Intel Broadwell Thread

[Abwx]

TDP is TDP.

Turbo boost can exceed TDP for short periods,

The times it takes to complete CB 11.5 is surely not a short period,during all this time it pump 15W, so its TDP is 4.5W for the datasheet but 15W for benches and its claimed perfs, nice marketing spinning, if TDP is really 4.5W then bench it at 4.5W and then show us what it can do with extended TDP to 15W.

 

[kimmel]

I wish Intel would stop posting TDPs and SDPs like Apple does. Then maybe people would finally shut up about this whole stupid debate.

 

[bullzz]

@Abwx - as long as ur able to hold the tablet in hand, how does it matter if its sucking 4.5W for an extended period of time or 15W for short period. hell some desktop processors use more power during idle than advertised tdp

 

[Sweepr]

@Abwx - as long as ur able to hold the tablet in hand, how does it matter if its sucking 4.5W for an extended period of time or 15W for short period. hell some desktop processors use more power during idle than advertised tdp

It must use that much power for very short periods of time, not during the entire Cinebench / 3Dark sequence, otherwise you probably wouldn't be able to fit them in a 7.4'' tablet chassis without a fan that doesn't seem to get hot to touch (François video). All in all if they can deliver Core i5 Haswell-U like performance in a 4.5W TDP chip that fits thin fanless tablets / convertibles with great battery life - that's what really matters, the general public doesn't care about 15W spikes or specifics.

 

[witeken]

Stunning.

Broadwell must be the most impressive upgrade and silicon that I have ever seen.

Beats every other passive SoC by a factor of 2+: :thumbsup:
Beats Nvidia at its own game, the GPU: :thumbsup:

So I don't understand why the 4.5 W is not just SDP then.

Because the 4.5W is exactly what is determined by the definition of TDP.

Indeed, that 15W is essentially a very short cTDP up. But either way, this has got some serious muscle.

The 15W has nothing to do with TDP. It is turbo.

BTW, it' a bit disturbing that everyone believes a random review site with the 15W typo. Intel only released 3 SKUs, with at most a 4.5W TDP.

 

[NTMBK]

Beats every other passive SoC by a factor of 2+: :thumbsup:
Beats Nvidia at its own game, the GPU: :thumbsup:

Beats every other SoC on size of price by a factor of 10: :thumbsup:

 

[witeken]

Beats every other SoC on size of price by a factor of 10: :thumbsup:

The total price is only a few % more, so the price per dollar is unmatched by any other tablet.

 

[StinkyPinky]

Intel have the best engineers in the business. I'm not sure why people act so shocked about it all. Intel were caught out by the ultra mobile revolution in computing and therefore took half a decade to catch up. Now they're back in the game and will wipe the floor imo. They also have the best fabs.

 

[USER8000]

Sneak Peak by François: Core M-5Y70 running 3DMark Ice Storm in a 7.4'' fanless tablet config, >50k overall score and doesn't seem to get hot.

www.youtube.com/watch?v=FIKpypEeo7U&feature=youtu.be

60% faster than Tegra K1, ~3.25x faster than Apple A7.

Emm,have you watched the video??

All you can see is black screens and no actual test running(unless that little box with texture corruption is the test,but the test seems to not run anywhere as long as it needs to run too) and the tablet is mounted on a stand.

The video appears to have no editing as the chap doing the filming has his reflection in the bezel(unless he is very good at being a robot and standing very still with no movement for minutes at a time).

This is what one part of the Ice Storm sequence should look like when running:

https://www.youtube.com/watch?v=3gaL-RGRpz4

I think I am going to wait until an actual website does some proper tests on Core M based computers first.

 

[liahos1]

The total price is only a few % more, so the price per dollar is unmatched by any other tablet.

agreed. this is basically sp3 i5 performance at half the price and now thinner, lighter, cooler and fanless and with likely better battery life (HP sku says 13hrs). its an amazing value. once console os comes out it will be an unmatched device.

given apples claims of 50% better gpu performance and 25% better cpu performance in a8 vs a7. the A8 gets TROUNCED here. i wonder how hard it would be for apple to move ios to x86 compatibility and get the future successors of these chips in ipads.

 

[jj109]

Emm,have you watched the video??

All you can see is black screens and no actual test running(unless that little box with texture corruption is the test,but the test seems to not run anywhere as long as it needs to run too) and the tablet is mounted on a stand.

The video appears to have no editing as the chap doing the filming has his reflection in the bezel(unless he is very good at being a robot and standing very still with no movement for minutes at a time).

This is what one part of the Ice Storm sequence should look like when running:

https://www.youtube.com/watch?v=3gaL-RGRpz4

I think I am going to wait until an actual website does some proper tests on Core M based computers first.

Ice Storm Unlimited is rendered off-screen.

https://www.youtube.com/watch?v=XGEzGTTl1j8

 

[sontin]

The total price is only a few % more, so the price per dollar is unmatched by any other tablet.

So, we will buy a 8" Broadwell tablet for $299 like the Shield tablet?
Or wait, the SoC costs that alone...

 

[Eug]

I'd buy a $1000+ Core M Retina MacBook.

I'll stick with A8 for a tablet.

 

[witeken]

So, we will buy a 8" Broadwell tablet for $299 like the Shield tablet?
Or wait, the SoC costs that alone...

At which price would Broadwell start... $599?

So if you take half of that, your $300 Shield tablet, you get almost 5X lower performance.

This is probably best-case, but you get the point.

 

[liahos1]

At which price would Broadwell start... $599?

So if you take half of that, your $300 Shield tablet, you get almost 5X lower performance.

This is probably best-case, but you get the point.

what would be most interesting is getting a 2-1 device with android compatibility. then you can have it all i one package and the value of the device would be unmatched. i cant run my work apps on a shield

 

[erunion]

If native android isn't possible, core m should be fast enough to emulate android pretty well.

 

[liahos1]

If native android isn't possible, core m should be fast enough to emulate android pretty well.

i use bluestacks now and its performance / implementation on my sp3 leaves a lot to be desired. although to be honest most of the apps i had on my ipad/nexus 7 i have on windows. and the browsing experience is far better. i do think having core m as a windows only device limits its potential for sales though. i dont know why more of these dual os devices arent proliferating.

edit - nm i guess google and ms killed the idea.

 

[IntelUser2000]

15W is PL3, and its only there for 10ms, which is nothing. PL2 is the Turbo that's most relevant with Cinebench, and can run for few minutes, and that's 25% above TDP. But running even longer than that it will bring it back to PL1, which is TDP.

Good to see that its meeting hyped expectations. Very rarely products do. I assume other Broadwell chips won't be as impressive though.

 

[bullzz]

@IntelUser2000 by "other broadwell chips" which ones do u mean. i think the U chips will be equally impressive. broadwell desktop i dont care since skylake less than a year away

 

[Khato]

i use bluestacks now and its performance / implementation on my sp3 leaves a lot to be desired. although to be honest most of the apps i had on my ipad/nexus 7 i have on windows. and the browsing experience is far better. i do think having core m as a windows only device limits its potential for sales though. i dont know why more of these dual os devices arent proliferating.

edit - nm i guess google and android killed the idea.

Yup, Intel put forth the effort to make it possible and easy to implement only for both Google and Microsoft to do what they can (which is a lot) to discourage it. http://arstechnica.com/gadgets/2014...out-to-stop-dual-boot-windowsandroid-devices/

It makes sense, both Google and Microsoft are in pursuit of total world domination after all, right? While that struggle resulting in the death of dual boot it actually should be advantageous to Intel over all. Google's success coupled with Intel offering a low price platform is what drove Microsoft to finally adopting appropriate licensing for the tablet market. And in turn Microsoft actually becoming a threat in the low end tablet market pretty much requires Google to step up their level of cooperation with Intel, else more and more of Intel's gains in tablets will come at the expense of Android to the benefit of Windows.

 

[ams23]

At which price would Broadwell start... $599?

It appears that most high resolution Core M-powered 2-in-1 convertible tablets will be sold for closer to $800-$1000 USD, but pricing is not so clear at this point. Anyway, this is a very premium new SoC fabricated on a bleeding edge fabrication process that will appear in very premium products.

So if you take half of that, your $300 Shield tablet, you get almost 5X lower performance.

That is very misleading. The Core M tablet that was tested by PCPer used a completely different browser in Internet Explorer that is very highly tuned for Javascript Sunspider. Note that the Tegra 4-powered Surface 2 using Internet Explorer has a better Sunspider score than the A7-powered iPad Air using Safari and a much better Sunspider score than the TK1-powered Shield tablet using Chrome!

And in GPU-limited (rather than CPU-limited) scenarios such as GFXBench 3.0, Tegra K1 is expected to be quite close in graphics performance to Core M considering it's performance relative to the Core i5 variant in Surface Pro 3.

That said, Core M is clearly head and shoulders ahead in overall CPU/browser performance, and I would estimate that it is ~ 2-3x faster than any SoC using Cortex A15.

Core M is a heck of a performer, but at the end of the day it is more of a replacement to the 15w Haswell SKU's (ie. lower power at similar levels of performance), and it will be mostly utilized in higher end Windows tablet convertibles, Windows laptops, and MacBook laptops while Tegra K1 will be mostly utilized in more moderately priced Android tablets and Chromebooks.

 

[Roland00Address]

First I want to point out that I do not know if we can trust the pcper information. I point this out for there a8-7600 45w in this review shows 2.65 Link for cinebench 11.5 multithread and in there other review it shows 3.01 Link

That said the normal score the 45w amd a8 7600 gets is 3.00 or 3.01 from other sites. So it looks like they made an error on this slide and if they made 1 error they can make many errors. Most likely there slide is accurate with the core m info and they just made a copy paste / transcribing error. I understand what it is like to be human for I make these types of mistakes all the time.

-----

That said...Oh this just hurts, look at Intel vs AMD it makes me want to cry. Cinebench 11.5 numbers

2.77 Intel 5Y70 pcper 4.5 watt tdp llano mountain platform passively cooled in a sub <8mm thick tablet pcper

2.04 AMD 5350 25 watt-Anandtech
2.67 AMD FX-7600p 35watt-sweclockers
3.01 AMD A8-7600 45 watt mode-pcper
3.36 AMD A8-7600 65 watt mode-pcper
3.58 AMD A10-7850k 95 watt mode-pcper

This is what you get when you are two nodes ahead, you have actual engineering resources and thus you can hire talent and do not have a possible brain drain because you have to be stingy with what you pay your people, and you do not have an architecture that is crap and should never have been greenlighted. You get 92% of the performance in 1/10th the tdp, and your chip only has a die size of 82mm^2 vs AMD Kaveri die size is 245 mm^2

 

[IntelUser2000]

It appears that most high resolution Core M-powered 2-in-1 convertible tablets will be sold for closer to $800-$1000 USD,

It won't be competitive in pricing but that's not the point. It does reduce the prices though. The Lenovo Helix sold for $1500 while the Core M version brings it down to $999. In average, you'll probably see it $200 lower than previous generations. Essentially you are talking about taking the premium pricing of the convertibles away.

That is very misleading. The Core M tablet that was tested by PCPer used a completely different browser in Internet Explorer that is very highly tuned for Javascript Sunspider.

FYI, the Core M 5Y10a gets 194ms on Chrome. The 5Y70 should be able to get 150ms.

And in GPU-limited (rather than CPU-limited) scenarios such as GFXBench 3.0, Tegra K1 is expected to be quite close in graphics performance to Core M considering it's performance relative to the Core i5 variant in Surface Pro 3.

That's probably true. Intel never touted about GPU gains because they knew that CPU would be the real performer. But its still very good.

get 92% of the performance in 1/10th the tdp, and your chip only has a die size of 131mm^2 and it is a true soc with everything on that chip, pch and everything integrated.

131mm2 is the die size of Haswell-Y. Broadwell Y, or Core M is 82mm2. Of course that doesn't include the PCH.

I wonder what the deal is with the Cinebench results from the PcPer and HotH reviews. 2.77pts vs. 2.48pts.

2.48 points is likely the steady state performance, while 2.77 points is when its running the first time.

 

[Abwx]

@Abwx - as long as ur able to hold the tablet in hand, how does it matter if its sucking 4.5W for an extended period of time or 15W for short period. hell some desktop processors use more power during idle than advertised tdp

Because perf variability can be 50% according to Intel in fonction of the device temperature, i guess it was better to launch them on december, but anyway the TDP is temperature dependant, 4.5W is for a heated device that run will run close to 1-1.4GHz.

http://www.hardware.fr/news/13888/idf-details-changements-broadwell.html

15W is PL3, and its only there for 10ms, which is nothing. PL2 is the Turbo that's most relevant with Cinebench, and can run for few minutes, and that's 25% above TDP. But running even longer than that it will bring it back to PL1, which is TDP.

25% more TDP for 50% more perfs.??.

 

[Roland00Address]

131mm2 is the die size of Haswell-Y. Broadwell Y, or Core M is 82mm2. Of course that doesn't include the PCH.

You are right, I made a mistake, oops going to edit.