Intel Broadwell Thread

[witeken]

First of all, the supposed "myth" you are not even close to busting was the FACT that haswell had about the same IPC as Cyclone.

Okay, broadwell is a bit less superior to Cyclone.

You mustn't blame me for being able to read. It isn't my fault that Geekbench doesn't know how to read the clock speed of a CPU.

Second, It is and will always be a FACT

Nope. It seems you don't know what a facts it.

Third, you are using intel marketing slides and their word on clock speeds

I'm not using marketing slides. A slide that states the clock speed of a SKU is not a marketing slide.

and benching cyclone from 2013 against broadwell, a mid to late 2015 delivery product. That's not a good comparison, for many obvious reasons.

What does release date matter? BTW, the difference is only 1 year, not 2 years. Also, Broadwell = Haswell + 5% IPC, so in fact the difference is minus -3 months.

Furthermore, you ignore turbo frequencies and you rarely post anything that isn't an intel marketing slides. These are the facts.

I don't ignore anything.

Bust whatever myths you find after you stop taking intel marketing slides and doing mystery-math to get your weird numbers. The word that Cyclone IPC was as good as Haswell (or Ivy?) came from Anand himself. It's in his review, or do I need to link you? My guess is you have read it because I and many others have linked it to you specifically before when you went on these rants.

Sure, post your link from Anand that won't proof anything you say.

 

[witeken]

Edit: I doubt it will beat K1 or Denver and probably not A8 in GPU. Intel is strong in single core and multicore CPU but actually very weak in GPU generally but especially in mobile. I think Core is better than their mobile junk tho.

Mobile GPUs still have difficulties beating a 2.5 years old HD4000 with 16 Gen7 EUs.

 

[witeken]

edit: Multicore and GPU broadwell should outclass A7 but I don't know how anybody can expect Core M to beat Apple A8 when A7 was literally 2x as fast as A6 across the board (and 4x faster in GPU).

So Apple doesn't have to follow the laws of nature and can magically double CPU and GPU speed (it was 2X, not 4X) every year.

 

[witeken]

No, it's been pushed back to early 2015 for ultrabooks and even further for all the rest.

It hasn't been pushed back. Broadwell will be available in October.

The only people getting broadwell in 2014 will be Apple Consumers. I guess it counts as 2014 but I'd say it is more a 2015 product vs cyclone which sold a hundred million in 2013.
.

Did you miss IFA?

 

[liahos1]

i guess getting back on point. ipc or perf/mhz or whatever metric a7 looks very good. but it is still operating in a thermal limited environment w/ ipad and iphone. what could they do in a8.

is it realistic to expect a massive stepup ala a6 to a7. In that case they moved from 32nm to 28nm, they moved from 32bit to 64bit and they introduced a new core.

in this instance they go from 28nm to 20nm - which is a half node step. they are staying at 64bit. Could they put up a new core architecture? maybe implement turbo.

 

[jdubs03]

the good thing is we'll know broadwell-y vs. a8 in about a week. the 5y10 actually surprised me with its geekbench score (pleasantly), idk why the 5y70 had a lower score, i'll chalk that up to variance.

it's gonna be a close one between 5y10/a8 as far as geekbench is concerned.

 

[mikk]

Singlethread a doubling to Bay Trail Z3795 looks pretty good. Even Cherry Trail can't match this. Difference is too big.

 

[liahos1]

the good thing is we'll know broadwell-y vs. a8 in about a week. the 5y10 actually surprised me with its geekbench score (pleasantly), idk why the 5y70 had a lower score, i'll chalk that up to variance.

it's gonna be a close one between 5y10/a8 as far as geekbench is concerned.

i think intel17's thread on a7 vs sandybridge ipc had its share of detractors with respect to geekbench as a good indicator of cpu performance.

i do find it alarming though that people are expecting close cpu performance on a8 vs core m, despite the fact that intel is at a massive foundry advantage in terms of density vis a vis 22nm when they werent really in that position with 22nm vs 28nm. although perf/watt still needs to be determined. but at the end of the day intel's own slide show how relatively small the soc's are as a percent of total device tdp.

e.g. if we see core m faster than a8 by a marginal amount and battery life worse because of non soc design/os considerations, the average consumer is going to be like omg intel cpu's are worse than apple cpus. not a good place to be from a public opinion perspective.

that said i have a my surface pro 3 (i5) and I havent given my ipad 4 a second thought. If intel's partners can get some good designs out with core m and win 9 isnt a disaster then should be a great option for consumers. Also need these with android support for ppl who prefer inferior os's

 

[TreVader]

i guess getting back on point. ipc or perf/mhz or whatever metric a7 looks very good. but it is still operating in a thermal limited environment w/ ipad and iphone. what could they do in a8.

is it realistic to expect a massive stepup ala a6 to a7. In that case they moved from 32nm to 28nm, they moved from 32bit to 64bit and they introduced a new core.

in this instance they go from 28nm to 20nm - which is a half node step. they are staying at 64bit. Could they put up a new core architecture? maybe implement turbo.

Good point. It can't be known yet how well Cyclone will scale (or if it can).


I think it's still very much a mobile-only architecture. It's optimized for race to sleep vs Broadwell needs more legs. Not sure how hard it would be to do OSX on ARMv8

 

[Essence_of_War]

Not sure how hard it would be to do OSX on ARMv8

I would be SHOCKED if there wasn't a team at Apple testing OS X complied on a variety of ARM chips. They don't want to be permanently married to anyone's hardware if that hardware isn't guaranteed to do everything they want.

 

[Vesku]

Geekbench is just one benchmark app but it seems a 2GHz A7 wouldn't be that far behind Broadwell. Just reinforces my expectation that Haswell -> Broadwell will be similar to IvyBridge -> Haswell in terms of performance. Intel's 14nm is definitely helping them on the TDP and probably power efficiency front though. 2GHz Broadwell vs 1.4GHz Apple A7 both in tablets with Broadwell most likely having a very healthy lead in graphics power.

Still, we can see why Intel is worried about its markets. Apple A8 will be out soon and Nvidia is claiming its 64-bit Denver CPU is quite a bit faster than A7 especially in integer performance.

We'll probably see custom Qualcomm 64 bit ARM sometime next year. Not sure how the ARM reference A57 will compare to A8 and Denver.

Quite an industry switch up from 2-3 years ago when Intel thought Microsoft was holding it back somewhat and went through 2 to 3 revisions of testing Intel backed Linux based OSes. Now Intel seems to be relying heavily on the Windows x86 lock in to promote their CPUs with Windows tablets and 2 (3) in 1s while also working to price compete, spending money to gain product wins, in the Android market.

 

[witeken]

If you wanted to know how Broadwell would compare to Haswell, Intel has a slide for you saying it's 5% IPC difference.

 

[liahos1]

----- from an old post of mine

if you take the denver results from the ppt

Geekbench 3 Single-Core
Baytrail (Celeron N2910): 0.65x
S800 (Krait 400 8974AA): 0.80x
Tegra K1 (R3 Cortex A15): 1.00x
A7 (Cyclone): 1.20x
Haswell (Celeron 2955U): 1.20x
Tegra K1 (Denver): 1.65x

gets you a denver single core k1 score of 2050 = 1.65/1.2 x 1491

if you look at the i5 version of haswell at 2981 there still seems to be a pretty big performance gap even assuming slightly lower performance relative to i5 version of surface pro 3.

-- thats right on top of the core m result listed.

how is it possible they are doing this at 28nm in a tablet tdp....

although i guess we dont know tdp on those benchmarks or dice size / cost for denver k1.

 

[witeken]

It needs a 25% higher clock speed on a 2 nodes older process, though, so its performance per watt will be terrible. They probably used active cooling to keep it at 2.5GHz.

 

[Eug]

I want a Core M MacBook Air.

I want a Core M Retina MacBook.

 

[bullzz]

@Eug - I am torn on Core M macbook or wait for broadwell macbook pro

if single thread performance is twice that of bay trail, i think thats amazing

 

[ViRGE]

Take it down a notch, guys. Otherwise some of you will be getting infractions for thread crapping and hostility.

-Thanks
ViRGE

 

[IntelUser2000]

Based on this slide it seems that mobile CPUs are going to basically catch up with Intel chips in certain applications.

I think the point was put well in the Xeon E5 v3 article where it says in certain cases IPC gain is pretty much nothing, but some brings BIG gains.

Now Intel got the TDP pretty much matched up with the A7, but for its purpose(mobile applications and devices) Apple has an advantage regardless of what process its on.

Lots of sites are saying A8 will be a 2GHz CPU. It's not 5960X, QC DDR4, 1TB SSD, desktop, but average user desktop-class which few years ago were very powerful.

 

[jdubs03]

Here are the presentation slides for the Technology Insight: Intel's Next Generation 14nm Microarchitecture for Client and Server: https://intel.activeevents.com/sf14/connect/fileDownload/session/F7EBB2F85164E31DF41CDD863B86286F/SF14_SPCS002_101f.pdf

Enjoy parsing through this.

 

[IntelUser2000]

2Ghz Broadwell 2000

GeekBench scores scale pretty much linearly with clock speed. You can find same ratio on the Core i7 4770K as you do on the Core i3 15W parts.

Anything that scales linearly is suspect, but all mobile benchmarks do this, so for mobile application, A7 is the right design.

Perfect scaling means the bottleneck doesn't exist, which is flat out false with PC applications. You probably won't see A7 running PC applications, neither Core M running iOS applications though, so does it really matter?

 

[TreVader]

@Eug - I am torn on Core M macbook or wait for broadwell macbook pro



if single thread performance is twice that of bay trail, i think thats amazing

Don't wait. I don't think the people who bought MacBook airs in 2013 with haswell are regretting it. 13hr battery life!

 

[Eug]

@Eug - I am torn on Core M macbook or wait for broadwell macbook pro

if single thread performance is twice that of bay trail, i think thats amazing

Yeah, Bay Trail promised us this ultra-ultrabook market, but didn't quite come through. Core M seems to have achieved this though, pending real-world benchmarks, etc.

In the absence of real 3rd party benches, I'm trying to translate those slides:

---

The Intel Core M processor enables up to 50 percent faster compute performance and 40 percent faster graphics performance versus the comparable, previous 4th generation Intel Core processor.3 Consumers with older PCs will notice a more significant performance improvement. The Intel Core M processor delivers up to two times the compute performance and up to seven times better graphics compared to a 4- year-old PC, for example.2

---

This 4 year-old PC has an Intel Core i5-420UM, and the test is SYSmark 2014, which I guess makes sense since SYSmark is geared toward business applications.

---

2Intel® Core™ M-5Y70 Processor (up to 2.60GHz, 4T/2C, 4M Cache) vs. Normalized to a 4-year-old PC with Intel® Core™ i5-520UM. Performance based on SYSmark* 2014. Intel® Core™ M 5Y70 compared to Intel® Core™ i5-520UM. Weight based on Intel® Core™ M processor-based 2 in 1 based on Intel® FFRD Llama Mountain. Old PC is OEM laptop with Intel® Core™ i5-520UM and 62WHr battery, 3 lbs weight, 1.1-inch thick.

3Up to 50 percent faster vs. 4th generation Intel Core processors based on: Specfp_rate_base 2006 comparing Intel® Core™ M-5Y70 Processor compared to previous-generation Intel® Core™ i5-4302Y at 4.5W. Up to 40 percent faster graphics vs. 4th generation Intel Core processors based on: 3D Mark Ice Storm comparing Intel® Core™ M-5Y70 Processor with Intel HD graphics 5300 vs. Previous Generation Intel® Core™ i5-4302Y at 4.5W with HD Graphics 4200.

---

Also, to repeat the obvious, 4.5W is the SDP value for Core i5-4302Y. TDP is much higher at 11.5 W.


So, let's look at Intel Core i5-520UM:

http://ark.intel.com/products/47554/Intel-Core-i5-520UM-Processor-3M-Cache-1_06-GHz

1.066 GHz with turbo up to 1.866 GHz and 3 MB cache. Dual-core 4-thread. DDR3-800 dual-channel.

It gets 1336 in PassMark (or 615 single thread):

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5+520UM+@+1.07GHz


My 5-year old 2.26 GHz Core 2 Duo P7550 13" MacBook Pro gets 1468 (849 single) in PassMark:

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core2+Duo+P7550+@+2.26GHz


In comparison, the 4.3 W TDP 1.46 GHz N2805 Bay Trail CPU has a PassMark score of 497 (307 single thread). Ouch!

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+N2805+@+1.46GHz


The 2.41 GHz J2850 Bay Trail CPU has a PassMark score of 1831 but it has a TDP of 10 W.

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Pentium+J2850+@+2.41GHz

---

If Core M 5Y70 gets twice as much as Core i5-520UM in PassMark, that'd put it in the 2670 range, which would correspond roughly to a 1.7 GHz Core i5-2557M.

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i5-2557M+@+1.70GHz&id=814

That's a mid-range mobile CPU from 2011, with a max turbo of 2.7 GHz and a 17 W TDP.

In fact, that's the very CPU in the mid-2011 13" MacBook Air.

http://www.everymac.com/systems/app...acbook-air-core-i5-1.7-13-mid-2011-specs.html

---

I know that's a lot of internet handwaving, and PassMark isn't SYSmark, but if even close to being in the right ballpark then that's not too bad, esp. if has vastly improved GPU performance.

I would gladly sacrifice some modern mid-range fan-endowed laptop CPU oomph to get an 1.1 kg 11.9" Retina MacBook with super long battery life.

With my light business-type usage, even my Core 2 Duo 2.26 P7550 is OK, except for the fact that it doesn't have Retina, among other things. If a 1.1 GHz Core M gave me an 80% boost in real-world CPU speed, that's a big jump for me. Even if it is "only" a 50% boost in real-world CPU-speed, but gives me Retina and USB 3, I'll still gladly take the plunge, provided the rest of the machine is to my liking.

 

[Homeles]

Based on this slide it seems that mobile CPUs are going to basically catch up with Intel chips in certain applications.

I think the point was put well in the Xeon E5 v3 article where it says in certain cases IPC gain is pretty much nothing, but some brings BIG gains.

Now Intel got the TDP pretty much matched up with the A7, but for its purpose(mobile applications and devices) Apple has an advantage regardless of what process its on.

Lots of sites are saying A8 will be a 2GHz CPU. It's not 5960X, QC DDR4, 1TB SSD, desktop, but average user desktop-class which few years ago were very powerful.

Do keep in mind that Broadwell and Cherry Trail will be receiving very substantial frequency boosts. Any catchup from competitors will be rather temporary, although I guess with all the delays, that might not be the case.

When you're talking about TDP being matched, are you talking about Bay Trail, or Haswell? If it's the former, I'd have to disagree and say that Intel has a very substantial lead there, of course terribly undermined by the fact that they couldn't get Moorefield and the like to market at a decent time.

It's painful to see such impressive technology being held back by so much time. I remember reading an article somewhere where the author opined that Intel needs to cut back on their legendary validation in order to get to market quicker, and frankly, they're going to have to. They aren't pumping out designs once a year like they used to be. They really are going to have to focus so much harder than they already are on mobile, because the competition is just so vicious.

 

[IntelUser2000]

Do keep in mind that Broadwell and Cherry Trail will be receiving very substantial frequency boosts.

I can't quite agree with the former, except for maybe Core M. I think we'll see some boosts, hopefully more than the minute gains we had with Ivy and Haswell.

When you're talking about TDP being matched, are you talking about Bay Trail, or Haswell? If it's the former, I'd have to disagree and say that Intel has a very substantial lead there,

Well, I meant Broadwell against A7, probably A8. I am not sure if Bay Trail has the lead over A7, I mean the latter is in Smartphones. I do think most of the mobile SoCs are in the ballpark range, and Core M isn't far off either(probably leans toward the high TDP ones).

They really are going to have to focus so much harder than they already are on mobile, because the competition is just so vicious.

Ah, I don't know what the heck they are doing there. I came to accept in the long run Intel is a boring, old, stable company.

 

[Homeles]

I can't quite agree with the former, except for maybe Core M. I think we'll see some boosts, hopefully more than the minute gains we had with Ivy and Haswell.

Ah, well I meant for the ULV stuff, like the Celeron 2955U in that chart. That should go from 1.4 -> 1.7 GHz, or so.

And as competition starts to heat up, we may see Celerons that aren't so terribly neutered. Really, I wish turbo boost were enabled across all SKUs.

Well, I meant Broadwell against A7, probably A8. I am not sure if Bay Trail has the lead over A7, I mean the latter is in Smartphones. I do think most of the mobile SoCs are in the ballpark range, and Core M isn't far off either(probably leans toward the high TDP ones).

The A7 is actually a really power hungry chip, for what it's used in; it's just massively downclocked in the iPhone. Definitely a different approach.