The Official Kaveri Review Thread (A10-7850K, etc)

[frozentundra123456]

The following graph clearly illustrates that A10-6700T 45W TDP Richland reaches Tcase Max after approximately 35 second and the processor reduces the power(gets slower) in order to keep it within the TDP rating of 45W and bellow Tcase Max temperature.

On the other hand, the Kaveri 45W TDP A8-7600 keeps the same power for the duration of the benchmark run. That indicates that the processor didnt reach the Tcase Max thermal limit and was able to use more power thus higher performance and higher power usage.

So people for the last time, TDP is not the max power usage. You never should compare the TDP rating with CPU power consumption.

Those graphs make it look like the CPU is either running at Max power or idle. That is, there is never an intermediate level, which is what I would expect if it throttled back to some lower intermediate clock when the Max temp was exceeded. Even after a fairly long period, the power usage never comes back up, as I would expect after the CPU cools off somewhat. Or does it just mean the job is finished when the power usage drops back to idle levels.

 

[Dresdenboy]

How is going back a decade relevant?

I just wanted to bring throttling into the discussion.

If a CPU has no throttling or other temperature related safety measurements, a TDP (influencing the thermal design) has to be max. power. With throttling or other mechanisms, TDP can be defined at a lower level than absolute max (e.g. ICC_max * VCC_max at max. t_junction).

 

[el etro]

(facepalm) so....answer then: why even put TDP rating on CPUs if power consumption may exceed it ?

You are not understanding his statement.

TDP is a estimative of the heat processor will generate.

 

[Ajay]

Two threads discussing this right now. Maybe a mod can merge them into one, on topic thread.

 

[Abwx]

Two threads discussing this right now. Maybe a mod can merge them into one, on topic thread.

Wonder why such a pollution is allowed since the debate
is not even about Kaveri s TDP....

 

[9enesis]

You are not understanding his statement.

TDP is a estimative of the heat processor will generate.

first of all, just to be clear , the CPU does not "generate" heat , what it does is: turns electrical energy into heat(energy) with COP = 1

 

[Enigmoid]

TDP is very relevant especially in Mobile. What we see with Kaveri is that at the same TDP and that means at the same THERMAL restrictions that applies in Mobile devices, it can have consistently higher performance than Richland. That means it can have higher base clocks and higher Turbo clocks. It also means that it will also be able to keep Turbo clocks for longer periods. All that translates in to extremely higher single thread performance than Richland, higher multithread performance and higher iGPU performance.

So to sum up,
I used the lower TDP 45W Desktop SKU to show that a 35W TDP Mobile Kaveri will have a substantial performance gain over the same TDP Richland because TDP matters the most in Mobile. I have never talked about power consumption but performance at the same TDP level.

Look at the actual power consumption of kaveri. Look at the clocks. Do you see a huge room for improvement?

Yes it does, and the further down you go in TDP, the less likely you are to be able to maintain boost clocks in full load (both CPU and iGPU).


No it doesn't, you've just been had by the marketing dep. Boost is the natural step you would take after being able to check&control how much power the chip uses at a certain time. It's a natural move and makes sense in any scenario where performance is needed.


When we are talking about a 45W TDP CPU the border between desktop and mobile is hardly there anymore. My mobile CPU has a 47W TDP.


Limiting TDP leads to dropping clocks in certain scenarios, and changing clocks changes efficiency. It also affects how the CPU performs in mobile.

Or alternatively maximum st performance at a given power target.

I'm saying that you cannot look at performance at boost clocks, then compare it to power at dropped clocks.

 

[Slomo4shO]

first of all, just to be clear , the CPU does not "generate" heat , what it does is: turns electrical energy into heat(energy) with COP = 1

I suppose your body doesn't "generate" heat either but instead turns chemical energy into kinetic energy... :whiste:

Also, since you can't seem to comprehend what "generate" implies, let me provide a definition:

generate

[jen-uh-reyt]
verb (used with object), gen·er·at·ed, gen·er·at·ing.
1. to bring into existence; cause to be; produce.
2. to create by a vital or natural process.
3. to create and distribute vitally and profusely
4. to reproduce; procreate.
5. to produce by a chemical process.

 

[9enesis]

.....turns chemical energy into kinetic energy... :whiste:

LOL....and into heat too for that matter


oh dear....ok...as far as my ENG goes:

1) nope
2) nope
3) nope
4) not sure
5) no

in order for something to create/generate heat , thus energy - it has to be an energy source itself CPU/APU is not one , it is a load (or a sink if you like).

 

[Erenhardt]

nm, not going to fuel this offtopic discussion

 

[Slomo4shO]

LOL....and into heat too for that matter

Heat is the transfer of thermal (kinetic and potential) energy between two systems...

Since the CPU increases its kinetic energy (as observed by the rise in temperature), the act of the CPU dissipating the thermal energy onto a secondary system, such as the atmosphere or heatsink, does imply that it is generating heat.

If you want to argue semantics, having an understanding of language becomes critical :whiste:

 

[9enesis]

TDP is a estimative of the heat processor will generate.

ok, to be serious. it is not estimative. It is something that had been calculated and measured.

there are two ways of doing that(as to my know):

1)calorimetry

2) knowing that CPU=heater thus power in = power out , therefore P=U*I (for DC circuits)

 

[9enesis]

Heat is the transfer of thermal (kinetic and potential) energy between two systems...

emmm.....yes and no(in this case) - heat itself is energy.

Since the CPU increases its kinetic energy (as observed by the rise in temperature), the act of the CPU dissipating the thermal energy onto a secondary system, such as the atmosphere or heatsink, does imply that it is generating heat.

nope. CPU is just a heater in this case.

not sure what you mean by "CPU increases its kinetic energy " sounds ridiculous:biggrin: does it start jumping around or what?

If you want to argue semantics, having an understanding of language becomes critical :whiste:

agreed. but i did not , i got tired to repeat the same thing over again , which no one seem to read or care

edited

 

[Slomo4shO]

emmm.....yes and no(in this case) - heat itself is energy.

Seems you need a crash course in elementary physics...

not sure what you mean by "CPU increases its kinetic energy " sounds ridiculous:biggrin: does it start jumping around or what?

Yes it does, pull out your atomic microscope or get an understanding of what temperature actually quantifies... BTW, your troll attempts are 2/10 at best

 

[9enesis]

Seems you need a crash course in elementary physics...

Yes it does, pull out your atomic microscope or get an understanding of what temperature actually quantifies... BTW, your troll attempts are 2/10 at best

Oh...i see...where it is going...big brother came to rescue his mate by nitpicking, right ?:biggrin:

i was not trolling ... yet besides i'm not physicist either , therefore do not punish me for that atomic microscope thing i beg you:biggrin:

i'm just an EE. never bothered with all those molecules and things you know:biggrin:

 

[Slomo4shO]

Oh...i see...where it is going...big brother came to rescue his mate by nitpicking, right ?:biggrin:

i was not trolling ... yet besides i'm not physicist either , therefore do not punish me for that atomic microscope thing i beg you:biggrin:

i'm just an EE. never bothered with all those molecules and things you know:biggrin:

There is much irony in this post considering that the entirety of your contributions in this tread have, thus far, been based on ill-conceived notions of concepts you have an absolutely abysmal understanding of... Good day Sir.

 

[9enesis]

LOL..phew ,how rude was that?! i'm terribly hurt :biggrin:

 

[inf64]

What is going on in this topic??

 

[9enesis]

it is all about TDP :biggrin: Kaveri has one , right?:whiste:

it's just that damn physicist came and spoiled all the fun:biggrin:

 

[SViscusi]

Those graphs make it look like the CPU is either running at Max power or idle. That is, there is never an intermediate level, which is what I would expect if it throttled back to some lower intermediate clock when the Max temp was exceeded. Even after a fairly long period, the power usage never comes back up, as I would expect after the CPU cools off somewhat. Or does it just mean the job is finished when the power usage drops back to idle levels.

Look at the red and orange lines where they go up to over 70, run for a period then drops down to 60 where it stays until it finishes up. There rest of the chips are fairly stable (excluding the i7) in terms of draw until the work is finished.

Whether that's proof that Richland initially goes above max until it hits a thermal limit then draws down to a stable level, I can't be sure.

 

[9enesis]

Whether that's proof that Richland initially goes above max until it hits a thermal limit then draws down to a stable level, I can't be sure.

the problem with that graph is that it is system power....there is no way to tell what the exact APU's power consumption is.

 

[Dresdenboy]

Just as an example, how flexible TDPs are being used nowadays:

And it makes sense. As I said above, TDP is just a design point. And heat transfer is a relatively slow process, giving some room to exceed TDP for short time intervals.

After throttling came multiple improvements up to fine grained power control. The more sophisticated this became (think of more sensors and dedicated µC), the smaller any safety margins could be made and the closer a chip will operate at it's limits.

 

[Enigmoid]

the problem with that graph is that it is system power....there is no way to tell what the exact APU's power consumption is.

Everything is the same in the AMD test platform with the exception of the RAM.

Look at the red and orange lines where they go up to over 70, run for a period then drops down to 60 where it stays until it finishes up. There rest of the chips are fairly stable (excluding the i7) in terms of draw until the work is finished.

Whether that's proof that Richland initially goes above max until it hits a thermal limit then draws down to a stable level, I can't be sure.

Doubt richland is hitting a thermal limit.

 

[frozentundra123456]

Look at the red and orange lines where they go up to over 70, run for a period then drops down to 60 where it stays until it finishes up. There rest of the chips are fairly stable (excluding the i7) in terms of draw until the work is finished.

Whether that's proof that Richland initially goes above max until it hits a thermal limit then draws down to a stable level, I can't be sure.

Ok, I think I see.

 

[Dresdenboy]

Look at the red and orange lines where they go up to over 70, run for a period then drops down to 60 where it stays until it finishes up. There rest of the chips are fairly stable (excluding the i7) in terms of draw until the work is finished.

Whether that's proof that Richland initially goes above max until it hits a thermal limit then draws down to a stable level, I can't be sure.

It likely goes over an estimated TDP minus a safety margin, since its measurement accuracy and number of temp sensors/circuits are not enough to go closer to the real max power limit.