TDP and core speed

[kprasad]

Hi,

I recently was reading a article about Intel's new processor . In that article, the author had explained about the TDP and how the core speeds are affected by it.

Link: http://www.anandtech.com/cpuch...howdoc.aspx?i=3634&p=4

The bottom line is, more core/cpu means slower performance (if all the cores are utilized at a given time).

Which means, for example 2Ghz dual core is faster that 2Ghz quad core (assuming both have same TDP rating) ?

Am I correct or am I missing something.

Thanks,
Prasad.

 

[VirtualLarry]

I think that you are missing something. A 2Ghz quad-core, on the same architecture as a 2Ghz dual-core, will always be faster, for those loads that are multi-threaded. On single-threaded tasks it will be the same.

I think what that article was talking about was the TDP "envelope", which on Nehalem CPUs, the CPU's PCU (power control unit) seeks not to exceed.

The thing is, four cores takes twice as much power as two cores. Now, Nehalem can shut down cores that aren't in use, due to it's new power gating transistors.

So, if two cores only takes half as much power as four cores, and there is a limited TDP envelope, then that means that with two cores, you can run them at a faster speed before you exceed TDP. That's what the new "Turbo mode" does.

So you not only have a fast quad-core, but you have in essence, a much faster dual-core CPU as well.

 

[cusideabelincoln]

What was previously said, but I'll attempt to show you where you went wrong.

You said we were assuming both a dual core and a quad core have the same TDP rating and are both running at 2 GHz. We'll ignore the reality of this situation, and it also seems to be implied these processors are built on the same architecture - the quad core just has two extra cores on its die. Well from those assumptions, you cannot say the dual core will be faster. The quad core will either be just as fast or faster when running at the same frequency as the dual core.

However with Core i5 and i7, the clock speed fluctuates with Turbo Mode as per VirtualLarry's explanation. It is not constant, hence why when only dual cores are in use they are clocked higher than when quad cores are in use, and thus dual threaded applications enjoy the benefit of the higher clocked dual cores. Meanwhile, a quad threaded application can enjoy the benefit of having more cores, so performance does not suffer.

 

[kprasad]

Thanks for the reply VirtualLarry / cusideabelincoln.

My understanding on that article is , all the cores in a cpu cannot run at max Ghz (at the same given time), since the power has to be shared between the cores (due to TDP limit which is fixed).
The single core can run at max. GHz only when other cores are idle(the power is switched off by the gating transistors, so all the power is utilized by that one core)

Also, Now I have a doubt on my understanding on speed calculation on multi-core CPU's
If a vendor says his quad core CPU runs at 4Ghz, does it mean "4 cores X 4 Ghz" or "4 cores X 1 Ghz"

Thanks,
Prasad

 

[dguy6789]

All cores run at maximum speed. A 4Ghz quad core has all four cores each running at 4Ghz, not each running at 1Ghz. It's possible that the turbo boost function has caused the misunderstanding. That's just a feature on the Core i5 and i7 that overclocks one of the cores when it's the only one running software.

 

[Idontcare]

Originally posted by: kprasad
Hi,

Prasad.

Welcome to the forums Prasad!

Originally posted by: kprasad
The bottom line is, more core/cpu means slower performance (if all the cores are utilized at a given time).

more cores/cpus can mean slower relative performance if we require certain boundary conditions to be true. But what you say there in that quote above is not exhaustively robust as we have examples where it is not true.

Originally posted by: kprasad
Which means, for example 2Ghz dual core is faster that 2Ghz quad core (assuming both have same TDP rating) ?

Remember that TDP rating is not the same as power-consumption, TDP rating can be used as a rule of thumb for estimating what the practical upper-limits of power-consumption could be for a given processor, but saying an i7 has a TDP of 130W for example does not mean we should expect an i7 to actual consume 130W when running fully loaded.

Here is some data to mull over when considering actual practical power-consumption at full load versus the TDP rating for the chips of interest:
http://www.lostcircuits.com/ma...1&limit=1&limitstart=8

Now let's amend your line of thought to be something a little more interesting:

Originally posted by: kprasad as amended by IDC
Which means, for example 2Ghz dual core is faster that 2Ghz quad core (assuming both are operating at the same power-consumption level under load) ?

Now then again this statement still doesn't hold true (2GHz dual-core won't be faster than 2GHz quad) because we've already made it a boundary condition that both chips are operating at 2GHz (meaning the quad core is more power-efficient per core).

So we need another restriction of some kind to be invoked that would require the 2GHz quad to operate at a lower clockspeed than the 2GHz dual-core chip before we can get to the point that the 2GHz quad is effectively running so slow that it's performance is less than that of the dual-core.

 

[kprasad]

@dguy6789: Thanks for clearing my doubt.

@Idontcare :

Originally posted by: Idontcare
Remember that TDP rating is not the same as power-consumption, TDP rating can be used as a rule of thumb for estimating what the practical upper-limits of power-consumption could be for a given processor, but saying an i7 has a TDP of 130W for example does not mean we should expect an i7 to actual consume 130W when running fully loaded.

Which means, when all the cores in a cpu run at full speed, it will not cross TDP but consumes more power ?

 

[Schmide]

Originally posted by: Idontcare

Remember that TDP rating is not the same as power-consumption, TDP rating can be used as a rule of thumb for estimating what the practical upper-limits of power-consumption could be for a given processor, but saying an i7 has a TDP of 130W for example does not mean we should expect an i7 to actual consume 130W when running fully loaded.

The I7 also has its un-core that can consume up to 45w in addition to the chip consumption (90w); thus the need for advanced power management. In reality the I7 is one processor that is regulated to stay within its TDP but can easily exceed it.

 

[Idontcare]

Originally posted by: Schmide

Originally posted by: Idontcare

Remember that TDP rating is not the same as power-consumption, TDP rating can be used as a rule of thumb for estimating what the practical upper-limits of power-consumption could be for a given processor, but saying an i7 has a TDP of 130W for example does not mean we should expect an i7 to actual consume 130W when running fully loaded.

The I7 also has its un-core that can consume up to 45w in addition to the chip consumption (90w); thus the need for advanced power management. In reality the I7 is one processor that is regulated to stay within its TDP but can easily exceed it.

Earlier I too subscribed to that line of thinking as I was educated about i7 power-consumption by this lostcircuits article - Core i7 Power Plays - but then someone was kind enough to take the time to lead me to the water thru some pm's and they capped it off with this:

Editor's Corner: Overclocking Core i7

Intel has added an Overspeed Protection to its Core i7 processors, keeping them from exceeding 130W or 100A. The company says it?s a safeguard in environments where that much power or current could result in a problem. The workaround is a BIOS switch. Overclockers will want to disable it before trying their hand at faster frequencies.

http://www.tomshardware.com/re...king-core-i7,2063.html

Checkout the tests they did in that link, the data is there to confirm that overspeed protection kicks in and not only do you lose your "turbo boost" multipliers above the stock multipliers but you also can end up having your cpu multiplier downgraded to a value below its stock value, whatever the PCU needs to do in order to prevent your CPU from exceeding its TDP rating.

The i7 (and all nehalem derivatives tmk) cannot exceed its rated TDP unless you intentionally defeat the over-current limits thru the bios.

You may briefly coerce it to spike above its TDP as the PCU circuitry of course needs some non-zero amount of time to respond after confirming the elevated power-consumption is not a spike but rather is intended to be a sustained load level.

Ever since I saw that data from the THG efforts to exceed TDP on an i7 I've been a changed man. The PCU is pretty cool if you expand the application to the mobile space. Shove a lynnfield into a laptop and set the over current limits such that it simply won't allow itself to clock high enough to consume more than the microcode-set TDP limit.

A self-governing self-regulating mobile chip that basically "bins itself" for operating at the highest clockspeed it can while not exceeding the thermal design limits of the laptop itself. How awesome is that?

 

[Schmide]

Yes - agree with all the above when the limits are disabled.

However from the original quote you gave, it inferred that the TDP would be much less. I was basically saying the I7 will reach its TDP and in fact dance around it.

 

[Idontcare]

Originally posted by: Schmide
Yes - agree with all the above when the limits are disabled.

However from the original quote you gave, it inferred that the TDP would be much less. I was basically saying the I7 will reach its TDP and in fact dance around it.

They had to push that 920 up to 3.8GHz and up the Vcc to 1.475V in order to get it to the point of being TDP limited.

I don't expect too many people are experiencing 130W power consumption with their 130W TDP rated i7's operating at stock...

 

[IntelUser2000]

On the Xeon 5500-based server platforms they have a feature called "Power Cap". Basically, you can set bunch of your racks to a set amount of power and the CPU and the platform will adjust itself to stay below the set limit.

 

[Idontcare]

^^^
| | |

The guy who was kind enough to hold my hand thru many pm's till I got to the point of starting to realize just some of the potential of the circuitry involved in the TDP self-governance.

 

[Schmide]

Originally posted by: Idontcare

Originally posted by: Schmide
Yes - agree with all the above when the limits are disabled.

However from the original quote you gave, it inferred that the TDP would be much less. I was basically saying the I7 will reach its TDP and in fact dance around it.

They had to push that 920 up to 3.8GHz and up the Vcc to 1.475V in order to get it to the point of being TDP limited.

I don't expect too many people are experiencing 130W power consumption with their 130W TDP rated i7's operating at stock...

You do not know how the chip works and what effect increasing the voltage has on the equations that regulate the chip. For all you know you were pushing out 200w. The fact is stock chips do get near 130w total chip usage. The lost circuits article showed that.

 

[alyarb]

it depends on how you are measuring. If you put an ammeter to the 8-pin CPU power connection, you will find your load watts to be quite a bit shy of TDP, however, the uncore does not receive power from this line. the uncore is powered by the 24-pin ATX plug, but there are a lot of other board components that take power off that line too, so it's especially difficult with i5/i7 to compare measured amps against TDP.

http://www.xbitlabs.com/articl...ay/core-i7-870_14.html

 

[kprasad]

Wow thanks for the replies guys.
I'm going to need sometime to go thru the replies and links.

Thanks,
Prasad.