AMD TDP definition

[Magic Carpet]

If the spec is not released you cannot say if the device is out of it or not. And it is the guy who claim that the FX chips are out of spec who has to back his statements.

Good point. If AMD's TDP doesn't mean actual power consumption then it can't, obviously go out of spec. Personally, I haven't seen a proper datasheet since Deneb/Thuban.

 

[mrmt]

Good point. If AMD's TDP doesn't mean actual power consumption then it can't, obviously go out of spec. Personally, I haven't seen a proper datasheet since Deneb/Thuban.

Read again:

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.

It isn't me or someone else claiming that TDP = Power consumption, but AMD itself on a server manual.

One thing is AMD to withhold data from 15h family, and with that we cannot verify whether AMD TDP numbers are real, or at least in which conditions they are real. But another very different thing is to claim that TDP is not power consumption, because this claim comes from AMD itself, and unless you can bring official documentation stating that AMD changed the TDP definition, then it's safe to consider that TDP definitions stated in the K8 and K10 manuals still stand.

 

[galego]

Read again:

It isn't me or someone else claiming that TDP = Power consumption, but AMD itself on a server manual.

I know this was said to you before, but I will try again. You omit to bold "for a thermally significant period while running commercially useful software".

Intel makes something similar. Read the Intel definition given above.

Again TDP != Power consumption.

 

[mrmt]

I know this was said to you before, but I will try again. You omit to bold "for a thermally significant period while running commercially useful software".

Translate this definition to your native language or ask AMD for a translated version of the manual, and it might start to make sense for you. You can't seriously be arguing against a straight definition like that.

 

[Magic Carpet]

Translate this definition to your native language or ask AMD for a translated version of the manual, and it might start to make sense for you. You can't seriously be arguing against a straight definition like that.

No actually, galego has a valid point. Running commercially useful software may not lead to the same power consumption seen in benchmarks like LinX, etc.

It's similar to what we have with GPUs, average gaming consumption is a lot less than synthetic stress tests like Furmark, etc.

I see nothing wrong there.

 

[ShintaiDK]

No actually, galego has a valid point. Running commercially useful software may not lead to the same power consumption seen in benchmarks like LinX, etc.

It's similar to what we have with GPUs, average gaming consumption is a lot less than synthetic stress tests like Furmark, etc.

I see nothing wrong there.

So what you say is, that its perfectly ok for the system to overheat, throttle, shutdown and otherwise not function as it should. just because you run something that wasnt under the very vague and random definition of "comemrcially useful software".

Thats not gonna hold in court. GPU makers are already on the borderline for a huge smack there.

 

[Enigmoid]

No actually, galego has a valid point. Running commercially useful software may not lead to the same power consumption seen in benchmarks like LinX, etc.

It's similar to what we have with GPUs, average gaming consumption is a lot less than synthetic stress tests like Furmark, etc.

I see nothing wrong there.

The problem is often certain types of commercial software use almost as much power as linx.

For example on GPUs.

Looking at a commercially useful software program (bitmining) vs torture test the difference in power consumption is basically 0 (3% more).

The problem is the definition of "commercially useful software".

 

[SlowSpyder]

I don't see why AMD doesn't just publish the numbers. Its no secret that AMD chips use a good deal more power than Intel chips these days, there's no reason to hide this info.

 

[ShintaiDK]

The HD6970 for example throttled not only at Furmark. But also Metro 2033 and 3Dmark. Possible more.

 

[Idontcare]

I think all of this was debunked before.

Someone debunked the fact that the thermal power dissipated by an integrated circuit is temperature dependent?

Uhm, no, device physics don't work like that.

You can't specify a TDP value without simultaneously specifying the max allowed operating temperature. Without the temperature info the TDP number itself is meaningless.

 

[mrmt]

No actually, galego has a valid point. Running commercially useful software may not lead to the same power consumption seen in benchmarks like LinX, etc.

It's similar to what we have with GPUs, average gaming consumption is a lot less than synthetic stress tests like Furmark, etc.

I see nothing wrong there.

Video encoding is not commercially useful software?

 

[Exophase]

The whole point about TDP being meaningless w/o temperature specifications is technically correct, but personally I think that when AMD publishes a TDP spec w/o a temperature that this TDP should be applicable for a) the CPU running with their stock HSF b) in a case that has an average amount of cooling c) in an environment close to room temperature. In other words, not something that'd require more extreme much less exotic cooling that a normal user would not think necessary to run the thing at stock.

Anything less is skirting by on a technicality, and sometimes (at least where legal action is not pursued) being grossly misleading is almost as bad as outright lying.

We could argue about how Intel defines their TDP but their high end desktop processors have no problem consuming less power than that under heavy workloads at stock speeds under stock/typical cooling in a normal environment.

 

[Charles Kozierok]

There seems to be a whole lot of semantic parsing and "rules lawyering" going on over what appears to me to be a fairly simple concept. It's made clear in the name of the metric: thermal design power, meaning, the power level for which the thermal system should be designed.

That doesn't mean it is the absolute maximum power that can be drawn, because the thermal system is designed based on overall, steady-state heat production. But at the same time, it should be a level that, if the cooling system is designed to match it, will allow the system to function properly under a wide range of use scenarios.

Since thermal design power inherently represents a power-temperature tradeoff point, it's questionable as to why a company would specify TDP without also making clear what temperature the TDP is based on. And if a chip is routinely, steady-state, drawing more power than its rated TDP, it's entirely valid to ask what the basis was for that determination.

ETA: Exophase said some similar things as I was writing my post. Also, companies that care about quality build in safety margins into their specifications; IMO a good chip should not only not be exceeding its TDP regularly, it should have headroom.

 

[Rakehellion]

Now compare with Intel definition:

The Intel and AMD definition are exactly the same. Where's the outrage?

 

[galego]

Someone debunked the fact that the thermal power dissipated by an integrated circuit is temperature dependent?

Uhm, no, device physics don't work like that.

You can't specify a TDP value without simultaneously specifying the max allowed operating temperature. Without the temperature info the TDP number itself is meaningless.

There is no unambiguous definition of TDP. Don't pretend that Intel has one.

The Intel and AMD definition are exactly the same. Where's the outrage?

Apart from releasing eight-core cpus clocked up to 5 GHz and winning the best award at computex with Temash apus I cannot see any other outrage.

galego, enough with the thread crapping. This has nothing to do with TDPs and you know it
-ViRGE

 

[Charles Kozierok]

There is no unambiguous definition of TDP. Don't pretend that Intel has one.

I see no ambiguity in Intel's definition.

Apart from releasing eight-core cpus clocked up to 5 GHz and winning the best award at computex with Temash apus I cannot see any other outrage.

What relevance does this have to the topic at hand?

 

[mrmt]

The key here is the reference to "under the conditions of all cores operating at...Tcase Max".

IDC,

Digging into AMD website I found this:

http://products.amd.com/en-us/Opter...f6=G34&f7=C0&f8=32nm&f9=115+W&f10=6400&f11=8&

Max Temps (C) 71.7'C

This would be a good guess for Tcasemax for that Opteron processor, no? But look at how the clocks are "conservative" on the server side: To achieve 3.2GHz they have to spend 115W at a rather low Tcase Max (compared to Intel's). This also means that they need almost the same power that Intel Core needs to get to the same frequency with 6 cores. And this is the best bin, they have another one with 2.8GHz base clock.

Hmmmm.... after reading this, I cannot but be impressed by AMD commitment to the enthusiasts community. Instead of selling their best Opteron bins (4GHz with 125W) for thousands of dollars in the server market, they are almost giving them away in the consumer market, just to keep the enthusiasts community happy. I'll buy one FX9570 to support AMD.

 

[Ajay]

Someone debunked the fact that the thermal power dissipated by an integrated circuit is temperature dependent?

Uhm, no, device physics don't work like that.

You can't specify a TDP value without simultaneously specifying the max allowed operating temperature. Without the temperature info the TDP number itself is meaningless.

The AT forum is going to have to come up with a saint hood icon for you - for repeatedly answering these posts.

 

[Exophase]

Hmmmm.... after reading this, I cannot but be impressed by AMD commitment to the enthusiasts community. Instead of selling their best Opteron bins (4GHz with 125W) for thousands of dollars in the server market, they are almost giving them away in the consumer market, just to keep the enthusiasts community happy. I'll buy one FX9570 to support AMD.

So you're saying FX-8350 is rated to emit, over thermally significant time periods while running commercially relevant software, no more than 125W on average when the case (not Tjmax, case) is 71.7C. If that's how it is then it's not giving enthusiasts their best bins, it's giving enthusiasts a lie.

Probably the "real" TcaseMax for FX-8350 would be dramatically lower than that.

 

[galego]

I see no ambiguity in Intel's definition.

Others see it. Maybe you would pay attention to ambiguous words such as "near", "significant",... used by Intel in its definition of TDP.

What relevance does this have to the topic at hand?

More or less the same that your question here.

 

[Exophase]

The definition is for all intents and purposes the same.

When AMD says "maximum power consumption" they are referring to the maximum average power consumption over a thermally significant period of time running commercially useful software.

Intel refers to the exact same power consumption. When they call it "near maximum" the maximum that they are talking about is the one that accounts for short power spikes that would go over on a thermally insignificant scale.

AMD uses the word thermally significant too, so I don't see how you could possibly make this an Intel-only fault. Nor do I see anything especially wrong with this language.

 

[galego]

The definition is for all intents and purposes the same.

When AMD says "maximum power consumption" they are referring to the maximum average power consumption over a thermally significant period of time running commercially useful software.

Intel refers to the exact same power consumption. When they call it "near maximum" the maximum that they are talking about is the one that accounts for short power spikes that would go over on a thermally insignificant scale.

AMD uses the word thermally significant too, so I don't see how you could possibly make this an Intel-only fault. Nor do I see anything especially wrong with this language.

Who said "Intel-only fault"? Link please. As shown in #43 both definitions are equivalent (and both are ambiguous).

Are others here who pretend that the fault is on AMD camp only, but they are wrong.

 

[mrmt]

At contrary others here pretend that AMD definition of TDP is ambiguous whereas that Intel definition is not. Moreover one poster even pretends that AMD claims that "TDP = power consumption" whereas other pretend that they are going out of specs... Both claims are plain false.

You are funny, really. You put an argument, that AMD TDP definition is ambiguous, and then you start to fight against your own argument as if we have put it here.

Stop the BS, nobody is saying that AMD TDP definition is ambiguous, quite the opposite, AMD TDP definition is very clear and straightforward:

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.

What we are saying here is that AMD is playing with their own definitions with the FX8350, either because at a reasonable Tcase Max power consumption cannot be 125W, or because you would need ridiculously low Tcase Max for the FX8350 TDP be 125W.

There is a reason why those specs aren't published by AMD for their desktop line and are timidly put on the Opteron line. They didn't have to hide those specs if they weren't up to some shenanigan, but they do, and they sure as hell are up to some shennanigan. There is not a chance in hell that fusing off hypertransport links would amount to an increase of 800Mhz in clock frequencies and only 10W more power consumption.

 

[Charles Kozierok]

Others see it. Maybe you would pay attention to ambiguous words such as "near", "significant",... used by Intel in its definition of TDP.

That's not an explanation.

Intel's definitions are explained here. What's ambiguous? Be specific.

More or less the same that your question here.

I'm trying to discuss TDP issues, per the thread title. That has nothing to do with winning awards at Computex. If you're going to mindlessly cheerlead for a company, at least keep it on-topic.

 

[AtenRa]

Again for people not familiar with TDP, its not the maximum power of the CPU not for AMD or Intel.

Have a look at the bellow picture taken from the OP link (page 18 in the pdf).
IF TDP was maximum power it would not change with Thermal resistance. Well, actually power can change UP but not down as we have in the tables bellow.
The tables are for the same processor, If TDP was maximum power, at the same tcaseMax we would have the same TDP.
From the tables bellow its clear that TDP IS Not maximum power but Thermal Design Power because its depended on the THERMAL resistance of the Heat-Sink Fan.
That means that with a better heat-sink fan we can have lower HIGHER TDP at the same tcase Max.

Look what happens in the two Heat-Sink designs in relationship to the tcaseMax.


Edit: ok fixed
Thermal Profile A heat-sink can dissipate 65W with a tcaseMax of 63.1c when at the same tcaseMax Thermal Profile B can only dissipate 45W. At their maximum performance, Thermal Profile A can dissipate 95W with a tcaseMax of 70.0c when Thermal Profile B Heat-Sink can only manage to dissipate 65W at the same tcaeMax of 70.0c.


Plz read the Note at the bottom of the pic. After that if its not clear what TDP is, you shouldnt even talk about it.