Intel Broadwell Thread

[witeken]

This is unacceptable.

If Core M is the benchmark for being unacceptable, then what vocabulary do you use to describe the iPad, all Qualcomm SoCs ever produced, etc.?

BTW, what are you comparing with? 15W Haswell-U? How much of a difference does it make in real world usage? And lastly, why is it unacceptable that a 3.5W device can't compete with real laptops?

 

[dahorns]

If Core M is the benchmark for being unacceptable, then what vocabulary do you use to describe the iPad, all Qualcomm SoCs ever produced, etc.?

BTW, what are you comparing with? 15W Haswell-U? How much of a difference does it make in real world usage? And lastly, why is it unacceptable that a 3.5W device can't compete with real laptops?

It is unacceptable because it is in an ultrabook form factor, with a fan, and gets nowhere near the advertised performance promised with Broadwell-M.

I whole heartedly agree. It is unacceptable. What I don't know is who to blame, Lenovo or Intel. I'd think Lenovo only because Intel would be flirting with securities fraud based on its representation of how Broadwell-M would perform. I mean, the numbers are really shockingly bad. And it sounds like real world use is worse.

 

[witeken]

I haven't read the review yet, so I can't comment.

 

[Space69]

Again the point is that GB isn't trying to get the best algo/implementation for each of the test. And because of one of the subtests not being fully tuned for your preferred architecture, should all of GB be dismissed?

IMO, yes! I'm interested in knowing how the architecture performs and not the performance of some suboptimal implementations of various algorithms.

I would prefer a modular benchmark system with various opensource algorithms with unknown datasets and with atleast 40+ realtime modulating parameters. Each hardware vendor deliver a highly optimized binary for a specific architecture - the datasets and modulating parameters are randommized for each run and is therefor unknown for the vendors (and everyone else). The system will first calculate the result with the opensource (generic) algorithm for verification of the results from the vendor binaries. Each run could run a single algorithm or multiple algorithms working on the same randomized dataset in serial.

If you really think software developers optimize code, you can't be more wrong. The vast majority of devs either don't have time for that (all what matters is time to market) or simply don't even know what a cache or a pipeline is or are using languages that hide low-level details required to really tune things. And again that's not the point of GB...

Most doesn't - I agree, but we're some that actually do use the time to optimize our code.

It's strange that Cinebench is used, since Maxon have moved to the Embree (Intel) library which is 25%-250% faster than the original algorithm used in Cinebench.

 

[North01]

Apparently a device called the ASUS T90 Chi has appeared:

PDF Link

This might be a ~10" version of the ASUS T300 Chi (12.5"), possibly based on Intel's 10.1" Llama Mountain tablet with Intel Core M.

 

[liahos1]

Apparently a device called the ASUS T90 Chi has appeared:

PDF Link

This might be a ~10" version of the ASUS T300 Chi (12.5"), possibly based on Intel's 10.1" Llama Mountain tablet with Intel Core M.

ya i need that aluminum backplane. not really sure what is going on with that lenovo device but seems to be way off what intel was saying core m could do on llama mountain

 

[North01]

Dell Venue 11 Pro and Latitude 13 coming soon with Intel Core M.

 

[Cerb]

It's strange that Cinebench is used, since Maxon have moved to the Embree (Intel) library which is 25%-250% faster than the original algorithm used in Cinebench.

Where on Maxon's website can one download a tool to test with this thing they've moved to?

P.S. TBF, I find it strange that Cinebench is used as a non-synthetic benchmark, more than anything else .

 

[Space69]

Where on Maxon's website can one download a tool to test with this thing they've moved to?

You can't, but if you're abit adventurous you can download embre here :

http://embree.github.io/index.html (remember the renderer)

For maximum performance you need the SPMD compiler found here :

http://ispc.github.io/

You just use Visual Studio. Unfortunately if you want to target AVX/AVX2 you need the Intel Compiler - not free.

If you're not that adventurous then, I have done most of it - AVX2, although with ISPC 1.7 and Embree 2.3. It's a 50mb (90% scene files) rar file - I don't know where to upload, so we have to figure that out.

 

[Cerb]

That was basically a rhetorical answer. If the whole package can't be easily gotten and used, then it's not strange at all.

 

[jpiniero]

You know, I wonder if Core M is contributing to Intel's problems with the 14 nm node. After all, they did have to create an entirely separate process in order to get the voltage down; and it was so difficult that it distracted them from doing what was needed for the 'normal' process.

 

[Abwx]

And lastly, why is it unacceptable that a 3.5W device can't compete with real laptops?

It is set at 4.5W in the Yoga, the 3.5W are an average, max is at 4W during thoses benches, CPU utilisation being at most 80% we can conclude that the max thermal envelloppe is 5W, this should eventualy include 10% losses on the VRMs wich lead right to 4.5W...


Anyway look like i was spot on on about everything...

the back cover was a piece of work impossible to duplicate in mainstream designs and that thermal storage would be very limited.

if the YOGA is benched several times using CB you ll see the consecutive scores collapsing down to the level at wich the device thermal resistance is enough to sustain 6.5W or eventualy more, it all depend of the dissipation capability of the notebook.

• First run: OpenGL 19.54 fps, CPU 171 cb
• Second run: OpenGL 18.84 fps, CPU 153 cb
• Thrid run: OpenGL 13.4 fps, CPU 121 cb

 

[SickBeast]

Are the Broadwell i3s going to be quad core? They're going to have to do it at some point.

 

[jpiniero]

Are the Broadwell i3s going to be quad core? They're going to have to do it at some point.

No. Nor do they need to really. Like the quads, they will probably continue to buff up the iGPU instead.

 

[Enigmoid]

It is set at 4.5W in the Yoga, the 3.5W are an average, max is at 4W during thoses benches, CPU utilisation being at most 80% we can conclude that the max thermal envelloppe is 5W, this should eventualy include 10% losses on the VRMs wich lead right to 4.5W...


Anyway look like i was spot on on about everything...

Nope and Nope. There is no such thing as set to 4.5W then drops to 3.5W. TDP is set, fixed, irrespective of workload; clocks dropping are a result of thermals, not a changing tdp. The fact that the yoga uses 3.5W over a long period means that 1) 3.5W is the tdp, or 2) thermals cannot be maintained at 4.5W. If TDP is set to 4.5W and thermals can be maintained then the chip will use 4.5W.

Also VRM's are on die. They are included in tdp.

I agree with you in that intel's reference tablets were the absolute best that we would see with core m.

 

[Abwx]

Nope and Nope. There is no such thing as set to 4.5W then drops to 3.5W. TDP is set, fixed, irrespective of workload; clocks dropping are a result of thermals, not a changing tdp. The fact that the yoga uses 3.5W over a long period means that 1) 3.5W is the tdp, or 2) thermals cannot be maintained at 4.5W. If TDP is set to 4.5W and thermals can be maintained then the chip will use 4.5W.

Also VRM's are on die. They are included in tdp.

I agree with you in that intel's reference tablets were the absolute best that we would see with core m.

Right for the fully integrated VRMs.

In principle the thing is set to 4.5W according to the data displayed, we can see that max TDP was 4W, wich means that the device can sustain such power if there s enough thermal headroom, since these are figures that encompass a serie of benches it s quite possible that in a heavy CPU+GPU loading such amount is sustained, CB wont fully load the device anyway and it is displayed that CPU max usage was 80% during theses tests, that is, 80% at 4W power, also they could had heated the device such that there was not much thermal headroom left when running several benches, wich is obviously easy to get looking at the successive CB scores.

 

[Fjodor2001]

A not so positive review of the Broadwell-M based Lenovo Yoga 3 Pro:

http://www.extremetech.com/computin...ressing-mediocre-picture-for-intels-broadwell

The first Core M laptop paints a depressing, mediocre picture for Intel’s Broadwell

To say there’s a great deal riding on the launch of Intel’s Core M is something of an understatement. The chip — and Intel’s 14nm hardware — is nearly a year late. The delays have raised investor questions about Intel’s ability to maintain or leverage a technological advantage over its rivals,
[...]
While many reviewers praise its diminutive size, low weight, and display quality, complaints about battery life, heavy throttling, and poor performance are common.
[...]
Curiously, reviewers are completely split on whether or not the system even has a fan. Some reputable outlets make specific note that the computer lacks one, while others complain about the sound level. What everyone agrees on is that the system throttles constantly, possibly in part because Lenovo chose to set a 3.5W target for the chip rather than the 4.5W TDP that Intel specifies
[...]
Lenovo’s claim of “up to nine hours” on battery life is farcical. WindowsCentral.com claims 5-6 hours, at the very most. PC Pro hit eight hours, but only by turning screen brightness to its lowest levels; turn brightness up and battery life plummets. UltraBookReview reports that under various workloads battery life ranges from 6 hours to 4 hours 40 minutes depending on workload.

And then finally there's a rant on OEM laptops in general...

One of the reasons we can’t tell if the problems reviewers are experiencing are fundamental issues with Broadwell, or caused by Lenovo’s bad system design, is because the history of PC laptops is basically the history of terrible design decisions writ large upon an unsuspecting but increasingly unhappy populace.
[...]
Why?
Because laptop OEMs are gutless and sell on specs, not on experience. Because even when they build $1200 hardware, they infuse that price point with $300 thinking. Crank up the resolution, but use a panel with a bad color gamut. Slash the thickness, but gut the battery life. Charge four figures, but refuse to remove the spyware and shovelware that infest most OEM laptops like a bad case of fleas. Include a terrible webcam, because you can get away with saving 20 cents on the part. Improve the specs on individual parts, but don’t combine those improvements into superior products.

So possibly it can be just a poorly designed laptop by Lenovo that is the cause of the problems. I guess we'll have to wait until more Broadwell-M based laptops have been reviewed...

 

[SAAA]

Don't worry too much guys, the problem isn't core-M or Intel lying apparently, just the ridiculous tdp set by the OEM (it's French but images are self-explanatory):
http://www.clubic.com/processeur/processeur-intel/actualite-734359-intel-core-m-tdp-variable.html

Basically step up the TDP to 5.5, not even 6W and the chip runs at a nice & constant 1.7 GHz after a few seconds of turbo. Just for fun look how low it clocks to keep a 2W tdp... or better how in both tests its never temperature limited (technically it could reach 95'C) but mostly power limited hence the lower power you allow the more it downgrades.

 

[Enigmoid]

I would not prematurely judge core m yet. The yoga is only one product and it seems that lenovo's recent success in the market is causing them to skimp on quality.

The y series went from decent quality at good prices to poor quality at ok prices. The new y50 throttles to base clock for example, making it slower than its predecessor.

 

[Arachnotronic]

Don't worry too much guys, the problem isn't core-M or Intel lying apparently, just the ridiculous tdp set by the OEM (it's French but images are self-explanatory):
http://www.clubic.com/processeur/processeur-intel/actualite-734359-intel-core-m-tdp-variable.html

Basically step up the TDP to 5.5, not even 6W and the chip runs at a nice & constant 1.7 GHz after a few seconds of turbo. Just for fun look how low it clocks to keep a 2W tdp... or better how in both tests its never temperature limited (technically it could reach 95'C) but mostly power limited hence the lower power you allow the more it downgrades.

Good find. Thank you.

Why would Lenovo purposely cripple the chip? Was it to save battery life, or was this just a *really* silly oversight?

 

[dahorns]

Good find. Thank you.

Why would Lenovo purposely cripple the chip? Was it to save battery life, or was this just a *really* silly oversight?

To me, it looks like they designed a chassis without any consideration for heat. They thought if they just went ahead and stuck a small fan in, it would all work out.

It may also be a really silly oversight. The reviews are still coming in from journalist samples, no? I mean, is this confirmed performance from a shipped consumer product?

 

[Abwx]

Good find. Thank you.

Why would Lenovo purposely cripple the chip? Was it to save battery life, or was this just a *really* silly oversight?

43.3Wh battery life is 280min, scores numbers are here :

http://www.clubic.com/pc-portable/ultrabook/article-733119-1-lenovo-yoga-3-pro-test.html

 

[witeken]

Here's something to consider for everyone who's on the Core M disappointment bandwagon:

Image if 28nm=22nm and instead of shrinking from 22nm to its 14nm process, Intel would go from TSMC's 28nm to its 20nm process; and from 11.5 to 3.5W.


Would Broadwell now be a great chip? No, it would be even worse; the TDP (at iso performance) would only shrink to 9.2W.

So instead of comparing a 3.5W Core M SKU to 15W chips (which is utterly insane) that thus have a 4.3x lower TDP (but note that it isn't even half as bad!), why not compare it to other chips with a 3.5W TDP?

But I guess people will pretend that the iPad Air (2) has a lower TDP. The only thing that has gone wrong is the availability. Intel is almost half a year late, and yet they can't manage to bring out a compelling (launch) device with Core M at a compelling price point.

 

[Qwertilot]

Fundamentally though, is there really very much space for them to do that? It almost certainly isn't in replacing mainstream Haswell ULV stuff in 1000+ notebooks.

Goodness, the ULV broadwells are going to be very power efficient/power down when needed anyway, so even CoreM isn't actually going to gain you a huge chunk of battery life.

If you go into tablets etc then you've got problems with working out why you'd want that sort of premium vs Cherry Trail etc.

Apple would likely do something nice which would sell well if they use it. Not obvious that they will though. They didn't announce it at that last event of theirs, which you'd think they'd probably have done were they going to use CoreM?

 

[Nothingness]

But I guess people will pretend that the iPad Air (2) has a lower TDP. The only thing that has gone wrong is the availability. Intel is almost half a year late, and yet they can't manage to bring out a compelling (launch) device with Core M at a compelling price point.

Who cares about TDP when it seems iPair Air battery life is as good or better than the only existing Core M machine available? The TDP number in isolation is a completely useless number in particular in "closed" devices like this.

Anyway let's wait for other Core M devices before spitting at the CPU face, IMHO Yoga 3 is just a bad design