"7 Watt" Ivy Bridge my arse!

[ShintaiDK]

More information > less information. I'm more than happy to sort through the garbage to find the valuable bits.

I dont have the patience for it. I mainly stick with credible sources and take it from there.

 

[Phynaz]

SDP = Solitaire Design Power (power draw while playing solitaire)

Which is the average load for most peoples' computers, so I guess Intel was correct after all.

 

[Haserath]

I guess SDP can be considered more like an active idle than a semi-load.

 

[Homeles]

I dont have the patience for it. I mainly stick with credible sources and take it from there.

Fair enough. I should probably cut down on the hardware hobby and focus on other things...

 

[ShintaiDK]

Fair enough. I should probably cut down on the hardware hobby and focus on other things...

I dont see how those 2 things are really related. Its more a study in journalism, or the lack of the same. Sensationalism sells, truth or no truth.

 

[NTMBK]

I dont see how those 2 things are really related. Its more a study in journalism, or the lack of the same. Sensationalism sells, truth or no truth.

Yeah, I remember that time he was totally wrong about Facebook starting to use ARM servers. OH NO WAIT

 

[ShintaiDK]

Yeah, I remember that time he was totally wrong about Facebook starting to use ARM servers. OH NO WAIT

I can also make 20 wild predictions. And when one of them gets true I gonna ignore all the rest and slam you in the head over and over again with the one I had right so you never doubt my greatness again.

See the point?

And the ARM servers?

“Facebook continuously evaluates and helps develop new technologies we believe will improve the performance, efficiency or reliability of our infrastructure. However, we have no plans to deploy ARM servers in our Prineville, Oregon data center.”

 

[NTMBK]

I can also make 20 wild predictions. And when one of them gets true I gonna ignore all the rest and slam you in the head over and over again with the one I had right so you never doubt my greatness again.

See the point?

And the ARM servers?

That Facebook quote very specifically talks about only one data centre. How many datacentres do you think they have? And why do you think that quote only specifies a single one?

As for the predictions/news/rumours,

-first to call it that the 680 would be clearly beat the 7970 in performance/watt
-first to say that the 28nm Bobcat was cancelled
-first to tell us about Haswell's on package DRAM
-first to call it that AMD was out of Apple and nVidia was back in
-first to say that the big Kepler wasn't going to make it into product in the first round

That's not a bad hit rate.

 

[ShintaiDK]

LOL, keep believing then. I am sure his bank account will enjoy it.

Shame we dont mention the regular crap he releases.

http://semiaccurate.com/2011/05/05/apple-dumps-intel-from-laptop-lines/
http://semiaccurate.com/2012/05/08/nvidias-five-new-keplers-raise-a-red-flag/

 

[NTMBK]

LOL, keep believing then. I am sure his bank account will enjoy it.

Shame we dont mention the regular crap he releases.

http://semiaccurate.com/2011/05/05/apple-dumps-intel-from-laptop-lines/
http://semiaccurate.com/2012/05/08/nvidias-five-new-keplers-raise-a-red-flag/

That article about Intel being out of Apple laptops actually seems more plausible as time passes. Apple just made a world-beating ARM core, and they have been steadily ramping up their level of chip prowess (from straight ARM off the shelf design, to ARM core and Apple uncore, to Apple core and uncore). They did this without anyone realising it was coming- even Anand thought the A6 had a Cortex A15 at first. Apple suddenly turning round and announcing that the next Macbook Air will ship with an in-house ARM design is far from implausible. Not to mention that the story specifically states that the time frame for this change is still in the future, so we can't call it yet either way.

 

[ShintaiDK]

That article about Intel being out of Apple laptops actually seems more plausible as time passes. Apple just made a world-beating ARM core, and they have been steadily ramping up their level of chip prowess (from straight ARM off the shelf design, to ARM core and Apple uncore, to Apple core and uncore). They did this without anyone realising it was coming- even Anand thought the A6 had a Cortex A15 at first. Apple suddenly turning round and announcing that the next Macbook Air will ship with an in-house ARM design is far from implausible. Not to mention that the story specifically states that the time frame for this change is still in the future, so we can't call it yet either way.

Ahahahahaa....

 

[cbn]

An article on the Intel North Cape Reference design (using Haswell Processor):

http://www.theverge.com/2013/1/7/38...ands-on#ooid=41MWM2ODojT509oe7XGmcrOPZvdeqQvb

If you scroll down to the bottom of the article you will see a video. At 34 seconds into the video something called "Smart frame" is described. Apparently when the tablet is detached from the laptop a black bezel forms around the thin border screen lowering both screen size and resolution. (Not sure how much of a difference this will make on GPU loads, but it is interesting to note)

P.S. That reference design really does look nice!!

 

[cbn]

For those wondering what Core @ 800 Mhz (vs. 1.7 Ghz A15) will do in some benchmarks check out the Anandtech Acer C7 review:

http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

Granted the cores are Sandy Bridge, not Ivy Bridge, so there will be some IPC difference.

 

[IntelUser2000]

For those wondering what Core @ 800 Mhz (vs. 1.7 Ghz A15) will do in some benchmarks check out the Anandtech Acer C7 review:

http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

Granted the cores are Sandy Bridge, not Ivy Bridge, so there will be some IPC difference.

I think there will be an additional ~10% difference being a Celeron with only 2MB L3.

 

[Exophase]

For those wondering what Core @ 800 Mhz (vs. 1.7 Ghz A15) will do in some benchmarks check out the Anandtech Acer C7 review:

http://www.anandtech.com/show/6476/acer-c7-chromebook-review/3

Granted the cores are Sandy Bridge, not Ivy Bridge, so there will be some IPC difference.

I wish for absolute proof that this is really 800MHz and not 1.1GHz (I left something about this in the comments, it's near the bottom currently)

 

[Ryaxnb]

Okay, update from Intel! Have a read of this:



http://www.theverge.com/2013/1/9/3856050/intel-candid-explains-misleading-7w-ivy-bridge-marketing (Emphasis mine)

So yeah, big surprise, they sacrifice a lot of performance to get it to 7W.

From MS:
If you want to run Windows 7 on your PC, here's what it takes:

• 1 gigahertz (GHz) or faster 32-bit (x86) or 64-bit (x64) processor

So this core i7 can't run windows 7?

:whiste::biggrin:
linky: http://windows.microsoft.com/systemrequirements

 

[FwFred]

From MS:
If you want to run Windows 7 on your PC, here's what it takes:

• 1 gigahertz (GHz) or faster 32-bit (x86) or 64-bit (x64) processor

So this core i7 can't run windows 7?

:whiste::biggrin:
linky: http://windows.microsoft.com/systemrequirements

I know you said this tounge-in-cheek, so not really directing this at you

I'd imagine that due to turbo, most interactions with the Windows OS would be > 1GHz. Also let's be real, that 1 GHz guideline was not set for an IPC monster like Ivy Bridge.

I'm still not sure I've found where the reliable source that says you will get 800MHz base clock if setting the PL1 value to 7W. Does anyone have a link?

Further reading: http://arstechnica.com/gadgets/2013/01/the-technical-details-behind-intels-7-watt-ivy-bridge-cpus/

 

[IntelUser2000]

I wish for absolute proof that this is really 800MHz and not 1.1GHz (I left something about this in the comments, it's near the bottom currently)

I searched about it.

The comment is saying that because of low CPU load, it rarely clocks at its full frequency.

Relating back to SDP, if Intel ever decides to make a Core-based Android or future Chromebook, it won't ever have to reach TDP figure.

 

[Exophase]

I searched about it.

The comment is saying that because of low CPU load, it rarely clocks at its full frequency.

That isn't what he said in his comment. He said that he tested it under lots of conditions and couldn't see it go beyond 800MHz, therefore was confident (his exact words) that that was the maximum frequency. Not that it rarely goes over it, but that it never goes over it because it can't.

More to the point, the claim is made in the article that the benchmark scores were attained with the CPU capped at 800MHz. Given other Sunspider scores with similar browsers (other Chrome OS devices) I find it unlikely that it was running at 800MHz and not 1.1GHz. This actually is relevant, since people are drawing perf/MHz comparisons from it.

Relating back to SDP, if Intel ever decides to make a Core-based Android or future Chromebook, it won't ever have to reach TDP figure.

I don't know why you think an Android or Chromebook device can't ever experience full load, even if only for short times. Just because these OSes can run on relatively weak hardware doesn't mean there are not times when the software wants to get something done ASAP.

 

[cbn]

if Intel ever decides to make a Core-based Android

I hope they do. (re: I don't think all my x86 devices would need to have Windows)

 

[IntelUser2000]

That isn't what he said in his comment. He said that he tested it under lots of conditions and couldn't see it go beyond 800MHz, therefore was confident (his exact words) that that was the maximum frequency. Not that it rarely goes over it, but that it never goes over it because it can't.

http://emkey1.blogspot.ca/2012/11/low-tech-acer-chromebook-unboxing-review.html

Read the comment section below. Also I find it strange that in Sunspider, how a 1.3GHz Celeron 867-using Chromebook gets 400ms, when that benchmark scales nearly linearly with clock frequencies. If the Acer C7 is really running at 800MHz maximum, it should be getting close to 650ms, not ~540ms.

I don't know why you think an Android or Chromebook device can't ever experience full load, even if only for short times. Just because these OSes can run on relatively weak hardware doesn't mean there are not times when the software wants to get something done ASAP.

TDP isn't full power either. For mobile Core 2, the X9100 has 44W TDP, but it has DC specification of 1.325V in Enhanced Dynamic Acceleration(early form of Turbo) and 1.275V in High Frequency Mode, with 59A Icc. That's well over 44W.

For Ivy Bridge, the 17W chip has Max VID of 1.2V, and ICCmax of 33A. It also states Thermal design Icc, which is at 15.8A.

But both chips have advanced power management, which doesn't allow such figures to be sustained, making it a practical maximum for system design purposes and such.

For Sandy Bridge and Ivy Bridge, it specifically states Power greater than even peak power during thermal response time can happen for an extremely brief under 10ms.

So for TDP values, it only counts when its done under some sustained duration, which I pointed out few pages ago. If you have no application that is capable of reaching TDP values(the practical max power set by the CPU), but only does SDP, the latter would be accurate as well.

 

[Exophase]

http://emkey1.blogspot.ca/2012/11/low-tech-acer-chromebook-unboxing-review.html

Read the comment section below. Also I find it strange that in Sunspider, how a 1.3GHz Celeron 867-using Chromebook gets 400ms, when that benchmark scales nearly linearly with clock frequencies. If the Acer C7 is really running at 800MHz maximum, it should be getting close to 650ms, not ~540ms.

Okay, but did you read the comments on the AT review? Of course I wasn't going to know you were referring to a comment someone made on another review, but in this one the reviewer said outright that he's confident that's the maximum clock. I said exactly what you did, that the Sunspider scores don't add up for 800MHz.

Like I said, this is a big deal and I wish Jason Inofuentes would take my comment more seriously because if the tests were performed at 1.1GHz instead of the 800MHz he says his conclusions are drawing a pretty unfair uarch comparison. To be frank I wasn't that impressed with some other things he said in the review either, like how he was certain that if vsync wasn't on the Intel IGP would gain a much bigger lead, and that he believes the pre-dev browser he ran wasn't even using hardware acceleration at all..

TDP isn't full power either. For mobile Core 2, the X9100 has 44W TDP, but it has DC specification of 1.325V in Enhanced Dynamic Acceleration(early form of Turbo) and 1.275V in High Frequency Mode, with 59A Icc. That's well over 44W.

It's supposed to be full power over a sustained period, which is what I meant. Although I said "short times" I wasn't very clear on just how short I was thinking more several seconds, not a few milliseconds.

I still think it's highly naive to think any use case that's even remotely removed from an embedded application can't ever see full load. It just takes one program behaving badly for a while. I'm sure a website with a bunch of flash ads can do it.

This is even more true on Android where you can run any old user app, and some of them actually can be properly designed to perform compute workloads that aren't immediately over even on a relatively fast computer.

 

[IntelUser2000]

Like I said, this is a big deal and I wish Jason Inofuentes would take my comment more seriously because if the tests were performed at 1.1GHz instead of the 800MHz he says his conclusions are drawing a pretty unfair uarch comparison.

It's weird because SpecCPU scores wouldn't match up either. 1.7GHz A15 should get around ~900 or so in SpecInt2K while an Ivy Bridge Core at 800MHz would get 1000 or so. But being a Celeron, we'd end up maybe 10% less at 900.

I still think it's highly naive to think any use case that's even remotely removed from an embedded application can't ever see full load. It just takes one program behaving badly for a while. I'm sure a website with a bunch of flash ads can do it.

I don't think flash ads would make it reach TDP, while it would still count as being on load.

I have a software power meter on my Windows 8 Ultrabook tested with flash videos, and while it would peak to ~14W for the entire system and reach 1.6-1.9GHz frequency, it would drop down to 12.5W and 800MHz-1GHz in 20 seconds or so. In Android you are talking about a system doing even more optimization and thus hardware acceleration.

Back with Pentium 4, 3rd party stress applications like WPrime would make it go shoot above its TDP, which is why the whole debate about whether TDP was accurate was brought up in the first place. But with Core 2, anything outside of Linpack wouldn't even reach the TDP mark.

If they take SDP as something of a "cTDPdown-down" with 800MHz as the base frequency and 7W being the real power cap, that would be fine. They are not that inflexible in their metrics. On the Windows power manager setting, my device is set at 15W(rather than being cTDPdown=13W, Base=17W). That means manufacturers get some leeway on how they want it with firmware/BIOS or something.

 

[Exophase]

It's weird because SpecCPU scores wouldn't match up either. 1.7GHz A15 should get around ~900 or so in SpecInt2K while an Ivy Bridge Core at 800MHz would get 1000 or so. But being a Celeron, we'd end up maybe 10% less at 900.

That's because Intel's SpecCPU scores get a ~20% boost from being compiled with ICC, remember?

I don't think flash ads would make it reach TDP, while it would still count as being on load.

I have a software power meter on my Windows 8 Ultrabook tested with flash videos, and while it would peak to ~14W for the entire system and reach 1.6-1.9GHz frequency, it would drop down to 12.5W and 800MHz-1GHz in 20 seconds or so.

Plain videos or more obnoxious flash stuff? That was really just one example.

Not really going to argue at length for browser-only stuff since Chromebooks are pretty niche, although there's definitely a bigger push to get more web content. For instance, Javascript emulators are getting bigger these days (there's an N64 one that I'm sure will happily peg at least a core, along with give the IGP at least some load)

In Android you are talking about a system doing even more optimization and thus hardware acceleration.

How can this apply to a system with many thousands of apps that can do pretty much anything? Hell, if people really want to make it hit full load for a while they can just run a benchmark. There are plenty of those.

Back with Pentium 4, 3rd party stress applications like WPrime would make it go shoot above its TDP, which is why the whole debate about whether TDP was accurate was brought up in the first place. But with Core 2, anything outside of Linpack wouldn't even reach the TDP mark.

It helps that the processors actually do properly throttle.

If they take SDP as something of a "cTDPdown-down" with 800MHz as the base frequency and 7W being the real power cap, that would be fine. They are not that inflexible in their metrics. On the Windows power manager setting, my device is set at 15W(rather than being cTDPdown=13W, Base=17W). That means manufacturers get some leeway on how they want it with firmware/BIOS or something.

Sure, if there's a hardware regulated power cap it won't go above that.. I'm kind of feeling lost, what were we talking about again? >_> I thought it was that you can comfortably spec a system below its TDP (configurable or otherwise) if it runs Android or Chrome OS because you can be confident no one will ever want to use it under heavy load. If that's the case why even bother with a relatively expensive ULV Core series derived processor in them?

 

[Nemesis 1]

Bottom line . It doesn't matter what intel rates these at. If they do not have battery life the product has no life.