solder vs. TIM (some noob questions)

[Anarchist420]

1. Why was TIM used for Ivy instead of the solder method which worked so well for Sandy?

2. Why doesn't nvidia use solder?

3. Would Ivy have run as cold as Sandy if intel had used AC MX2 and applied it well or does solder always beat TIM?

I'm sorry if any of these questions were dumb and I'll appreciate any answers

 

[ShintaiDK]

1. Because it fits the quality testing for repeated heatcycles, pressure etc. And most likely easier to apply.

2. They got exposed dies. So no solder needed.

3. No.

 

[IntelEnthusiast]

1. Because it fits the quality testing for repeated heatcycles, pressure etc. And most likely easier to apply.

2. They got exposed dies. So no solder needed.

3. No.

+1 for ShintaiDK.

I don't know all the reasons for the change to TIM but the simple fact is that the 3rd generation Intel® Core™ processors are running reasonable temperatures at stock speeds and it isnt until you start to overclock that you start to see any temperatures issues. Even overclocking them you are getting performance that is right up there with the 2nd generation Intel Core processors that are overclocked to higher speeds.

 

[moonbogg]

Question 1 is all that matters.

Answer: They did it to control overclockability and to make Sandy Bridge-E more attractive to the enthusiast market.

 

[nismotigerwvu]

http://forums.anandtech.com/showthread.php?t=2261855

Read through IDC's posts and you should have a better grip on this. Yes there are (major) improvements to be had with (major amounts of) time and work invested. As stated, the decisions made really didn't effect stock settings, but more or less eat into the headroom of the chips. While I personally feel Intel would be better served to toss some bones to the enthusiast crowd (since I'm a member of said crowd and could only stand to benefit), it really isn't going to effect their bottom line all too much. If they were losing performance wise to AMD, yeah sure sound the alarm, but they have been lapping them around the track for quite some time now and cheaping out/gimping the OC headroom on chips that are already the fastest X86 products on the market isn't going to bite them too hard. Chalk it up to "Intel being Intel".

 

[pantsaregood]

tin foil hat mob incoming

The TIM was a poor design decision, but it was likely more financially sound in some way. Intel has no reason to turn people towards LGA 2011 - it is a server platform. Enthusiasts account for an absolutely negligible amount of Intel's revenue, and even fewer of them would purchase an i7-3820. Ivy Bridge, even with better thermals, wouldn't hold a candle to six-core Sandy Bridge.

 

[BonzaiDuck]

tin foil hat mob incoming

The TIM was a poor design decision, but it was likely more financially sound in some way. Intel has no reason to turn people towards LGA 2011 - it is a server platform. Enthusiasts account for an absolutely negligible amount of Intel's revenue, and even fewer of them would purchase an i7-3820. Ivy Bridge, even with better thermals, wouldn't hold a candle to six-core Sandy Bridge.

. . . Which provokes the speculations of how IB-E will be released. Gulftown hex-core processors included the Intel 3000 iGPU, or so my sources indicate [I personally never had a Gulftown.] SB-E didn't. Do we know anything more specific about the anticipated IB-E?

 

[aigomorla]

http://forums.anandtech.com/showthread.php?t=2261855

Read through IDC's posts and you should have a better grip on this. Yes there are (major) improvements to be had with (major amounts of) time and work invested. As stated, the decisions made really didn't effect stock settings, but more or less eat into the headroom of the chips. While I personally feel Intel would be better served to toss some bones to the enthusiast crowd (since I'm a member of said crowd and could only stand to benefit), it really isn't going to effect their bottom line all too much. If they were losing performance wise to AMD, yeah sure sound the alarm, but they have been lapping them around the track for quite some time now and cheaping out/gimping the OC headroom on chips that are already the fastest X86 products on the market isn't going to bite them too hard. Chalk it up to "Intel being Intel".

i think he's talking about with the IHS intact.

IDC only showed what we used to do back in the old days called delidding.

And to be honest, on the new cpu's it might be a merit.
However on the i7's it was a waste of effort.

And op, whats better solder or tim?
Lemme ask you this question then.. what do you think can transfer better? metal or oil?
You just answered your own question.

Solder transfers better, and intel's solder they used to use, turned liquid at around 75-77C.

Also having a better transfer branch isnt worth it, if your sink cant keep up with the large delta to benefit from it.
Also delidding does come with its own can of worms... i have cracked a few DIE's before from mounting on a waterblock too hard for mount pressure...
Something which isnt possible if i had an IHS.

 

[ShintaiDK]

. . . Which provokes the speculations of how IB-E will be released. Gulftown hex-core processors included the Intel 3000 iGPU, or so my sources indicate [I personally never had a Gulftown.] SB-E didn't. Do we know anything more specific about the anticipated IB-E?

Gulftown did not have any iGPU.

Still no such thing as IB-E.

 

[BonzaiDuck]

Gulftown did not have any iGPU.

Still no such thing as IB-E.

Instead of going to the Intel website, I did a quick search at a reseller we all know and love . . . . checked the "details" tab. It said "Intel 3000 integrated graphics" or something close to that. Certainly, the reseller could have been wrong, and I didn't think worth it to search and further confirm.

But that's what it said.

There's an article recently posted on some site like "TechPowerUp" or Tech Report or some similarly named . . . reporting the planned release of an "IB-E" in mid-2013.

It wasn't someone's "forum post," but a review or article.

 

[pantsaregood]

Gulftown was on the LGA 1366 platform. Even if a GPU was on the Gulftown die (it wasn't), it couldn't be used because of the platform.

Some variants of Westmere had GPUs on them. None had HD 3000, however.

 

[BonzaiDuck]

Gulftown was on the LGA 1366 platform. Even if a GPU was on the Gulftown die (it wasn't), it couldn't be used because of the platform.

Some variants of Westmere had GPUs on them. None had HD 3000, however.

If such is surely true, then the reseller made a mistake with its product description.

I skipped socket-1366, although I initially thought I might build something with a 980X or similar Gulftown model, but hadn't looked at it since January, 2011.

So if you say so . . . [somebody should complain to those guys at the Egg.]

 

[BonzaiDuck]

Gulftown did not have any iGPU.

Still no such thing as IB-E.

I finally troubled myself to track down that news-item:

http://www.techspot.com/news/47849-ivy-bridge-e-delayed-until-second-half-of-2013.html

 

[Ferzerp]

I finally troubled myself to track down that news-item:

http://www.techspot.com/news/47849-ivy-bridge-e-delayed-until-second-half-of-2013.html

The supporting documentation there doesn't even suggest the existence.

An accurate story would say "no IB-E in 1H 2013" because that's all the documentation suggsts.

 

[ShadowVVL]

I am starting to think intel isn't going to make a IB-E and will move to haswell instead.

I dont see a point in launching IB-E When haswells launch is close.

 

[BonzaiDuck]

You can ask "What sort of self-interest would cause the Tech-spot folks to spin a complete falsehood?"

I can't answer these things, but I know that SB-E involved the release of socket-2011. Why would it be so troublesome to release an Ivy Bridge for socket-2011? Otherwise, the Haswell processor coincides with a new socket design.

As late as spring 2010, they released the last of Wolfdale processors for LGA-775. There is no way to make that a perfect analog to the socket-2011 SBE and some chance they would do the same with Ivy Bridge. But it happened.

I'm almost indifferent to whether this unfolds according to what I hear on the forums or according to the Tech-Spot item. I'm just saying I came across the web-page, and I don't see how their information would be a complete fiction or why it would be such.

Also, I can see what I can imagine is a mistake, since they say that Haswell will be "22nm." My understanding was that Haswell would be another die-shrink. Maybe I have that wrong, too . . .

 

[ShintaiDK]

You can ask "What sort of self-interest would cause the Tech-spot folks to spin a complete falsehood?" .

Considering how much technews these days are utter rumours and sensationalism in the hunt for ads revenue. I would say a huge interest.

Plus they usually copy/cat one another without any regular journalistic critisism and tend to forget to link their "source".

There is zero information in that article that support that there will be an IB-E. "Well its not on the roadmap, so we just think its delayed.".

Or simple BS like this:

The slide also indicates that the next generation Haswell micro architecture, the successor to Sandy Bridge, is being pushed back until the first half of 2013.

How can it be pushed back if its on schedule? And the slide even says march-june launch.

 

[ShintaiDK]

I am starting to think intel isn't going to make a IB-E and will move to haswell instead.

I dont see a point in launching IB-E When haswells launch is close.

Haswell-E/EN/EP will most likely come in late 2013 or early 2014.

 

[ShadowVVL]

Haswell-E/EN/EP will most likely come in late 2013 or early 2014.

assuming hasswell will be 30% faster then IB quad. Wont haswell quad be almost as fast as IB-E since IB quad is only about 30%-40% slower that SB-E in multi threaded apps.

so wouldn't haswell quad be a better value in performance/watt/ price?

 

[pantsaregood]

30% faster? Haswell won't be more than 15% faster than Ivy Bridge per clock - I doubt it will manage that much.

Ivy Bridge-EX/EP/EN (provided they're released - I would assume they will be) parts at 6+ cores will very easily outperform quad-core Haswell.

 

[ShadowVVL]

30% faster? Haswell won't be more than 15% faster than Ivy Bridge per clock - I doubt it will manage that much.

I am assuming it will be 30% faster.Haswell hasn't come out yet so we don't know how fast it will be.

 

[pantsaregood]

Intel hasn't made a revolutionary step like that in an incredibly long time.

Pentium III -> Pentium M -> Core -> Core 2 -> Nehalem -> Sandy Bridge all yielded around a 15% performance improvement. Furthermore, Intel is concerned with doing things other than increasing performance/clock with Haswell. There is exactly a 0% chance that Haswell will be 30% faster than an equivalently clocked SB/IB CPU - it also won't be capable of reaching significantly higher clocks than Ivy Bridge.

 

[Anarchist420]

Intel hasn't made a revolutionary step like that in an incredibly long time.

Pentium III -> Pentium M -> Core -> Core 2 -> Nehalem -> Sandy Bridge all yielded around a 15% performance improvement. Furthermore, Intel is concerned with doing things other than increasing performance/clock with Haswell. There is exactly a 0% chance that Haswell will be 30% faster than an equivalently clocked SB/IB CPU - it also won't be capable of reaching significantly higher clocks than Ivy Bridge.

Some apps will be more than 30% faster on Haswell because of avx2. Avx2 is a giant leap over avx. I would guess that Haswell will have 25% higher performance per clock than IB on average because of TSX and all of the things brought with avx2 like FMA/scatter/gather. I don't know what the cache latency will be, but theoretical performance of avx2 is significantly higher than that of avx so unless the cache latency is significantly worse than IB, then there should be at least a 50% performance increase on average in apps that use avx2 over avx.