So long as sustained performance isn't way lower than shown, this is def a huge win for Surface Pro. Similar performance to i5 SP3 at 1/3 the power use, awesome.
If it was, I imagine they'd just keep their current fan-cooled chassis. Still, the TDP is 5 Watts. At full power, it would only display a third of the heat.
Eh. I'm SUPER excited about a potential Surface Pro that's fanless. Frankly right now I wish there was an Atom model, as I love everything about the Surface Pro 3 save that I don't want a fan since I'll be using it heavily as an ereader.
I think 2nd gen Atom already has the performance I need...but hey, I'll take a 4.5-6 watt chip with the performance of a 15 watt Haswell!
Lame. Unless you just want to check your facebook.
Honestly Would like to see a Windows Convertible laptop/tablet with one of these chips. Though even a new Surface Pro using one would be nice. Though I hope the keep the fans, and give the machines less chance of throttling. Would be great during those long Civ 5 matches.
Would this enable an even thinner/lighter Surface Pro 4? Or perhaps they could retain the same chassis but have the benefit of reduced thermal throttling.
Well... does the TDP for Broadwell Y include the southbridge? I think it's on package, but isn't it still on a separate die? I realize that's not exactly a fair comparison since Broadwell Y offers more IO options, but still something to consider. Also, K1 is built on 28nm, so there's a huge disadvantage there.
I created this account to say that you had me in tears with that comment. Good old winamp.
Anandtech writers - thank you so much for covering these chips in depth - as a current T100 owner who loves the fact that it's silent I watch with great interest. Good work!
Wow. I might actually be tempted into buying my first tablet, if this can truly be the level of performace in a $300 1080p mobile device running Win 8.x . Pretty sweet!
Intel's challenge long term in breaking into the mobile market isn't getting power down/claiming the performance/watt crown at the high-end. Clearly they can do that. The question is can they sustain their ultra-high profit margin business model in an era of sub-$10 SoCs.
Intel stated last week that these would be showing up in "premium" tablets and 2-in-1s at the $799 level (and higher), but suggested that thicker models (less premium) may show up as low as $599.
Supposedly the tray prices on these chips alone are around $280 ea, but intel has ample wiggle room given these are just downclocked dual core parts with possibly a varied fab process.
Are these the ridiculous $280 mobile chips? You previewed those a few weeks ago, I think. Intel has way too many product code names / lines to keep track of.
you are right on both accounts, but the pricing isn't all that ridiculous when you see that they can outperform similarly priced haswell ULV-chips at a third of the power.
If you haven't seen it, Tech Report includes the ASUS Transformer T100 (Atom Z3740) in their charts: http://techreport.com/news/27039/preliminary-core-...">Preliminary Core M benchmarks hint at Broadwell's potential
So based on the SunSpider test, a Broadwell-M mobile chip with a 2.0GHz turbo isn't just matching a Haswell-U processor with a 2.9GHz turbo and 3x TDP, but is almost doubling performance too?
i think the 5y70 is clocked at 1.1ghz with a turbo of 2.6ghz for both cores. performance seems incredible here, but those two quick benchmarks won't tell us the whole story just yet.
It is hard to believe if both cores were turbo at 2.6Ghz that this chip could just consume 4.5w TDP!!!. This does NOT seem possible at all. Yeah, if that throttling is just a few seconds, it would jack up the overall score but it could alternate from 2.6Ghz an 1.3Ghz for a large portion of the benchmark to give the indicated score. I would expect the TDP would be somewhere around the 12w zone. We need Anandtech to do power measurements during the benchmarks to see where the power spike were and how long they lasted.
It's the turbo to 2.6Ghz version, plus the Core M tablet seems to be running IE11 where the older Surfaces are either in a power saver mode or ran in Chrome instead. SP2 here, I get 115ms in IE11 in balanced power profile.
It's also important to keep in mind that the turbo mechanism is quite strong. The power draw can spike to 15W or more in some scenarios but the TDP is constant at 4.5W.
The second picture is actually the Surface Pro while the third pic is the Llama Mountain reference PCB. Shows the big difference in size and cooling needed.
this picture had me confused too for a sec, but then i read on: "For reference, the photo above is from our Surface Pro teardown. The PCB takes up almost the entirety of the tablet, and there's a great deal of extra thickness from the fans in the device."
I don't think this chip will be able to sustain that performance in the chassis it's going to ship in. I expect there to be quite a bit of thermal throttling for extended high-performance use cases.
We definitely can't eliminate that possibility as these graphs also show the Surface Pro 3 results as better than the Pro 2 in 3DMark; this is patently untrue in extended real-world use so we can conclude that 3D mark is probably not sufficient to induce extended-use throttling.
Yes, it's unlikely it was 4.5W since most Intel CPUs can run at 125% or more of their official TDP for some time. Also as others noted it is way too fast to be true on SunSpider, so this is obviously due to the usual browser differences and tricks (unlike the other results the 5Y50 scores are provided by Intel). People will be quite disappointed when the actual CPUs become available as they of course won't be twice as fast as Surface Pro!
Speculating here: Core M was ran on IE11, other Surfaces in Chrome. I get 115ms with my SP2 in IE11. For a quick test, being a hair faster than a 4200U based device makes sense as they turbo to the same frequency so the few percent better IPC will pull it ahead.
Wow. Just wow. I'd been thinking of the Surface Pro 4, and I assumed it would be like having the current i3 at the current i5's performance (plus a little). Maybe they actually *can* redo the chassis again to make it fanless?
(As a buyer of the original Surface Pro, I'm kinda eyeing the market, and this might be the difference between buying in February and buying in October.)
Good thing I bought the replacement warranty on my Surface Pro 3. I might have to "accidentally" break it when these show up in a revised SP3 or a SP4. Holy crap.
"One of the biggest surprises for me was how much smaller the PCB was in Llama Mountain. For reference, the photo above is from our Surface Pro teardown. The PCB takes up almost the entirety of the tablet, and there's a great deal of extra thickness from the fans in the device."
To be fair, that Surface Pro board isn't very dense at all. Look at Apple's Haswell-equipped logic boards, and you'll get a really good idea of what can be done with Haswell. Still, that Broadwell board is much smaller, and it'll be amazing to see how much more they can condense things over the next few generations.
That sun spider test for Core M has been run under IE and Surface Pros on Chrome obviously. Running that test on 5Y70 on Chrome would give score around 172. Optimally one should not compare web benchmarks between different browsers. IE is heavily optimized for sun spider while in other tests looses to Chrome almost twofold.
I wonder what's the chances of someone picking Core M up and using it in a high-end Chromebook.
Probably not much, manufacturers are scared of high-priced Chromebooks after the Pixel, but I think the market has markedly changed, and if someone picked this up, put it in an attractive-but-not-over-the-top body and matched it with a nice keyboard and trackpad and a 1080p IPS display and sold it for $699, it would do very well. If an OEM excluded fancy things like WiGiG or Wireless Charging or touchscreen panels or any of that, and sticks to the basics, I think it would be possible to hit that price and still have decent margins.
Where are all those people that were claiming the A8 was going to be as fast as this? This looks frankly fantastic and I find it hard to believe that an iPad Pro is on the cards against a tablet like this. Watch Apple surprise us and release a touch enableled MacOS tablet running this chip!
With such a simple and modest pcb we should be looking at a complete tablet BOM of around $100 for a 7" with 32GB of NAND. Plus the cost of the cpu of course. Now the question is, how is intel going to charge $200 for the cpu. In the past they have always offered "good enough" celerons and pentiums for a <$100 tray price. (Such as the 3560Y) So when the first broadwell sub 5 watt pentium is released, they might actually start gaining market share.
in comparing the PCB from Pro 3 to the one in the new tablet, does that mean I can expect the extra space to be replaced by a larger battery so we get a double win - less power hungry AND larger battery for a much increased tablet battery life without an increase in weight!
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
81 Comments
Back to Article
liahos1 - Wednesday, September 10, 2014 - link
are you going to be able to do a deeper benchmark analysis on the reference device?B3an - Wednesday, September 10, 2014 - link
This NEEDS to be in the Surface Pro 4 now...frostyfiredude - Wednesday, September 10, 2014 - link
So long as sustained performance isn't way lower than shown, this is def a huge win for Surface Pro. Similar performance to i5 SP3 at 1/3 the power use, awesome.mkozakewich - Wednesday, September 10, 2014 - link
If it was, I imagine they'd just keep their current fan-cooled chassis. Still, the TDP is 5 Watts. At full power, it would only display a third of the heat.craighamilton - Saturday, December 6, 2014 - link
There are a number of higher rated phones, I recommend seeing http://www.topreport.org/tablets/ among others.nofumble62 - Thursday, September 11, 2014 - link
No, these chips must go in the Macbook Air first.Wolfpup - Monday, September 29, 2014 - link
Eh. I'm SUPER excited about a potential Surface Pro that's fanless. Frankly right now I wish there was an Atom model, as I love everything about the Surface Pro 3 save that I don't want a fan since I'll be using it heavily as an ereader.I think 2nd gen Atom already has the performance I need...but hey, I'll take a 4.5-6 watt chip with the performance of a 15 watt Haswell!
Spunjji - Monday, November 10, 2014 - link
Late reply... but FYI, the Surface Pro 3 fan doesn't engage when you're using it that lightly.halo37253 - Monday, November 3, 2014 - link
Lame. Unless you just want to check your facebook.Honestly Would like to see a Windows Convertible laptop/tablet with one of these chips. Though even a new Surface Pro using one would be nice. Though I hope the keep the fans, and give the machines less chance of throttling. Would be great during those long Civ 5 matches.
Stochastic - Wednesday, September 10, 2014 - link
Would this enable an even thinner/lighter Surface Pro 4? Or perhaps they could retain the same chassis but have the benefit of reduced thermal throttling.mkozakewich - Friday, September 12, 2014 - link
Also, more cooling would mean more TurboBoost.Roland00Address - Wednesday, September 10, 2014 - link
Which surface pro 3? The i5, the i3, the i7?Roland00Address - Wednesday, September 10, 2014 - link
And I see the i3 labeled as a separate bar after I hit send :(Laxaa - Wednesday, September 10, 2014 - link
There's both the i3 model and another model that's not labeled. I'm guessing it's the i5 version.dragonsqrrl - Wednesday, September 10, 2014 - link
If I remember correctly there's not much of a performance difference between the i5 and i7 versions.mtalinm - Thursday, September 11, 2014 - link
MSFST did not sample any reviewers an i7. anand's review says it's an i5.monstercameron - Wednesday, September 10, 2014 - link
yeah right, so let me get this straight...uses less power than a tegra k1, yet offers more performance than a sufrace pro 3. impossibru!shing3232 - Wednesday, September 10, 2014 - link
5y70 is 14nm part, so it is possibledragonsqrrl - Wednesday, September 10, 2014 - link
Well... does the TDP for Broadwell Y include the southbridge? I think it's on package, but isn't it still on a separate die? I realize that's not exactly a fair comparison since Broadwell Y offers more IO options, but still something to consider. Also, K1 is built on 28nm, so there's a huge disadvantage there.IntelUser2000 - Wednesday, September 10, 2014 - link
Yea... it does. The 4.5W is the package TDP. It's kinda like how it went from 17W + 3W in Ivy Bridge to 15W combined in Haswell.Wilco1 - Wednesday, September 10, 2014 - link
Indeed. And at a far lower frequency as well. Look, another flying pig!This is why one shouldn't rely on cheated SunSpider scores for benchmarks...
Alexey291 - Friday, September 12, 2014 - link
Especially since the sunspider benchmarks seem to have been run on different browsers. Welpkron123456789 - Friday, September 12, 2014 - link
It's possible, but this chip costs $281. There won't be ~$300 tablets with that chip, only $800-$1000.piroroadkill - Wednesday, September 10, 2014 - link
Goddamn, this really whips the Llama's ass!Pissedoffyouth - Wednesday, September 10, 2014 - link
I created this account to say that you had me in tears with that comment. Good old winamp.Anandtech writers - thank you so much for covering these chips in depth - as a current T100 owner who loves the fact that it's silent I watch with great interest. Good work!
thepaleobiker - Wednesday, September 10, 2014 - link
"throws money at screen. Nothing happens"Wow. I might actually be tempted into buying my first tablet, if this can truly be the level of performace in a $300 1080p mobile device running Win 8.x . Pretty sweet!
liahos1 - Wednesday, September 10, 2014 - link
i think these are for the premium tablet, 2-1, and fanless laptop market. pricing probably around 600 for tablets w/ windowsfokka - Wednesday, September 10, 2014 - link
considering the cpu alone is said to cost around 280$ i'd say it's unlikely we'll see a 300$ core-m tablet anytime soon.but if you want cheap you can look into atom, those things aren't the bottlenecks they were in the netbook days either anymore.
name99 - Wednesday, September 10, 2014 - link
Given that the chip costs $281, good luck with that $300 tablet dream...Theodore41 - Thursday, September 11, 2014 - link
OK,but I prefer to give the €600.00 for this,instead of switching to a iPad Air 2,from the Air I now have...Stochastic - Wednesday, September 10, 2014 - link
Intel's challenge long term in breaking into the mobile market isn't getting power down/claiming the performance/watt crown at the high-end. Clearly they can do that. The question is can they sustain their ultra-high profit margin business model in an era of sub-$10 SoCs.nathanddrews - Thursday, September 11, 2014 - link
Intel stated last week that these would be showing up in "premium" tablets and 2-in-1s at the $799 level (and higher), but suggested that thicker models (less premium) may show up as low as $599.MikeMurphy - Thursday, September 11, 2014 - link
Supposedly the tray prices on these chips alone are around $280 ea, but intel has ample wiggle room given these are just downclocked dual core parts with possibly a varied fab process.FelixDraconis - Wednesday, September 10, 2014 - link
Are these the ridiculous $280 mobile chips? You previewed those a few weeks ago, I think. Intel has way too many product code names / lines to keep track of.fokka - Wednesday, September 10, 2014 - link
you are right on both accounts, but the pricing isn't all that ridiculous when you see that they can outperform similarly priced haswell ULV-chips at a third of the power.JDG1980 - Wednesday, September 10, 2014 - link
You don't really think OEMs pay anywhere near sticker price, do you?Alexey291 - Friday, September 12, 2014 - link
Are you trying to say that these chips will be sold separately in a box?tipoo - Saturday, October 11, 2014 - link
Erm, that's for 1K tray pricing. Yes, that is the price for OEMs.ScottSoapbox - Wednesday, September 10, 2014 - link
What size screen is this reference tablet?JoshHo - Thursday, September 11, 2014 - link
I believe it is a 12.5" display, it was definitely on the large side.Pissedoffyouth - Wednesday, September 10, 2014 - link
Joshua, could you please add the results for the Atom z37xx chips, they are the closest Intel competitors at that kind of TDP?Pissedoffyouth - Wednesday, September 10, 2014 - link
The SDP is 2 watts, but the TDP of the Atom is 4w in case anyone is confusedStanand - Wednesday, September 10, 2014 - link
If you haven't seen it, Tech Report includes the ASUS Transformer T100 (Atom Z3740) in their charts: http://techreport.com/news/27039/preliminary-core-...">Preliminary Core M benchmarks hint at Broadwell's potentialJoshHo - Thursday, September 11, 2014 - link
Done, they should be updated soon.Pissedoffyouth - Thursday, September 11, 2014 - link
Thanks - that really puts it into perspective!OrphanageExplosion - Wednesday, September 10, 2014 - link
So based on the SunSpider test, a Broadwell-M mobile chip with a 2.0GHz turbo isn't just matching a Haswell-U processor with a 2.9GHz turbo and 3x TDP, but is almost doubling performance too?fokka - Wednesday, September 10, 2014 - link
i think the 5y70 is clocked at 1.1ghz with a turbo of 2.6ghz for both cores. performance seems incredible here, but those two quick benchmarks won't tell us the whole story just yet.fteoath64 - Saturday, September 13, 2014 - link
It is hard to believe if both cores were turbo at 2.6Ghz that this chip could just consume 4.5w TDP!!!. This does NOT seem possible at all. Yeah, if that throttling is just a few seconds, it would jack up the overall score but it could alternate from 2.6Ghz an 1.3Ghz for a large portion of the benchmark to give the indicated score. I would expect the TDP would be somewhere around the 12w zone. We need Anandtech to do power measurements during the benchmarks to see where the power spike were and how long they lasted.Wilco1 - Wednesday, September 10, 2014 - link
Yes, if anyone else believes that Broadwell provides 3x the IPC of Haswell, they are going to be quite disappointed...frostyfiredude - Wednesday, September 10, 2014 - link
It's the turbo to 2.6Ghz version, plus the Core M tablet seems to be running IE11 where the older Surfaces are either in a power saver mode or ran in Chrome instead. SP2 here, I get 115ms in IE11 in balanced power profile.OrphanageExplosion - Wednesday, September 10, 2014 - link
Thanks for that somewhat crucial clarification! Even parity with Haswell would be amazing, but we need the right data really.JoshHo - Thursday, September 11, 2014 - link
It's also important to keep in mind that the turbo mechanism is quite strong. The power draw can spike to 15W or more in some scenarios but the TDP is constant at 4.5W.nodak - Wednesday, September 10, 2014 - link
Is it actually a fanless tablet? The second photo has two seashell-shaped black disks that look just like the fans in a Macbook Pro...nodak - Wednesday, September 10, 2014 - link
Grey disks. Need to learn my colors.bepo - Wednesday, September 10, 2014 - link
The second picture is actually the Surface Pro while the third pic is the Llama Mountain reference PCB. Shows the big difference in size and cooling needed.tipoo - Wednesday, September 10, 2014 - link
"For reference, the photo above is from our Surface Pro teardown. "fokka - Wednesday, September 10, 2014 - link
this picture had me confused too for a sec, but then i read on: "For reference, the photo above is from our Surface Pro teardown. The PCB takes up almost the entirety of the tablet, and there's a great deal of extra thickness from the fans in the device."-> it's a photo of the surface pro
liahos1 - Wednesday, September 10, 2014 - link
this chip is magicaldanstek - Wednesday, September 10, 2014 - link
What's the display resolution on the reference tablet?dragonsqrrl - Wednesday, September 10, 2014 - link
...Wow, that Sunspider Benchmark... And at 4.5W TDP?! Even the 3DMark score is impressive considering the much lower TDP.Stochastic - Wednesday, September 10, 2014 - link
I don't think this chip will be able to sustain that performance in the chassis it's going to ship in. I expect there to be quite a bit of thermal throttling for extended high-performance use cases.Spunjji - Thursday, September 11, 2014 - link
We definitely can't eliminate that possibility as these graphs also show the Surface Pro 3 results as better than the Pro 2 in 3DMark; this is patently untrue in extended real-world use so we can conclude that 3D mark is probably not sufficient to induce extended-use throttling.fokka - Wednesday, September 10, 2014 - link
i'm really wondering at what TDP and clocks this thing ran during the benchmarks.Wilco1 - Wednesday, September 10, 2014 - link
Yes, it's unlikely it was 4.5W since most Intel CPUs can run at 125% or more of their official TDP for some time. Also as others noted it is way too fast to be true on SunSpider, so this is obviously due to the usual browser differences and tricks (unlike the other results the 5Y50 scores are provided by Intel). People will be quite disappointed when the actual CPUs become available as they of course won't be twice as fast as Surface Pro!frostyfiredude - Wednesday, September 10, 2014 - link
Speculating here: Core M was ran on IE11, other Surfaces in Chrome. I get 115ms with my SP2 in IE11. For a quick test, being a hair faster than a 4200U based device makes sense as they turbo to the same frequency so the few percent better IPC will pull it ahead.OrphanageExplosion - Wednesday, September 10, 2014 - link
Thanks for the data - yes, like-for-like is kinda important...Krysto - Wednesday, September 10, 2014 - link
Isn't that the "higher end" version of Core M? Doesn't it mean it uses more than 4.5W TDP then?JDG1980 - Wednesday, September 10, 2014 - link
So when can we expect the fanless Surface Pro 4?mkozakewich - Wednesday, September 10, 2014 - link
Wow. Just wow. I'd been thinking of the Surface Pro 4, and I assumed it would be like having the current i3 at the current i5's performance (plus a little).Maybe they actually *can* redo the chassis again to make it fanless?
(As a buyer of the original Surface Pro, I'm kinda eyeing the market, and this might be the difference between buying in February and buying in October.)
evilspoons - Wednesday, September 10, 2014 - link
Good thing I bought the replacement warranty on my Surface Pro 3. I might have to "accidentally" break it when these show up in a revised SP3 or a SP4. Holy crap.Homeles - Wednesday, September 10, 2014 - link
"One of the biggest surprises for me was how much smaller the PCB was in Llama Mountain. For reference, the photo above is from our Surface Pro teardown. The PCB takes up almost the entirety of the tablet, and there's a great deal of extra thickness from the fans in the device."To be fair, that Surface Pro board isn't very dense at all. Look at Apple's Haswell-equipped logic boards, and you'll get a really good idea of what can be done with Haswell. Still, that Broadwell board is much smaller, and it'll be amazing to see how much more they can condense things over the next few generations.
jhoff80 - Thursday, September 11, 2014 - link
Yeah, a better comparison would've probably been the Pro 3 board, which is small in itself, and Broadwell is still much smaller. Pro 3 (from iFixit):https://d3nevzfk7ii3be.cloudfront.net/igi/nEKPhsVy...
Infy2 - Thursday, September 11, 2014 - link
That sun spider test for Core M has been run under IE and Surface Pros on Chrome obviously. Running that test on 5Y70 on Chrome would give score around 172. Optimally one should not compare web benchmarks between different browsers. IE is heavily optimized for sun spider while in other tests looses to Chrome almost twofold.aryonoco - Thursday, September 11, 2014 - link
I wonder what's the chances of someone picking Core M up and using it in a high-end Chromebook.Probably not much, manufacturers are scared of high-priced Chromebooks after the Pixel, but I think the market has markedly changed, and if someone picked this up, put it in an attractive-but-not-over-the-top body and matched it with a nice keyboard and trackpad and a 1080p IPS display and sold it for $699, it would do very well. If an OEM excluded fancy things like WiGiG or Wireless Charging or touchscreen panels or any of that, and sticks to the basics, I think it would be possible to hit that price and still have decent margins.
Speedfriend - Thursday, September 11, 2014 - link
Where are all those people that were claiming the A8 was going to be as fast as this? This looks frankly fantastic and I find it hard to believe that an iPad Pro is on the cards against a tablet like this. Watch Apple surprise us and release a touch enableled MacOS tablet running this chip!willis936 - Thursday, September 11, 2014 - link
That really would be a surprise because it would make half of their current product line obsolete.kron123456789 - Friday, September 12, 2014 - link
this chip itself costs more than half iPad.Shadowmaster625 - Thursday, September 11, 2014 - link
With such a simple and modest pcb we should be looking at a complete tablet BOM of around $100 for a 7" with 32GB of NAND. Plus the cost of the cpu of course. Now the question is, how is intel going to charge $200 for the cpu. In the past they have always offered "good enough" celerons and pentiums for a <$100 tray price. (Such as the 3560Y) So when the first broadwell sub 5 watt pentium is released, they might actually start gaining market share.iwod - Saturday, September 13, 2014 - link
4.5W TDP, Imagine additional work going into this for 10nm, it may well fit into a phone.simard57 - Monday, September 15, 2014 - link
in comparing the PCB from Pro 3 to the one in the new tablet, does that mean I can expect the extra space to be replaced by a larger battery so we get a double win - less power hungry AND larger battery for a much increased tablet battery life without an increase in weight!Sweet!
ewpelleg - Tuesday, September 16, 2014 - link
My SP3 gets 100ms for the Sunspider Test in IE. What's up?