

 

[mrmt]

Intel tried to more or less make their TDP be like their new SDP, and backtracked, when some P4s could be found using much more power running real software (TBH, I think a userspace-settable SDP that the processor would adhere to would be awesome, and it looks like we may not be far from having that).

I think that SDP was a good idea wrapped in awful marketing message. They should have said something like:

- Look, we have those 13W notebook processors but under certain circunstances (insert huge list of caveats here) you can have just 8W power consumption, meaning that you can design battery and plan your battery life using these parameters.

I don't think they freely floated the 7W number to their partners, otherwise they would have FX 8350 issues in a market they sorely need to establish themselves, and that would be very bad for business.

But the way SDP was presented, a clean "7W" announcement was awful, it hurt credibility in the tech community.

 

[LogOver]

I don't follow every troll thread religiously so can somebody explain to me in a few words what the relevancy of this thread is, or is it simply an information broadcast on AMD's definition of TDP? (something I can find myself through google).

I guess the goal of this topic is to warn people from making mistakes, when the next time they want to write something like "Intel TDP = Average Power, AMD TDP = Max. Power." Unfortunately, this wrong idea was inspired by some (former) AMD employees:
http://www.semiaccurate.com/forums/showpost.php?p=7949&postcount=86

Most power a system can draw:
Intel = Max power
AMD = TDP

Typical power draw for standard applications:
Intel = TDP
AMD = ACP

In fact AMD definition almost identical to the Intel's definition. Personally I didn't know that.

 

[BrightCandle]

Until Intel and AMD have genuinely the ability to limit the processors TDP this is always going to be an issue. If they keep defining TDP by setting the clockspeed and voltage and then measuring using some real world software its always going to be flawed when someone optimises software for the architecture of that chip. The software will then cause the CPU to exceed its TDP and then presumably it will throttle as the temperature climbs and then hit a steady state of consumption based on what the cooling can actually do.

That isn't really what TDP is meant to about. But they don't really have the ability to measure heat transferred and more to the point for the most part it doesn't matter. If the cooling is really designed to only dissipate TDP/SDP then in theory at least the CPU will just throttle based on the limits imposed by the cooling system.

 

[Cerb]

I think that SDP was a good idea wrapped in awful marketing message.

I agree, there. They presented it much as an alternate TDP.

What would be ideal, and should be possible, now (but may take a couple hardware generations, and OS generations, to fully iron out), is to make parameter sets like what makes up SDP configurable. Not necessarily you going and adjusting Watts used in an applet (but, that would be cool), but the OS/firmware power management actually being able to track approximate Wattage real-time, for better battery and fan management, since temperature alone tends to be too coarse-grained, already. That would allow for such limitations on the hardware, but without the rough throttling we're used to, now, and with less of the, "oh no, there's a high load, go fans, go!"

 

[mrmt]

What would be ideal, and should be possible, now (but may take a couple hardware generations, and OS generations, to fully iron out), is to make parameter sets like what makes up SDP configurable. Not necessarily you going and adjusting Watts used in an applet (but, that would be cool), but the OS/firmware power management actually being able to track approximate Wattage real-time, for better battery and fan management, since temperature alone tends to be too coarse-grained, already. That would allow for such limitations on the hardware, but without the rough throttling we're used to, now, and with less of the, "oh no, there's a high load, go fans, go!"

I don't think we'll get rid of the TDP as we know it. No matter how useful SDP could be, how coarse grained your thermal/load management is and and how far from TDP we stray in our daily using, engineers still need a ceiling to target when designing cases/coolers, and this ceiling is the TDP.

 

[CHADBOGA]

So do AMD still use ACP in their advertising?