Broadwell-M specification

[mikk]

So it maxes out at 1.1 GHz.

Are you trolling?

 

[Fjodor2001]

It maxes out at 2.6GHz.

Then they could have set the turbo to 5 GHz, so it could peak at that for a few milliseconds before starting to throttle. And you would have said it maxes out at 5 GHz. Come on, get real.

It's at 1.1 GHz, and that's not sufficient for a MBA laptop. It's good enough for tablet duty though.

 

[jpiniero]

It's at 1.1 GHz, and that's not sufficient for a MBA laptop. It's good enough for tablet duty though.

Believe it. Apple wants a fanless Retina MBA, and this is the only option available to them at the moment. Apple hasn't merged OSX and iOS yet so A8 isn't an option.

Edit: Or they could use Airmont, but that's not coming out until next year, and the rMBA is coming out this year.

 

[witeken]

Then they could have set the turbo to 5 GHz, so it could peak at that for a few milliseconds before starting to throttle. And you would have said it maxes out at 5 GHz. Come on, get real.

It's at 1.1 GHz, and that's not sufficient for a MBA laptop. It's good enough for tablet duty though.

Well, that's exactly what everyone does. Or do you think Qualcomm's chips really max out their TDP at 2.5GHz? They don't; they do a lot earlier. Everyone does this, and Intel also gives the base clock. Bay Trail has a base clock smaller than 1.5GHz, but in reality, it easily does max out its TDP at 2.4GHz.

But that doesn't matter in the end. My point is that the performance will be limited by the available thermal headroom. The real question is how much performance does this chip have within this power budget. And the fact is that Intel has a massive advantage with the most energy efficient transistors in existence.

So if your Apple MBA has any thermal headroom left at 1.1GHz, Core M has another 1.5GHz left to do so. If that's not enough, Intel also has Broadwell-U.

 

[Fjodor2001]

Believe it. Apple wants a fanless Retina MBA, and this is the only option available to them at the moment. Apple hasn't merged OSX and iOS yet so A8 isn't an option.

Edit: Or they could use Airmont, but that's not coming out until next year, and the rMBA is coming out this year.

Yes, maybe you're right. After all the lowest end Haswell MBAs are 1.3/2.6 GHz.

I guess it depends on how long the Broadwell-M can turbo to 2.6 GHz @ 4.5W before throttling, and how much better the IPC is on Broadwell vs Haswell.

I wonder what tasks it will be sufficient for though. Perhaps mail, web browsing, and such?

 

[Fjodor2001]

I wonder when Intel will go public on details about Broadwell. For previous CPU generations usually that was way ahead of actual launch. But now we're just ~2.5 months away from launch. So now I'm thinking Intel won't be presenting details until IDF in September, i.e. more or less at launch date...

 

[meloz]

http://www.cpu-world.com//news_2014/2014070201_Details_of_first_Broadwell_Y_mobile_processors.html

Imagine if we could get these for desktop, socketed or otherwise. These CPUs are ideal for fanless HTPC and also general internet browsing, office work etc. Sadly Broadwell is not coming to desktop, AFAIK, and a Skylake build will likely be more expensive than its hypothetical Broadwell equivalent considering migration to DDR4.

 

[AtenRa]

how are you getting those estimates?

Have a look at ATOM Z3770 (4W TDP) vs Core i5 4202Y (11,5W TDP). The ATOM is close to 70% the performance of the 11,5W TDP Core i5.

Im expecting the new 14nm ATOM to be very close to 80% or more(CPU) than 14nm Broadwell-Y at the same TDP.

Also, Haswell 11.5W TDP SKUs are priced at $280, ATOM is priced at $35. That is 8x times the price for ~30% more performance, ouch

 

[jdubs03]

Have a look at ATOM Z3770 (4W TDP) vs Core i5 4202Y (11,5W TDP). The ATOM is close to 70% the performance of the 11,5W TDP Core i5.

Im expecting the new 14nm ATOM to be very close to 80% or more(CPU) than 14nm Broadwell-Y at the same TDP.

Also, Haswell 11.5W TDP SKUs are priced at $280, ATOM is priced at $35. That is 8x times the price for ~30% more performance, ouch

Yeah there is quite a large disconnect there in terms of price per performance. I wouldn't expect the new Y-series chips to be any cheaper, so the ratio of $/perf is in flux, especially if performance suffers compared to Haswell-Y 11.5W.

The Haswell-Y's (i5) at 11.5W were all clustered around 1.5-1.6 Ghz with turbos varying from 1.9-2.3 Ghz. So, the only chip that would be comparable to those chips would be the 5Y70, it has a higher turbo, which could yield higher benchmark figures.

Point is: Broadwell-Y has to be cheaper if they want to raise volumes, especially if performance stays the same compared to Haswell.

Those figures paint a positive picture for Broadwell-U though, I expect clock's and turbos to be raised considerably at the 15W level.

 

[Fjodor2001]

But what about the iGPU in Atom Z3770 vs Broadwell-Y at 4.5 W? Will it also have 80% of the iGPU performance?

 

[witeken]

The IGP of Broadwell-Y and Cherry Trail will be the same.

 

[Fjodor2001]

So someone could easily have made an Atom based variant of the coming Broadwell-Y based fanless MBA, at 80+% of the performance but much cheaper and released much earlier?

 

[witeken]

Yes, but the single threaded performance of Atom is much less than 80% of Core's.

 

[jdubs03]

Yes, but the single threaded performance of Atom is much less than 80% of Core's.

I think what we all need to recognize is the specific Core SKU. Atom should always be compared to Y-series chips. So for that case single-thread performance is about 50-70%. I would expect that ratio to remain the same (as I have mentioned before, reiterating Krzanich's statements of the perf ratio between Core and Atom), but the TDP's have now changed, so I'm not as certain that the ~50% ratio will hold; if it does, it will be the 5Y70 SKU that upholds that ratio.

iGPU perf should beat everything else on the market. It's CPU perf that may not (Logan 64bit).

 

[witeken]

Krzanich's statements of the perf ratio between Core and Atom)

Source?

iGPU perf should beat everything else on the market. It's CPU perf that may not (Logan 64bit).

Compared to Airmont?

 

[beginner99]

Starting at $600 means Core M's price isn't low enough for something like an iPad Air or an Android tablet.

Why should it be? It's far better than any ARM SOC especially CPU wise.

 

[jdubs03]

Source?


Compared to Airmont?

http://blogs.barrons.com/techtraderdaily/2013/11/21/intel-ceo-krzanich-talks-expanded-foundry-biz-new-broxton-sofia-atom-chips/

There is too much uncertainty with regard to the ratio of Broadwell-Y and Airmont, it would be logical that Intel would be concentrating on keeping the ratio the same. Otherwise then we would have to compare Airmont to Broadwell-U, and the ratio is significantly higher than 2x, for single-thread.

I'm optimistic about Logan-Denver, I think that they may have a chance even if Intel has years of experience on them. It will have higher single-thread perf than Airmont assuredly, multi-thread I doubt it. The A8 could be quite close to maybe a bit more than Logan, only because of the switch from 28nm to 20nm. uArch wise Tegra should win out this time around (against the A8).

 

[SAAA]

Have a look at ATOM Z3770 (4W TDP) vs Core i5 4202Y (11,5W TDP). The ATOM is close to 70% the performance of the 11,5W TDP Core i5.

Im expecting the new 14nm ATOM to be very close to 80% or more(CPU) than 14nm Broadwell-Y at the same TDP.

Also, Haswell 11.5W TDP SKUs are priced at $280, ATOM is priced at $35. That is 8x times the price for ~30% more performance, ouch

You're paying for the best here, single thread especially, possible in a tablet fanless form factor. The premium is big but understandable.
C'mon single haswell at 2.6Ghz won't be matched by any other soc in years.

The IGP of Broadwell-Y and Cherry Trail will be the same.

Isn't the GT2 of Core M a 24EUs part? Maybe it's clocked lower than the 16EUs of Cherry Trail to save some watts but should still beat it with 50% more units (like HD5000 vs HD4600 in constrained TDP).

 

[SAAA]

http://blogs.barrons.com/techtrader...ded-foundry-biz-new-broxton-sofia-atom-chips/

There is too much uncertainty with regard to the ratio of Broadwell-Y and Airmont, it would be logical that Intel would be concentrating on keeping the ratio the same. Otherwise then we would have to compare Airmont to Broadwell-U, and the ratio is significantly higher than 2x, for single-thread.

I'm optimistic about Logan-Denver, I think that they may have a chance even if Intel has years of experience on them.

Remember that Goldmont is coming too and at the same clocks of Cherry Trail, so we should expect minimum some IPC impovements.
At the same time there's also Skylake that doesn't need to improve as much because it starts from already better performance (numbers are made up now):
Silvermont 50% +30% is 65% for Goldmont
Haswell 100% +15% is 115% for Skylake
Then the ratio is 100% better to 77% better, still far ahead and not bad when you add things like IGP, AVX, DDR4, etc.

Denver is interesting but the power and extreme widness make me think especially when you compare it to current 2/3-wide architectures...
Is it worth really to add all that much (diminishing returns, etc)?

 

[Qwertilot]

Apologies for being sceptical, but this does look a little bit like a hangover from their previous strategy of trying to stay with really high margin chips and ignoring atom.

If you compare this vs the 14nm atom then they'll be on the same process of course and apparently being given similar levels of attention in terms of design. Meanwhile atom is specifically meant to run in this sort of power budget while core is really being pushed.

A priori I can't see why you wouldn't expect the atom to end up better. Massive die harvesting or something?

Bigger single thread performance for this of course.

 

[mikk]

The IGP of Broadwell-Y and Cherry Trail will be the same.

Broadwell GT2 has 24 EUs and CHT 16 EUs (possibly 8 EUs initially).

 

[NTMBK]

Apologies for being sceptical, but this does look a little bit like a hangover from their previous strategy of trying to stay with really high margin chips and ignoring atom.

If you compare this vs the 14nm atom then they'll be on the same process of course and apparently being given similar levels of attention in terms of design. Meanwhile atom is specifically meant to run in this sort of power budget while core is really being pushed.

A priori I can't see why you wouldn't expect the atom to end up better. Massive die harvesting or something?

Bigger single thread performance for this of course.

The "bigger single thread performance" is the key. In a normal light workload there will be lots of brief bursts of work, and during these bursts the bigger core can (briefly) stretch its legs and really make a difference. It cuts down on things like UI lag considerably, and just gives a nicer feel to the tablet.

 

[raghu78]

Looks great. Also: "The 5Y10 supports a configurable TDP feature, which allows it to run at lower 4 Watt TDP." I wonder if there's any other difference between the 5Y10 and 5Y10a. Also a bit unusual names.

These processors should be a lot faster than any ARM SoC.

I wouldn't call that so easily. Apple A8 could be a game changer. TSMC 20nm brings a full shrink with 1.8x - 1.9x transistor density increase. Apple can easily double the CPU core count and clock them low enough to fit the 5w TDP. Four Cyclone cores clocked at 0.8 - 1 Ghz and with dual core turbo upto 2.0 Ghz should give this Broadwell M a run for its money. Also Apple could be integrating the PowerVR Series 6XT Rogue GX6650 which is expected to compete with Tegra K1 in GPU performance. Intel's graphics is not known for efficiency. Underestimate Apple at your peril. :biggrin:

http://www.anandtech.com/show/7793/imaginations-powervr-rogue-architecture-exposed/5
http://www.imgtec.com/news/detail.asp?ID=867

 

[witeken]

I wouldn't call that so easily. Apple A8 could be a game changer. TSMC 20nm brings a full shrink with 1.8x - 1.9x transistor density increase. Apple can easily double the CPU core count and clock them low enough to fit the 5w TDP. Four Cyclone cores clocked at 0.8 - 1 Ghz and with dual core turbo upto 2.0 Ghz should give this Broadwell M a run for its money. Also Apple could be integrating the PowerVR Series 6XT Rogue GX6650 which is expected to compete with Tegra K1 in GPU performance. Intel's graphics is not known for efficiency. Underestimate Apple at your peril. :biggrin:

http://www.anandtech.com/show/7793/imaginations-powervr-rogue-architecture-exposed/5
http://www.imgtec.com/news/detail.asp?ID=867

Die size won't really be the problem, but worse yields + much higher wafer costs + only ~1.3x lower power consumption would make such a chip expensive and power hungry. There's no way Apple could ever make a quadcore Cyclone or even more power hungry architecture with a 50% higher clock speed fit within a planar 20nm ~5W power envelope. Even your 1GHz base clock wouldn't be so easily done as you say.

You say that Intel isn't know for its graphics efficiency, which I could understand, but you (and we) don't know anything about Gen8, except that it's one of the biggest Gen updates ever and the number of EUs has quadrupled.

 

[raghu78]

Die size won't really be the problem, but worse yields + much higher wafer costs + only ~1.3x lower power consumption would make such a chip expensive and power hungry. There's no way Apple could ever make a quadcore Cyclone or even more power hungry architecture with a 50% higher clock speed fit within a planar 20nm ~5W power envelope. Even your 1GHz base clock wouldn't be so easily done as you say.

You say that Intel isn't know for its graphics efficiency, which I could understand, but you (and we) don't know anything about Gen8, except that it's one of the biggest Gen updates ever and the number of EUs has quadrupled.

TSMC 20nm brings 15% higher performance at same power or 30% lower power for same performance wrt TSMC 28HPM. btw this comparison is with 28HPM which is the best foundry process and not Samsung 28nm gate first which is inferior in transistor performance.

http://www.eda.org/edps/edp2013/Papers/4-4 FINAL for Tom Quan.pdf

slide 19

16FF/28HPM 16FF/20SoC

Speed @ same total power 38% 20%
Total power saving @ same speed 54% 35%

16FF 20SOC 28HPM
Speed at same Power 1.38x 1.15x 1x
Power at same speed 0.46x 0.71x 1x

So a 1.4 ghz Cyclone will draw 30+% lower power (given Apple is moving from samsung 28nm which has lesser performance than TSMC 28HPM). Also at the same power Apple can clock the A7 at 1.6 Ghz at TSMC 20nm vs TSMC 28HPM and most likely 1.7 Ghz when you compare with Samsung 28nm. This is without any power efficiency improvements to the core and any power management improvement like finer grained power gating etc. Apple also can easily have better turbo speeds if they implement a quality DVFS (Dynamic voltage and frequency scaling) system like the one found on Intel Baytrail/AMD Mullins.

By clocking the 4 cores at 1 Ghz on TSMC 20nm Apple can run at lower voltage and easily run 4 Cyclone cores within the same power as 2 Cyclone cores at 28nm. At 1.4 Ghz itself Apple has cut power by more than 30% by process transition. Add power efficiency improvements and better DVFS and Apple could even do better.

As for the GPU you are talking about PowerVR which is the leader in mobile graphics and power efficiency. Rogue 6XT will be competitive against Kepler K1. Intel graphics against Rogue 6XT is a mismatch which Apple will easily win.