AMD Carrizo Pre-release thread

[Abwx]

You should learn to read.

His intentionally perverted truths isn't worth any debate, especially when he can't prove his lies. There is no basis. And you should know it better.

The only lies here are your deffamatory claims...

Contradict me if you can :

Schematicaly a transistor max frequency can be estimated with its resistance R and its input capacitance C, with max frequency being the ratio C/R, undimensionaly of course since we dont need at this point to introduce the pulsation 2pi as its not a variable and can be neglected if further computations are done with linear operators.

Intel stated that they reduced C by a 0.65 factor, let take this as granted and this would increase max frequency by 1.54x or improve perf/Watt by the same amount, and that s indeed the origin of their perf/watt claim.

Now what they didnt say is that the transistor resistance increased such that it could be brought down to Haswell process value only by raising the voltage by 1.13x IIRC such that its resistance is decreased by the square of this ratio (That s a fundamental law of mosfets, transconductance increase as a square of the voltage ).

The result is that the charge energy stored in 0.6C is factored by 1.13^2 = 0.83 less power, that s 20% improvement at best since my computations neglect some details...

Here to help you build your demonstration :

https://en.wikipedia.org/wiki/MOSFET

In case you are in trouble i sort the relevant formulae about mosfets law of conduction so you know from where to start :

It is perhaps too simplistic and doesnt account for the transistor dynamicaly varying caracteristics, but this can be cured thanks to a more precise approximation with channel size modulation in respect of voltage accounted:

Of course all parameters can be broken into sub parameters that better modelise the behaviour but for our need we wouldnt even need the second formulae...


I hope you realize the total discretanpcy between your arguments and my own..?..

Of course it s not exactly directed to you, i would venture saying that without the constant pollution a lot of people would had have the occasion to learn something at least basic about semiconductors, in this respect your role is negative in all sense of the term, not only you are refusing to learn but you are keeping other people from doing so.

Now if AMD gear doesnt interest you for other than thread crapping there s Intel dedicated threads where you can bash AMD as much as you want while singing some gospels about imaginary perf/Watts, keep the AMD threads for people that are seeking other things that thread crapping...

 

[maarten12100]

What Intel said is that the TDP spec can be breached for a short period of time, but the spike would be so short that it wouldn't impact an average measured in any significant period of time. It is not what you said here:

They said the same thing I said how are you not seeing that?

30 seconds would be a thermally significant period of time by any standards.

Even if no energy was extracted from the reservoir(heatsink) 30 seconds at 10W is still only 300J of energy. Let's say the a 0,2 kg chassis as cooling and a uniform temperature rise let's assume the chassis is made of copper.

Copper has a heat capacity of about 400 J/(kg * K) It doesn't take a genius to finnd out the temperature rise if uniform and no extraction too place would be equal to 300/400 / 0,2 = 3,75 next to nothing and pretty possible in todays notebooks. A common scenario every recent notebook user is probably familiar with is the fan shutting down when the laptop is idle but then kicking in while idling all the time. There is not enough heat being removed from the reservoir and thus the energy of the reservoir increases eventually the temperature thresshold is reached and the fan kicks in.

Look at Core M's benchmarks scores decreasing if you run a bench for long periods or time after time. The reservoir is a high state when a new run is start obviously to keep the TDP the processor needs to be kept on a tight leash to make it happen.
In any recent product there is probably more going on than just TDP such as skin temp monitoring and all that fancy stuff but really this is just the way things are.

 

[monstercameron]

On a technical level, he's probably more likely full of peopleshit.

Generally only bulls make the bullshit. And chickens produce the chickenshit.
On a technical forum like Anandtech, the details matter.

As funny as this is, please don't feed the trolls.

 

[mikk]

The only lies here are your deffamatory claims...

Contradict me if you can :

There is no need for any debate. Claiming that going from HSW-Y to BDW-Y offered only a 10-15% performance per watt improvement is just plain wrong. Simple as that.

 

[monstercameron]

There is no need for any debate. Claiming that going from HSW-Y to BDW-Y offered only a 10-15% performance per watt improvement is just plain wrong. Simple as that.

Please continue this debate in the Intel thread please,

@abwx respond to him there and let's get back on topic.

 

[mrmt]

They said the same thing I said how are you not seeing that?


Even if no energy was extracted from the reservoir(heatsink) 30 seconds at 10W is still only 300J of energy. Let's say the a 0,2 kg chassis as cooling and a uniform temperature rise let's assume the chassis is made of copper.

Copper has a heat capacity of about 400 J/(kg * K) It doesn't take a genius to finnd out the temperature rise if uniform and no extraction too place would be equal to 300/400 / 0,2 = 3,75 next to nothing and pretty possible in todays notebooks. A common scenario every recent notebook user is probably familiar with is the fan shutting down when the laptop is idle but then kicking in while idling all the time. There is not enough heat being removed from the reservoir and thus the energy of the reservoir increases eventually the temperature thresshold is reached and the fan kicks in.

Look at Core M's benchmarks scores decreasing if you run a bench for long periods or time after time. The reservoir is a high state when a new run is start obviously to keep the TDP the processor needs to be kept on a tight leash to make it happen.
In any recent product there is probably more going on than just TDP such as skin temp monitoring and all that fancy stuff but really this is just the way things are.

The paper mentions power consumption over a thermally significant amount of time, so for every thermally significant amount of time power consumption = tdp.

 

[maarten12100]

The paper mentions power consumption over a thermally significant amount of time, so for every thermally significant amount of time power consumption = tdp.

So you finally agree? The statement was that TDP is only equal to power consumption when the duration of said measurement approaches infinite.

 

[CHADBOGA]

Yeah, nobody rigs benchmarks better than Intel -- at least according to the Federal Trade Commission:

http://sharikou.blogspot.com/2009/12/ftc-accuses-intel-of-rigging-benchmarks.html

Sharikou is a great man, you could learn a lot from him.

 

[Abwx]

There is no need for any debate. Claiming that going from HSW-Y to BDW-Y offered only a 10-15% performance per watt improvement is just plain wrong. Simple as that.

Get your facts straight, 10-15% performance per watt improvement was the number i stated for 15W TDP and recently for DT TDP in light of Hardawre.fr review..

As for your U/Y i already made a post for this corner case, it is here :

http://forums.anandtech.com/showpost.php?p=37570477&postcount=1717

And i m not interested in commenting my own post in said thread, my opinion is made, is scientifically motivated and as such wont change.

Please continue this debate in the Intel thread please,

@abwx respond to him there and let's get back on topic.

I did though but to better get back on topic...:biggrin:

 

[mrmt]

So you finally agree? The statement was that TDP is only equal to power consumption when the duration of said measurement approaches infinite.

Nope, by Intel itself power consumption = TDP for any thermally significant period.

 

[mikk]

Get your facts straight, 10-15% performance per watt improvement was the number i stated for 15W TDP and recently for DT TDP in light of Hardawre.fr review..

Learn to read:

You yourself referred to BDW-Y!

 

[Abwx]

Seems that AMD threads threadcrapping is litteraly an addiction for some people...

Some slide to clean this mess and get back on topic and power numbers :

And the FP perf improvement on a good ole ICC compiled Cinebench R15...

 

[maarten12100]

Nope, by Intel itself power consumption = TDP for any thermally significant period.

As in a cycle in which the average power consumption is the TDP number. In other words the bench runs for a long time hence thermally significant. You're throwing the equal sign into the equation on a scenario that is obviously a limit where the time t must go to infinite to have the power consumption and the TDP equal each other exactly.

Maybe I should not go at this with any math in mind and just state that the power consumption approaches the TDP when the load lasts for a longish time.

I already gave that example and I'm sure you know what boosting for short periods of time is. You get that to have the average consumption match the tdp the integral of the power as function of time on an interval t divided by that interval t must be the exact same value as the TDP number.

I can't think of a way to explain it better than this.

 

[mrmt]

I already gave that example and I'm sure you know what boosting for short periods of time is. You get that to have the average consumption match the tdp the integral of the power as function of time on an interval t divided by that interval t must be the exact same value as the TDP number.

You brought an example Broadwell. AFAIK Broadwell has Turbo, which is a breach of their TDP specs by definition.

 

[coercitiv]

30 seconds would be a thermally significant period of time by any standards.

You don't get to decide what period is thermally significant, physics does.

Based on the cooling solution and maximum power usage any period of time may or may not be thermally significant. For example even 5 seconds may be thermally significant for a high power chip without a heat sink, while 5 minutes may not be thermally significant for a very low power chip connected to a big heat sink.

You brought an example Broadwell. AFAIK Broadwell has Turbo, which is a breach of their TDP specs by definition.

Definitely not, for the reasons stated above.

Saying Intel's Turbo breaches TDP specs is the equivalent of admitting Intel CPUs use more power than advertised, as in a 15W BW-U is in fact a 25W CPU. Are you really prepared to promote such an idea in a thread about AMD products? This is mind boggling.

 

[IntelUser2000]

Guys, we are talking about companies, not individuals. Individuals usually do not change much, companies do, because the people working for them change.

Intel benchmarks were honest until couple of years ago. Right now they are shady at best. AMD is better than they used to, and obviously better than Intel now.

Nothing changes the fact that Core M is a big disappointment contrary to Intel's fancy powerpoint slide claims. Broadwell U is more efficient in terms of battery life, fanless isn't guaranteed, systems can be just as expensive as the U, multi-threading performance for most systems dwell in the range of high-end Atoms.

The slide here explicitely confirms what I said many times:

http://forums.anandtech.com/showpost.php?p=37578613&postcount=1754

Look at this!: http://images.anandtech.com/doci/8355/BDW-14nm.png

How do you believe this nonsense?! They state that its not Broadwell in general, but Broadwell-Y that got those specific improvements. They did not meet it.

Unfortunately while we don't have real review for Carrizo products, at least they've gone away from tactics like using paid employess in places like AT forums.

 

[CHADBOGA]

AMD is better than they used to, and obviously better than Intel now.

What on Earth are you basing that on?

 

[mikk]

He is an Intel hater, don't expect too much from him.

 

[mrmt]

You don't get to decide what period is thermally significant, physics does.

Yes, and that's why Intel talks about microseconds spike and not minutes spike. So for all intents and purposes, without turbo, TDP = power consumption. There is no way you can satisfy Intel TDP definition if you don't restrict thermally significant periods to a very, very brief, insignificant period.

Saying Intel's Turbo breaches TDP specs is the equivalent of admitting Intel CPUs use more power than advertised, as in a 15W BW-U is in fact a 25W CPU. Are you really prepared to promote such an idea in a thread about AMD products? This is mind boggling.

That's the problem with the fanboys, they can't read specs, just read what Intel has to say about Turbo:

http://www.intel.com/content/www/us...ology/turbo-boost/turbo-boost-technology.html

(...)

Note: Intel® Turbo Boost Technology 2.0 allows the processor to operate at a power level that is higher than its TDP configuration and data sheet specified power for short durations to maximize performance.

So yes, Intel is breaching its TDP definition when its processors are under Turbo, and that's why base clock is important, this is the guaranteed clocks your processor will run after burning the temperature buffer it has.

The issue with certain AMD processors is that they breach the TDP specs on normal conditions meaning they will throttle under standard conditions (default cooler, no overclock, running commercial applications), so you end up below base clock frequency, so there's something wrong with them. As AMD doesn't release the thermal datasheet of their CMT processors, AMD TDP numbers also become meaningless.

 

[Stennan]

All the vitriol in this thread is very sad actually and it makes it hard to follow the discussion about Carrizo. I tried reading through the 10 latest pages, but honestly I am starting to wonder if there is any point in getting back to these forums.

Has anyone spotted any 11-13 inch models at all? othervise the Lenovo Flex 3 seems to be the closest one if one wants something more portable. Hopefully the floodgates will open after Windows 10 launches... but some 15.6 inch models have already trickled out to stores in sweden so chances are that is all we are getting here

 

[mikk]

With their 12W-35W configurable TDP it is an even bigger mess now because OEMs usually don't give TDP details. Without reviews of the Notebook TDP and performance will be hard to know before purchase.

 

[coercitiv]

With their 12W-35W configurable TDP it is an even bigger mess now because OEMs usually don't give TDP details. Without reviews of the Notebook TDP and performance will be hard to know before purchase.

Please don't start this. Intel CPUs have had configurable TDP for quite a while now.

 

[coercitiv]

Has anyone spotted any 11-13 inch models at all? othervise the Lenovo Flex 3 seems to be the closest one if one wants something more portable. Hopefully the floodgates will open after Windows 10 launches... but some 15.6 inch models have already trickled out to stores in sweden so chances are that is all we are getting here

We still need to see what happens once Win 10 gets the green light, however I find the lack of available reviews on any Carrizo product a bit disheartening. The time frame is right though, Win 10 -> back to school should be ideal for a mobile platform launch.

 

[Abwx]

All the vitriol in this thread is very sad actually and it makes it hard to follow the discussion about Carrizo. I tried reading through the 10 latest pages, but honestly I am starting to wonder if there is any point in getting back to these forums.

Has anyone spotted any 11-13 inch models at all? othervise the Lenovo Flex 3 seems to be the closest one if one wants something more portable. Hopefully the floodgates will open after Windows 10 launches... but some 15.6 inch models have already trickled out to stores in sweden so chances are that is all we are getting here

Smaller model announced to this point is a Dell Inspiron 5000 14", but so far it still didnt pop up in stores, HP will undoubtly use Carrizo in their AMD based Elite Book 12"5 725 G2 to replace Kaveri although that will be at inflated prices.

And welcome to the forum...

 

[Abwx]

We still need to see what happens once Win 10 gets the green light, however I find the lack of available reviews on any Carrizo product a bit disheartening. The time frame is right though, Win 10 -> back to school should be ideal for a mobile platform launch.

It wont take long, Notebookcheck already received a Carrizo-L 7410, Carrizo will soon follow suite.

NBC is a german site and the HP 15"6 1080p is already at the top in the german pricer Geizhals notebook/tablet section, in a matter of a few days...

https://geizhals.de/?o=53