Ivy Bridge-E or what happens when there's no competitions

[crashtech]

I would not assume the highest performing chips take less work to produce, nor that they are necessarily plentiful enough to give away at discount prices.

Apparently not enough people feel Intel's pricing is out of hand, note no mass exodus toward AMD.

 

[Torn Mind]

Cutting of features is quite easy, or at least it sounds that way: they just blow the fuses to features they don't want on the chip.

Although, a silicon wafer might have a bunch of bad chips and only a few are adequate to put onto the market while the rest are just completely discarded, not even fit for a Celeron...

http://en.wikipedia.org/wiki/Semiconductor_device_fabrication#Wafer_test

 

[colonelciller]

hmm.

When the highest end chip is the baseline design from which features are then systematically blocked off... that blocking off bit is extra work. The more of work that is done to the chip (block off a core on one model, a core + some cache on another model, 2 cores on another model, 2 cores + some cache on yet another, 2 cores + even more cache + alter the core clock speed)... well that all sounds like extra work to me, considering that these are all modifications of the baseline (high end) model.

as to why people are not flocking to another brand... well that's because there is no other brand with comparable performance to intel's Xeon CPUs. That's a really bad attempt to argue that people are happy with the price & market manipulation.

Not sure how anyone can defend price-gouging.

 

[crashtech]

I don't defend price gouging, but then again I don't accept definitions of price gouging that don't seem well founded. Plenty of CPU power is available to those who want it at very reasonable prices. It's rare that anyone would accuse a luxury car maker like BMW of price gouging when affordable vehicles are available to most, would you?

 

[Torn Mind]

hmm.

When the highest end chip is the baseline design from which features are then systematically blocked off... that blocking off bit is extra work. The more of work that is done to the chip (block off a core on one model, a core + some cache on another model, 2 cores on another model, 2 cores + some cache on yet another, 2 cores + even more cache + alter the core clock speed)... well that all sounds like extra work to me, considering that these are all modifications of the baseline (high end) model.

as to why people are not flocking to another brand... well that's because there is no other brand with comparable performance to intel's Xeon CPUs. That's a really bad attempt to argue that people are happy with the price & market manipulation.

Not sure how anyone can defend price-gouging.

It's a capital intensive task. After the program the computer properly, they put the chip in whatever it is and then apply a certain voltage to blow the fuses. I would think testing for things that shouldn't work is much easier and less time-consuming that validation.

http://www.google.com/url?sa=t&rct=...o-t9wwXN2APURcNig&sig2=Nrbb2q8VcezjLtD3gAY5qA

 

[colonelciller]

 

[colonelciller]

It's a capital intensive task. After the program the computer properly, they put the chip in whatever it is and then apply a certain voltage to blow the fuses. I would think testing for things that shouldn't work is much easier and less time-consuming that validation.

http://www.google.com/url?sa=t&rct=...o-t9wwXN2APURcNig&sig2=Nrbb2q8VcezjLtD3gAY5qA

interesting, thanks for the link

 

[bononos]

...... What do you want, for them to give them away?

No, just for them not to disable HT.

 

[colonelciller]

Interesting comparisons:

• E5-2643 --> 3x the price of the E5-1620
• E5-2620, 2630, 2630L.. nice cores, threads, cache... but dismal clock speed. Then jump up one more "model" to the E5-2637... nice clock speed, compensated for by blocking off all but 2 cores. Just bad.
• Do you need the cores/threads/cache of the 2620 and the clock of the 2637 (for your non-luxury non-commercial application), well that's available in another "model" the E5-2667 for double the price.

 

[2is]

It brings more cores I suppose. 8 or even maybe 10, versus 6 SB-E cores.

Are you guessing or is this confirmed? Serious question.

 

[cytg111]

Intel has so many software/OS/compiler guys that they probably could get any OS to run on x86.

Indeed and the incentive is quite obvious.
If anyone should be researching and developing the next great thing in programming paradigms/code concurrency it should be intel, cause we are hitting a brick wall here! Pretty soon the cake is eaten and we have nothing else in the oven :-/.

 

[Fjodor2001]

It was a pure example that ARM couldnt deliver both worlds. And that performance increase for ARM means higher power consumption.

Not at all. Every CPU could benefit from a big.LITTLE design.

Show me an Intel CPU that covers the same performance range that big.LITTLE does, with the same perf/watt as big.LITTLE has across that complete range.

 

[ShintaiDK]

Not at all. Every CPU could benefit from a big.LITTLE design.

Show me an Intel CPU that covers the same performance range that big.LITTLE does, with the same perf/watt as big.LITTLE has across that complete range.

Uhm...any?

Its just the range thats different.

 

[ShintaiDK]

Are you guessing or is this confirmed? Serious question.

IB-E/EP/EX dies comes in 6, 10 and 15 cores.

 

[ShintaiDK]

Intel probably sows the seeds of distrust when they manipulate chip prices as they do.
In order to see the blatant price manipulation you just need to look at the chips, their specs, and the prices.

An excel spreadsheet for Intel CPUs with:

• the chips listed in a vertical stack according to chip model
• all chip attributes arrayed to the right of said stack

... make such a spreadsheet (I have) and BAM, there it is clear as day. The insane price manipulation (for the exact same product... with cheaper "models" having extra work put in to deactivate cores and features) becomes clear as crystal within the Xeon product line.

The only explanation for the price structure is "milking" the market, which they can do because they have no competitor... they have a market monopoly.
Behavior like this by the one company who completely dominates the market breeds distrust... and thus this thread. If you wonder why people are skeptical of Intel's motives, Intel's past behavior is your answer.

You do know that CPUs have never been so cheap in history as they are now? if you want to see price manipulation and milking, turn the clock back to 2004-2006 and look on K8 prices. Specially X2.

 

[SunRe]

Uhm no. Your thread is an epic fail of misunderstanding.

Sorry, my mistake, I guess I just wanted to stir things a bit, that's all. I am well aware the E series is a byproduct of the xeon line and that considering how well the manufacturing process works at Intel, this is mostly a tactical move that has nothing to do with the consumer market but rather with their status in the server market.

I was also not implying that lack of competition makes Intel sloppy in their execution. I was only suggesting (didn't brought any arguments though) that if Intel would've had some strong competition in the server and desktop high-performance market maybe they would be willing to alter their profit margin in favor of say market-share.

Anyhow, glad the discussion took off, too bad that it's more on the arm vs x86 subject and not the E series. But I guess it reflects the current fractures in the market.

 

[Fjodor2001]

Uhm...any?

Its just the range thats different.

So the 2500K has as good low power characteristics as a Cortex A7 when used in a mobile device? Really?

 

[ShintaiDK]

So the 2500K has as good low power characteristics as a Cortex A7 when used in a mobile device? Really?

Look at the range, not the numbers. Unless you wish to change your original claim.

Its the delta between the lowest and highest operating point.

 

[Fjodor2001]

Look at the range, not the numbers. Unless you wish to change your original claim.

Its the delta between the lowest and highest operating point.

I want you to mention an Intel CPU that covers the performance range that big.LITTLE (e.g. Exynos 5 Octa) does, while keeping the same or better perf/watt and low power characteristics across that complete range.

The thing is that you can always do better if you can have two different cores types, each covering one separate section of the complete performance range.

 

[ShintaiDK]

I want you to mention an Intel CPU that covers the performance range that big.LITTLE (e.g. Exynos 5 Octa) does, while keeping the same or better perf/watt and low power characteristics across that complete range.

The thing is that you can always do better if you can have two different cores types, each covering one separate section of the complete performance range.

Everyone else but ARM picked the 1 core solution. Its also been debated before in other threads. And you keep bringing it up for a new debate. Seems ARM simply lacked the engineering capacity to do anything better.

I am pretty sure any Intel desktop/server CPU covers a much wider range than big.Little for example. With a factor of up to >100x between lowest and highest operating point.

And with Silvermont Atoms its pretty much set for ARM. One core to rule. Atoms today already compete with A7.

Even most custom ARM designers avoid the big.LITTLE.

 

[NTMBK]

Everyone else but ARM picked the 1 core solution. Its also been debated before in other threads. And you keep bringing it up for a new debate. Seems ARM simply lacked the engineering capacity to do anything better.

I am pretty sure any Intel desktop/server CPU covers a much wider range than big.Little for example. With a factor of up to >100x between lowest and highest operating point.

And with Silvermont Atoms its pretty much set for ARM. One core to rule. Atoms today already compete with A7.

Even most custom ARM designers avoid the big.LITTLE.

Using "people haven't done it that way before" is a fairly bad argument for a tech enthusiast forum!

One thing I'm curious about- how much die size does BIG.little add compared to hyperthreading? In a high-power-use mode where all 8 cores are enabled, it would have much the same effect- doubling thread count, but half those threads are less effective.

 

[ShintaiDK]

Using "people haven't done it that way before" is a fairly bad argument for a tech enthusiast forum!

One thing I'm curious about- how much die size does BIG.little add compared to hyperthreading? In a high-power-use mode where all 8 cores are enabled, it would have much the same effect- doubling thread count, but half those threads are less effective.

Arcording to Glenn Hinton. Hyperthreading added less than 5% to the core size.

Example from the P4.

 

[Fjodor2001]

Everyone else but ARM picked the 1 core solution. Its also been debated before in other threads. And you keep bringing it up for a new debate. Seems ARM simply lacked the engineering capacity to do anything better.

I am pretty sure any Intel desktop/server CPU covers a much wider range than big.Little for example. With a factor of up to >100x between lowest and highest operating point.

But we are talking mobile devices, and low power performance ranges. Also, it was acctually you that brought up the subject again in this thread starting with this post.

Also, this has already been discussed in detail in other threads. So it seems like we're just throwing old arguments at each other.

I think there are two ways at looking at this:

1. big.LITTLE is a great concept since it means that you can get a CPU that gets good perf/watt and low power characteristics across a wider performance range.

or:

2. The big.LITTLE concept had to be invented because ARM couldn't create a single CPU core type that gets good perf/watt and low power characteristics across a wider performance range.

But either way, you can always do better if you can have two different cores types, each covering one separate section of the complete performance range. So our original statement saying:

"Anytime ARM shows performance, the power consumption goes crazy. Else we wouldnt have the "octocore"."

is simply not true. It could be the reason they created the octocore (or actually the big.LITTLE concept), but it could also be 1) above. We don't know.

 

[ShintaiDK]

"Anytime ARM shows performance, the power consumption goes crazy. Else we wouldnt have the "octocore"."

is simply not true. It could be the reason they created the octocore (or actually the big.LITTLE concept), but it could also be 1) above. We don't know.

What do you call this then?

 

[Fjodor2001]

What do you call this then?

Not sure what you're trying to say?

As I see it that diagram just shows why 1) in my previous post is likely.

Intel could actually also benefit from a big.LITTLE design (but using two different types of Intel cores instead of ARM cores of course). Or can you show me an Intel CPU that can provide the same power/performance curve as above, but using only a single core type?