[reuters 5-18-14] Intel CEO promises Broadwell on shelves for holidays 2014

[Fjodor2001]

You would be best served to not treat this as fact yet.

Who claimed it was a fact? Like a lot of other stuff on this forum it will not be a fact for sure until later communicated by Apple or Intel (if ever, or we just notice the delay). This is no different of course. But there are 50+ sites on Google News picking up the story.

It's funny how you always question the validity of articles only when the message does not appeal to you. It's really kind of pointless. People around here are educated enough to make their own judgment on the information, without you having to lecture them every time.

So, yes - from reading the article it has quite obviously not been confirmed by Apple or Intel yet. Now can we please leave that meta discussion aside?

 

[Fjodor2001]

Sure. But, the rumor does make sense if Apple planned on using anything but Broadwell-M. Why they would do that? I don't know. I always assumed Intel was focusing on Broadwell-M at Apple's insistence.

Yes, I agree. I also assumed Broadwell-M would be used for the Apple 12" fanless MBA. Maybe there will be two versions of the 12" MBA then, one fanless based on Broadwell-M, and one non-fanless based on Broadwell-U? But I've not heard anything about that. Or Broadwell-M is just delayed so sufficient volumes will not be available for the fanless 12" MBA until 2015. Any other possible explanations? :hmm:

 

[Phynaz]

It's funny how you always question the validity of articles only when the message does not appeal to you. It's really kind of pointless. People around here are educated enough to make their own judgment on the information, without you having to lecture them every time.

So, yes - from reading the article it has quite obviously not been confirmed by Apple or Intel yet. Now can we please leave that meta discussion aside?

So you agree that the article is nothing but rumor and speculation, but then attack me for pointing it out.

Your bias is showing.

 

[Fjodor2001]

So you agree that the article is nothing but rumor and speculation, but then attack me for pointing it out.

Your bias is showing.

It is quite obvious the article is not based on an official announcement from Apple or Intel, i.e. just like for a lot of other articles referenced on this forum. Since it's so obvious, there is no point in mentioning that every time someone references an article that is not an official announcement. Also, it's especially tiresome and biased that you only point it out when the message in the article doesn't suit you. If you instead would have complained on some technical aspect in the article sounding unreasonable, therefore making it less likely to be true, then I think that's another story. So that was my complaint, and that's all.

 

[Fjodor2001]

Core M 5Y10 2/4 0,8–2,0 GHz 100–800 MHz 1 600 MHz 4,5 W
Core M 5Y10a 2/4 0,8–2,0 GHz 100–800 MHz 1 600 MHz 4,5 W
Core M 5Y70 2/4 1,1–2,6 GHz 100–850 MHz 1 600 MHz 4,5 W

Just wondering, for how long is it likely the Core-M 5Y70 will be able to turbo to 2.6 GHz before throttling in a fanless slim Ultrabook design? 1s, 10s, 1m, 10m, ...?

 

[Roland00Address]

Just wondering, for how long is it likely the Core-M 5Y70 will be able to turbo to 2.6 GHz before throttling in a fanless slim Ultrabook design? 1s, 10s, 1m, 10m, ...?

We have no way to know that with the currently known information. The only way to figure this out is to actually test it, and one chip will not perform the same as another due to voltages as such. Furthermore it depends on the load you give the chip.

I have no doubt that you can make it throttle though. Anand could throttle the ipad air in about 60 seconds with his power virus test, and he could slowly make the ipad air throttle down with his more limited cpu intensive test that only targeted 2 threads. With his more limited throttle test the ipad air is slowly going down from 1.6 ghz to 1.4 ghz over the ten minute period.

http://www.anandtech.com/show/7460/apple-ipad-air-review/3

Turbo and throttling are pretty much the same thing with only minor differences. The only difference is intel is promising 800 mhz all the time with bursts to 2.6 ghz.

 

[Khato]

Just wondering, for how long is it likely the Core-M 5Y70 will be able to turbo to 2.6 GHz before throttling in a fanless slim Ultrabook design? 1s, 10s, 1m, 10m, ...?

That's more a function of the design of said fanless slim ultrabook/tablet than anything else. Even with the same steady-state thermal dissipation capability you could easily have one design that could last a a minute (just has good thermal coupling to a plastic chassis) before throttling while another lasts 10 minutes (good thermal coupling to other components, especially battery and display.)

 

[Fjodor2001]

Yes, I'm aware that the design of the unit will determine this to a large degree. But I'm just looking for "ballpark figures". Is it more likely to be 10s or 10m?

After all there should be experience from similar units that estimates could be based on. Also, to me it seems like 2.6 GHz is a rather high turbo frequency, so I find it hard to believe that it can last that long in a slim fanless Ultrabook chassis (MBA 12" style) at full load without throttling... :\

 

[bullzz]

@Fjodor2001

my guess - CPU perf will be atleast 20% more than silvermont. GPU pref should be 4X of bay trail

 

[dahorns]

@Fjodor2001

my guess - CPU perf will be atleast 20% more than silvermont. GPU pref should be 4X of bay trail

uh, this is Core. It better be more than that, otherwise it will be indistinguishable from cherry trail.

 

[jdubs03]

uh, this is Core. It better be more than that, otherwise it will be indistinguishable from cherry trail.

Maybe a 20% increase in IPC over Haswell makes more sense.

 

[teejee]

Maybe a 20% increase in IPC over Haswell makes more sense.

How could they do this? They've worked hard for years only to reach much lower IPC improvements, and now suddenly they will reach 20% improvement?
10% would impress me.

 

[bullzz]

@dahorns - i know this is core but I dont think its going to be 50% CPU perf over silvermont. also, from wat we have heard, cherry trail CPU is a shrink of bay trail

 

[Nothingness]

How could they do this? They've worked hard for years only to reach much lower IPC improvements, and now suddenly they will reach 20% improvement?
10% would impress me.

I agree that 10% would already be very good.

@dahorns - i know this is core but I dont think its going to be 50% CPU perf over silvermont. also, from wat we have heard, cherry trail CPU is a shrink of bay trail

Haswell Y CPU already are more than 50% faster than Silvermont. Here are Geekbench results for a 4200Y (1.9 GHz boost) vs a Z3770 (2.4 GHz boost).

 

[witeken]

It will be much closer to 5% IPC because Broadwell is a Tick.

 

[NTMBK]

It will be much closer to 5% IPC because Broadwell is a Tick.

Have we heard anything about CPU changes for Broadwell? It may even just be a straight shrink, Westmere style. (Not heard anything either way. *shrug* )

 

[ShintaiDK]

Have we heard anything about CPU changes for Broadwell? It may even just be a straight shrink, Westmere style. (Not heard anything either way. *shrug* )

There havent been a shrink without a performance boost. The design is further tweaked. Alot it contains new instructions, so its not a direct shrink.

 

[NTMBK]

There havent been a shrink without a performance boost. The design is further tweaked. Alot it contains new instructions, so its not a direct shrink.

Sorry, Broadwell contains new instructions, or are you referring to previous shrinks?

(But yeah, a shrink seems like an ideal chance to make minor tweaks, so we'll probably see at least a small performance boost.)

 

[ShintaiDK]

Sorry, Broadwell contains new instructions, or are you referring to previous shrinks?

(But yeah, a shrink seems like an ideal chance to make minor tweaks, so we'll probably see at least a small performance boost.)

Previous shinks usually had a few extra instructions too. But Broadwell certainly do.

 

[jdubs03]

Sorry, Broadwell contains new instructions, or are you referring to previous shrinks?

(But yeah, a shrink seems like an ideal chance to make minor tweaks, so we'll probably see at least a small performance boost.)

We should see a small boost for performance due to the any minor revisions, but the shrink should give a solid CPU performance boost overall, though that depends on how much of area is allocated to the GPU (20%).

We probably won't get a real idea until U-series SKUs come out in Q1 2015.

 

[Nothingness]

Previous shinks usually had a few extra instructions too. But Broadwell certainly do.

Indeed: they are listed in chapter 9 of Intel Architecture
Instruction Set Extensions Programming Reference, July 2012. Most of them are useful for cryptography and large integer arithmetic.

 

[witeken]

And there are lots of changes in the processor graphics of Broadwell. But the key feature of Broadwell is still the 14nm transistor, which brings a substantial increase in efficiency (>40%).

"I can show you the benefit of what 14nm is already delivering without a lot of tuning," said Skaugen. "We're seeing it comes to a 30% power reduction at the same performance level just by moving from the Haswell to the Broadwell SoC."
"And we're not done yet," explained Krzanich. "That's as far as we've been able to test it."

 

[Fjodor2001]

And there are lots of changes in the processor graphics of Broadwell. But the key feature of Broadwell is still the 14nm transistor, which brings a substantial increase in efficiency (>40%).

Hmm... didn't the article say 30% and not 40% ("We're seeing it comes to a 30% power reduction at the same performance level")?

Also, I wonder if that if valid for all chips. In that case they should be able to make a Broadwell version of the 4770K at 84W*0.6=50W TDP. Or 4590T 4C at 21W.

 

[NTMBK]

Indeed: they are listed in chapter 9 of Intel Architecture
Instruction Set Extensions Programming Reference, July 2012. Most of them are useful for cryptography and large integer arithmetic.

Aha, I hadn't heard about those, thanks!

 

[witeken]

Hmm... didn't the article say 30% and not 40% ("We're seeing it comes to a 30% power reduction at the same performance level")?

A 30% decrease in power equates to a 40% increase in power efficiency (1.4x lower power).

Also, I wonder if that if valid for all chips. In that case they should be able to make a Broadwell version of the 4770K at 84W*0.6=50W TDP. Or 4590T 4C at 21W.

This >30% reduction is obviously for any Broadwell-Y SKU and below (Airmont). At higher frequencies, I suspect the decrease becomes lower, but we don't have information about that. My guess is the TDP becomes 65-70W, if I take Ivy Bridge as reference.