IVY E uses solder tim as expected, see delidded (and killed) 4960X with solder tim

[tweakboy]

Can Ivy E do 5Ghz or 4.8Ghz ... Im getting the 4930k chip. thx gl

 

[Lepton87]

If 4.8GHz is very easy to achieve on that IVY-E I might buy one, my platform is aging or I might wait for Haswell-E. Did intel disclose how many cores will Haswell-E have for desktop? 8 cores for 500$ would be worth the wait on the other hand if only Extreme Edition gets 8 cores I won't buy it. 1000$ for one piece of hardware is too much. I would only consider it at that price if they released full EP core, which is supposed to have 10cores(12?) I heard conflicting rumors.



ps.
SOFTengCOMPelec

Could you please use normal black font in your posts? It's harder to read posts that use colorful fonts. People usually use colored font to highlight the most pertinent part of their post.

 

[aigomorla]

6) How do you drive a car without arms ? (Use Intel, rather than Arm)
7) All car accidents are caused by mechanical faults. The most common cause is the nut which is in the middle of the steering wheel, and holds it on the rack spindle. I.e. The nut behind the wheel causes accidents.

No worries!

8. Intel is telling the general public.. a Le Car is enough on the normal desktop regime...

 

[SOFTengCOMPelec]

If 4.8GHz is very easy to achieve on that IVY-E I might buy one, my platform is aging or I might wait for Haswell-E. Did intel disclose how many cores will Haswell-E have for desktop? 8 cores for 500$ would be worth the wait on the other hand if only Extreme Edition gets 8 cores I won't buy it. 1000$ for one piece of hardware is too much. I would only consider it at that price if they released full EP core, which is supposed to have 10cores(12?) I heard conflicting rumors.



ps.
SOFTengCOMPelec

Could you please use normal black font in your posts? It's harder to read posts that use colorful fonts. People usually use colored font to highlight the most pertinent part of their post.

Sorry, normal font engaged.

I think it is more likely that the 8-core version will be $999, rather than $499. I have seen stuff which seems to be saying that even the 6 core Haswell-E is going to be pricey (e.g. $799).
But it is probably all speculation at this point in time.

8. Intel is telling the general public.. A le car is enough on the normal desktop regime...

9) Intel decided to use the history books and read up on Ford model T cars.
So the new BestWell processor seriers has the following user features:-

1. You can have any color/colour TIM you like, as long as it's BLACK and a poor conductor of heat
2. You can have any clock speed you like, as long as it's 1 GHz
3. You can choose which model you want. 2 cores ONLY, 2 cores + 2 perm disabled cores or the delux 2 cores + 4 cores disabled
4. We are continually lowering prices. Our prices are so low now that they are the cheapest Intel(R) branded cpu's available for sale anywhere on the planet

 

[Lepton87]

Sorry, normal font engaged.

I think it is more likely that the 8-core version will be $999, rather than $499. I have seen stuff which seems to be saying that even the 6 core Haswell-E is going to be pricey (e.g. $799).
But it is probably all speculation at this point in time.

6 cores for 800$ would be an insane price, 8-cores at 1k$ would be actually a better value then that 6 core CPU. Historically EE CPUs were always horrible value compared to the CPU just below it. I think the worst we can expect is 6 cores priced the same as 3930K and 8 cores at 3970X price. It would be a shame that the only desktop 8 core CPU would be EE. Recently EE CPUs were only a smidgen faster then the tier just below at half the price. From profit margins perspective it would make sense, now Extreme CPUs are just terrible value and don't really offer anything over 3930K, but 33% extra cores and more cache would be a nice incentive to move to Extreme CPU.

 

[SOFTengCOMPelec]

6 cores for 800$ would be an insane price, 8-cores at 1k$ would be actually a better value then that 6 core CPU. Historically EE CPUs were always horrible value compared to the CPU just below it. I think the worst we can expect is 6 cores priced the same as 3930K and 8 cores at 3970X price. It would be a shame that the only desktop 8 core CPU would be EE. Recently EE CPUs were only a smidgen faster then the tier just below at half the price. From profit margins perspective it would make sense, now Extreme CPUs are just terrible value and don't really offer anything over 3930K, but 33% extra cores and more cache would be a nice incentive to move to Extreme CPU.

As a rule of thumb, I can't see Intel selling the cheapest 8-core Haswell-E part for less than 2 x 4 core model = 2 x 4770K (price) = Ballpark $700
but probably >>$700 because it is the Haswell-E platform.

In theory, it is probably considerably harder to make 8 "all working" cores on a single chip, rather than just 4 cores, because yield is considerably worse for 8. (Approx double the number of transistors, ALL have to be working, which is statistically considerably less likely).

Also, on the issue of pricing, I'm worried that DDR4 ram (especially high capacity, as there will only be 4 rather than 8 slots, so you need high capacity ram, just to get existing memory sizes, which are usually disproportionately more expensive), will be extremely expensive, at the time Haswell-E comes out.

 

[NTMBK]

In theory, it is probably considerably harder to make 8 "all working" cores on a single chip, rather than just 4 cores, because yield is considerably worse for 8. (Approx double the number of transistors, ALL have to be working, which is statistically considerably less likely).

Actually the yields on their 8 core die were so good that Intel had to make a special 4-core die for the 2011 platform, different both from the 2011 8-core die and the 1155 4-core and 2-core dies. http://www.anandtech.com/show/5276/intel-core-i7-3820-review-285-quadcore-sandy-bridge-e

 

[SOFTengCOMPelec]

Actually the yields on their 8 core die were so good that Intel had to make a special 4-core die for the 2011 platform, different both from the 2011 8-core die and the 1155 4-core and 2-core dies. http://www.anandtech.com/show/5276/intel-core-i7-3820-review-285-quadcore-sandy-bridge-e

Information SOURCE here

problems ranging from low-yields of 8-core Sandy Bridge-E processors

Also, Haswell-E will be new, which tends to worsen yields, at least initially, whenever a new process and/or design is started (no source, my own opinion).

 

[Idontcare]

Also, Haswell-E will be new, which tends to worsen yields, at least initially, whenever a new process and/or design is started (no source, my own opinion).

You are probably right, but just using words that state the matter it in a way that I am not familiar with...could you elaborate on what you mean here?

I might be able to confirm or deny your assertions (likely confirm, given you are an extremely logical individual) but also might be able to translate it into the word-speak of the fab people for your own future use.

 

[Khato]

Information SOURCE here

problems ranging from low-yields of 8-core Sandy Bridge-E processors

Also, Haswell-E will be new, which tends to worsen yields, at least initially, whenever a new process and/or design is started (no source, my own opinion).

Yeah, I'm sure it had nothing whatsoever to do with the fact that the C2 stepping which fixed VT-d wasn't going to be available until Q1 of 2012... (The fact that another stepping was necessary for the server parts to enabled VT-d showed up around a month after that article.)

 

[SOFTengCOMPelec]

Yeah, I'm sure it had nothing whatsoever to do with the fact that the C2 stepping which fixed VT-d wasn't going to be available until Q1 of 2012... (The fact that another stepping was necessary for the server parts to enabled VT-d showed up around a month after that article.)

The point that I was trying to convey, was that yields on single chip, 8 core processors, are potentially worse (i.e. more expensive) than e.g. single core processors.
Nothing about stepping's or bug fixes etc.

If you are trying to claim that everyone can go straight to 8 core, single chip, with out the slightest yield issue, whatsoever, then please explain why the Sony PS3 processor, has one core processor disabled on purpose, reportedly to reduce the costs by increasing yields. Since it means that up to one core can be faulty, and yet they can still sell it as a 100% working PS3 cpu.

 

[Khato]

The point that I was trying to convey, was that yields on single chip, 8 core processors, are potentially worse (i.e. more expensive) than e.g. single core processors.

Nothing wrong with that point at all. I just never would have guessed that such was your point given the context of the quote that you were responding to. But just because yields are twice as bad for an 8 core design versus a 4 core one doesn't mean that they're at all an issue. It all depends upon the defect rate of the process, and by the time that Intel is going into production on their large Xeon based products... well, it's safe to guess that their defect rate is quite low.

 

[SOFTengCOMPelec]

Nothing wrong with that point at all. I just never would have guessed that such was your point given the context of the quote that you were responding to. But just because yields are twice as bad for an 8 core design versus a 4 core one doesn't mean that they're at all an issue. It all depends upon the defect rate of the process, and by the time that Intel is going into production on their large Xeon based products... well, it's safe to guess that their defect rate is quite low.

Yes, "QUOTES" do sometimes end up bearing too little, about what the issue was.
I wish I knew more about the specific yields of Intel, AMD and the Arm related companies, but I never seem to ever come across information like that. There does seem to be lots of secrecy surrounding it, as far as I can tell.

 

[Idontcare]

Yes, "QUOTES" do sometimes end up bearing too little, about what the issue was.
I wish I knew more about the specific yields of Intel, AMD and the Arm related companies, but I never seem to ever come across information like that. There does seem to be lots of secrecy surrounding it, as far as I can tell.

The lack of info is on purpose. But you can get a good feel for the lower limits of their functional yields on the basis of reported D0 and known die-sizes.

 

[Lepton87]

8 cores Intel CPUs won't be 2x as big as 4 cores CPUs because the former lack IGP which can take up as much as half a die so I don't know if a native 8 core would end up any bigger then 4 core CPU+GT3+Crystalwell. However as far as I know there won't be any native 8 cores CPUs on 22nm process. Full IVY-EP will be 10 cores (some sources claim 12 cores) CPU.

http://wccftech.com/intels-leaked-r...epen-processors-12-cores-30mb-cache-130w-tdp/

 

[Khato]

I wish I knew more about the specific yields of Intel, AMD and the Arm related companies, but I never seem to ever come across information like that. There does seem to be lots of secrecy surrounding it, as far as I can tell.

There is indeed, and for good reason. You will on occasion see defect density graphs, typically designed to show how great the newest process is coming along. But those are usually just in relation to the previous process nodes and rarely put numerical values on the Y axis.

 

[SOFTengCOMPelec]

There is indeed, and for good reason. You will on occasion see defect density graphs, typically designed to show how great the newest process is coming along. But those are usually just in relation to the previous process nodes and rarely put numerical values on the Y axis.

[Deleted]...

Actually, I can understand the graph, now (select performance section on web page)

CORRECTION: I have looked again at the graph, and it now MAKES SENSE TO ME.

New Arm Architecture Performance Graph

8 cores Intel CPUs won't be 2x as big as 4 cores CPUs because the former lack IGP which can take up as much as half a die so I don't know if a native 8 core would end up any bigger then 4 core CPU+GT3+Crystalwell. However as far as I know there won't be any native 8 cores CPUs on 22nm process. Full IVY-EP will be 10 cores (some sources claim 12 cores) CPU.

http://wccftech.com/intels-leaked-r...epen-processors-12-cores-30mb-cache-130w-tdp/

I'm ONLY talking about 4 cores going up to 8 cores, on the SAME die. There may/may-not-be other stuff, which may or may not work on it, including faulty/disabled cores/IGP etc etc.

I was not going into the finer details, such as the IGP (which is not included on ALL 4 core, as they may sell ones with bust/disabled IGP), or possible bigger caches, or the fact that there can be even more than 8 cores for certain market segments (Xeon).

The cores you are referring to either are faulty or disabled, so I ignored them, as 'Haswell-E, consumer' is up to 8 cores, in that version (even if there are also possibly FAULTY cores as well).
The IGP is different as regards yields, because, as I understand it, it splits into different sections, which can be individually enabled or disabled (if faulty), hence the wide range of Intel SKU's.

So, I agree with you that it is not really a "pure" 4 core vs a "pure" 8 core situation. But think it would be making my posts way too long and complicated, if I went into every tiny detail for every post. So sorry if it caused any confusion.

The basic point was 8 cores is much bigger than 4 cores, without arguing about the exact % size increase that it incurs.

 

[SOFTengCOMPelec]

The lack of info is on purpose. But you can get a good feel for the lower limits of their functional yields on the basis of reported D0 and known die-sizes.

That's very interesting, thanks!

I C (excuse pun), so because we know that the basic defect rate would be somewhere around, 0.2 to 0.8 defects (my estimate) per square centre meter (you caught my interest too much, so I did some reading up on it), we can make an approximate guess as to the yield, per chip.

Analogy to Intels binning scheme. (The transistor is for automotive stuff, so this stilll counts as a car analogy).
(My understanding, and speculation) was that in the old days, they would have a basic component (transistor, spec, 100V, 1Amp, Gain (HFe) >= 100), and if 50% met all those specs, 500 out of every 1000 made, would be sold as an XYZ1.

To save throwing away the other 50%, they would bin sort them and have XYZ2 (only 50V max), XYZ3 (only HFe >=50), XYZ4 >= 0.5 Amps .

Hence XYZ1 = $0.10, and maybe XYZ2 etc, would be $0.08, for less demanding applications.

Therefore, they would only have to waste 15% (est), as the rest could still be sold, as long as ONLY one parameter was partly out of spec.

Intel seems to do the same. So that for each of their main logic blocks (where possible, some things are essential), there is a part which has (mostly) everything else working, except that particular logic block (e.g. Virtualization extensions).

Hence the (annoyingly) huge number of bewildering SKU's (Intel cpu units), and all the complaints along the lines of "why does my Intel lack feature xyz, when all the other cpu's seem to have it ?".

 

[ShintaiDK]

Features are not disabled due to defects.

Only cache is done on desktop and mobile chips. And cache plus cores on workstation and server chips.

 

[SOFTengCOMPelec]

Features are not disabled due to defects.

Only cache is done on desktop and mobile chips. And cache plus cores on workstation and server chips.

Maybe your are right, would you happen to have a source, as I would be interested to read it.

I've tried searching, but the only partial confirmation that (they do what I said to improve yields, is from another forum, and he does not make it 100% clear, either).

The reason so many choices exist is because Intel (and all other manufacturers) wants to maximize yields. A die that is perfect can be binned and sold for a premium, whereas

Another forum

Also, surely hyperthreading would also be done as well, especially as it is missing from most SKU's.

 

[ShintaiDK]

Maybe your are right, would you happen to have a source, as I would be interested to read it.

I've tried searching, but the only partial confirmation that (they do what I said to improve yields, is from another forum, and he does not make it 100% clear, either).



Another forum

Also, surely hyperthreading would also be done as well, especially as it is missing from most SKU's.

Features they disable are so tiny its extremely unlikely its due to defects. It would be, if anything, due to binning.

HT takes up less than 5% of a core, and most of those parts for non HT as well.

Example from P4:

 

[SOFTengCOMPelec]

Features they disable are so tiny its extremely unlikely its due to defects. It would be, if anything, due to binning.

HT takes up less than 5% of a core, and most of those parts for non HT as well.

I largely agree with what you are saying. I looked through the selectable/disable-able options (Intel website), and they don't really seem to cover that much functionality.

Although they may represent functionality, which Intel has found out from experience, to be the most problematic sections of the chip, most likely to cause it to have a defect failure.

On those lines, I have faint memories of reading a long time ago, that HT was just such a feature, and that it can cause major headaches for the chip manufacturer.

From what Idontcare was saying before, presumably, Intel would probably be VERY secretive about their strategy on improving yields, so maybe they do this, but it is NOT acknowledged publicly.

Anyway, thanks for the information (P4 diagram).

There is a good chance you are right.

 

[Michael Meio]

Sorry but I don't buy it.. Please have a look at this. It's the evolution of recent chip dies. I think I've never linked something on this forum and hope it works:

http://i1-news.softpedia-static.com...3-Haswell-Processor-Reportedly-Pictured-3.png

Those images used for this post don't look like a Haswell lid. It has no lips like all others so the LGA bracket can hold on. And the chip itself, looks more like an old model which maybe someone could identify.

One pic has a blade inserted while the TIM on top of the lid is intact -yeah, try that on a Sunday morning-, fully covering every detail regarding markings, model type, etc. On other images, the info is deliberately blanked.

Haswells are like Ivys and Sandys, rectangular shaped chips. Enthusiasts are cool peeps but not cool enough for Intel to make a totally different chip, retaining system and lid.

I call it a pile of ungulate dropping. -ʇıɥsןןnq :ןɐınboןןoɔ-

 

[Idontcare]

Sorry but I don't buy it.. Please have a look at this. It's the evolution of recent chip dies. I think I've never linked something on this forum and hope it works:

http://i1-news.softpedia-static.com...3-Haswell-Processor-Reportedly-Pictured-3.png

Those images used for this post don't look like a Haswell lid. It has no lips like all others so the LGA bracket can hold on. And the chip itself, looks more like an old model which maybe someone could identify.

One pic has a blade inserted while the TIM on top of the lid is intact -yeah, try that on a Sunday morning-, fully covering every detail regarding markings, model type, etc. On other images, the info is deliberately blanked.

Haswells are like Ivys and Sandys, rectangular shaped chips. Enthusiasts are cool peeps but not cool enough for Intel to make a totally different chip, retaining system and lid.

I call it a pile of ungulate dropping. -ʇıɥsןןnq :ןɐınboןןoɔ-

IB-E would be expected to be more square than rectangular because (1) it doesn't have the iGPU portion that the pedestrian IB has, and (2) it has more cores bolted onto the die versus that of the pedestrian IB.

It is true that Intel doesn't make a totally different chip just for enthusiasts, but they do make a totally different chip for their XEON product line and they repurpose that chip for use in their enthusiast platform.

 

[sushiwarrior]

One pic has a blade inserted while the TIM on top of the lid is intact -yeah, try that on a Sunday morning-, fully covering every detail regarding markings, model type, etc. On other images, the info is deliberately blanked.

ES chips have specific serial numbers, Intel sees this info posted outside of NDA, traces serial number on chip, this poster never receives another pre-release sample from Intel ever again. Not hard to figure that out.