"7 Watt" Ivy Bridge my arse!

[NTMBK]

Intel are resorting to marketing bamboozling in their presentations, again. http://arstechnica.com/gadgets/2013/01/power-saving-through-marketing-intels-7-watt-ivy-bridge-cpus/

7W isn't the TDP, its the "SDP". Lowest TDP is 10W on a 2C2T Pentium with no turbo at 1.1GHz.

 

[podspi]

I'd love to know how they calculate that number and what the implication is for product designs. Does this mean the cpu could seriously overheat if someone runs linx on their tablet? Or will it preemptively throttle to enforce the SDP?

 

[Charles Kozierok]

LOL. Didn't AMD try to play a similar game a few years ago?

Edit: Ah yes, here it is.

 

[CHADBOGA]

SDP > ACP :awe:

 

[Phynaz]

LOL. Didn't AMD try to play a similar game a few years ago?

Yep...here's the white paper.

http://www.amd.com/us/Documents/43761D-ACP_PowerConsumption.pdf

 

[Haserath]

SDP makes much more sense on these products. You won't be using them for heavy duty tasks, but max power definitely needs to stay as a marker for products...

I like having both measurements. Makes for easier comparison on power consumed.

 

[NTMBK]

Its not that they have a new metric that bothers me- if they'd introduced it with a 5 minute tangent on what it is and how its calculated, then presented it alongside the TDP, then yay. More information is a good thing.

But they didn't. They just said "7W Ivy Bridge" without saying that they were referring to an entirely different measurement than the one which gives us "17W Ivy Bridge".

EDIT: And yes, AMD pulls scummy marketing tricks from time to time too. But they need to be called out for what they are, whoever does it.

 

[Charles Kozierok]

Not only that, but as one of the commenters on the article said, "SDP" is defined in an entirely arbitrary way. TDP at least has some actual meaning that relates to something tangible.

 

[Homeles]

Not only that, but as one of the commenters on the article said, "SDP" is defined in an entirely arbitrary way. TDP at least has some actual meaning that relates to something tangible.

TDP is still a pretty arbitrary metric. Point in case: 7970 and 7970 GE. Both have the same TDP, but power draw is inarguably different.

 

[Ferzerp]

TDP is still a pretty arbitrary metric.

Especially on AMD's desktop processors where, as IDC has shown, it is thoroughly incorrect.

 

[Haserath]

TDP is still a pretty arbitrary metric. Point in case: 7970 and 7970 GE. Both have the same TDP, but power draw is inarguably different.

When either one hits its TDP, it throttles.

SDP is just general power use. They do need to give more info on it, but if it stays the same per generation, at least it should show what improvement there is.

 

[Charles Kozierok]

Sure, TDP isn't perfect. But SDP strikes me as a completely manufactured metric.

 

[piesquared]

SDP makes much more sense on these products. You won't be using them for heavy duty tasks, but max power definitely needs to stay as a marker for products...

I like having both measurements. Makes for easier comparison on power consumed.

Do you think ACP makes sense on servers?

 

[inf64]

Wow somehow this became AMD topic even though it's about "7W my arse" Ivy Bridge 😀. Some posters still amaze me 😉.

 

[piesquared]

Especially on AMD's desktop processors where, as IDC has shown, it is thoroughly incorrect.

Too bad IDC wasn't around to 'investigate' power draw on Nehalem processors when the uncore wasn't included in the TDP. Well, I guess if it didn't matter then, it doesn't matter now.

 

[Haserath]

Do you think ACP makes sense on servers?

They design the cooling and power system for TDP, but measure average cost with ACP.

TDP usually has nothing to do with actual power use in real world scenarios.

 

[Homeles]

When either one hits its TDP, it throttles.

Either way, it's a terrible indicator for power usage.

Then you have to remember that each company has its own idea for how to calculate TDP.

 

[Idontcare]

Too bad IDC wasn't around to 'investigate' power draw on Nehalem processors when the uncore wasn't included in the TDP. Well, I guess if it didn't matter then, it doesn't matter now.

I never owned a Nehalem, nor a Westmere for that matter.

I went from 65nm (Cedar Mill, QX6700, and Q6600) to 32nm (Sandy Bridge, Piledriver) to 22nm (Ivy Bridge).

What I do remember of Nehalem was that they bifurcated the power lines so traditional methods of measuring power consumption of whatever it was that was occupying the socket (be it nehalem or its successors) were no longer valid.

It was because of this reality that I moved on to my current method of deconvolving power data from "at the wall" measurements which unfortunately entails generating ~900 measurements from different combinations of voltage, temperature, and clockspeed :|

It is a PITA, but it can be done. I've no doubt Nehalem wasn't all that great when it came to power-consumption at the socket level.

 

[piesquared]

I never owned a Nehalem, nor a Westmere for that matter.

I went from 65nm (Cedar Mill, QX6700, and Q6600) to 32nm (Sandy Bridge, Piledriver) to 22nm (Ivy Bridge).

What I do remember of Nehalem was that they bifurcated the power lines so traditional methods of measuring power consumption of whatever it was that was occupying the socket (be it nehalem or its successors) were no longer valid.

It was because of this reality that I moved on to my current method of deconvolving power data from "at the wall" measurements which unfortunately entails generating ~900 measurements from different combinations of voltage, temperature, and clockspeed :|

It is a PITA, but it can be done. I've no doubt Nehalem wasn't all that great when it came to power-consumption at the socket level.

Yes, and the power draw was for all practical purposes, ignored by reviewers and forum goers alike (other than Lostcircuits of course, and Michael Shuette was berated for it by the horde. Mob rule mentality and all). Basically, intel was given another free pass. I guess intel has cleaned that mess up now and we can all start measuring in a way that produces the worst outcome for AMD, again.

So now we have a presumed 7W processor, that by normal standards is a minimum 10W processor. Using the same metrics that AMD used for ACP in servers, and was of course beaten to death for it. Go figure.

 

[Idontcare]

So now we have a presumed 7W processor, that by normal standards is a minimum 10W processor. Using the same metrics that AMD used for ACP in servers, and was of course beaten to death for it. Go figure.

I believe it really will be a 7W processor because they dramatically reduced TJmax for SDP versus TDP, that will cut the leakage power losses by a significant amount.

(can't find the specific numbers right now but I think I read somewhere the TJmax for SDP settings was around 50C or 55C)

 

[IntelUser2000]

I believe it really will be a 7W processor because they dramatically reduced TJmax for SDP versus TDP, that will cut the leakage power losses by a significant amount.

(can't find the specific numbers right now but I think I read somewhere the TJmax for SDP settings was around 50C or 55C)

Tjmax goes from 100C to 85C.

So that does indicate there is a relevance.

If you ask me, I would guess they are simply disabling Turbo Boost on both CPU and the GPU core to achieve it. Remember since Turbo Boost 2.0 on Sandy Bridge, Turbo is made to use the "thermal headroom" that isn't normally used by typical applications.

On Penryn chips, 25W chip may use maybe ~20W in most applications. Turbo Boost would mean it would use the 5W difference for that application for enhanced performance.

I think that also explains the Intel claim that ULT GT3 Haswell parts achieve Ivy Bridge GT2 17W 3D performance at half the TDP. Since for 3D, the iGPU takes up bigger portion of the power, it needs to cut MORE than half to achieve half the TDP.

Going from a ~1GHz sustained 16 EU to 350-400MHz Turbo-less 40 EU would be roughly equal in performance, and CPU part savings are achieved by simply running at base frequencies.

 

[ShintaiDK]

SDP is simply TDP at a lower temperature. However the tjmax is still 105C. And its up to the OEM to keep it below 80C, unless they want extra power consumption.

 

[tweakboy]

It's a budget chip. Not for premium enthusiast.

 

[Charles Kozierok]

SDP is simply TDP at a lower temperature. They could just as well have released them at 80C tjmax and be done with it. However the tjmax is still 105C. And its up to the OEM to keep it below 80C, unless they want extra power consumption.

Which is "simply" marketing bullshit.

Heck, why not define Tjmax at 50C and call it 3W or whatever? Who knows, maybe they'll do that next time.

They make it difficult enough to compare processors. Now we need every core or core revision having incompatible power ratings?

 

[ShintaiDK]

Which is "simply" marketing bullshit.

Heck, why not define Tjmax at 50C and call it 3W or whatever? Who knows, maybe they'll do that next time.

They make it difficult enough to compare processors. Now we need every core or core revision having incompatible power ratings?

The user will always get full performance. Its simply OEM information for their designs. Its abit like a rerun of Tcase that the majority of people dont have a clue what is and often confuse with tjmax.

So seems we got some drama for the sake of drama.