Confirmed: Haswell-E uses Soldered TIM

[lutjens]

While I understand why you are concerned, you have to understand that if Intel wants to hit a certain TDP, they have 2 options on a given node. They can either reduce the core count, or reduce the clock speed of said cores (assuming we keep cache sizes/iGPU shader counts etc consistent I am referring just to the CPU side)..Since we all want an Octacore in non Xeon trim, then Intel has to keep the clockspeed down to hit a certain TDP..

Now we all know even the worst silicon has headroom over the "stock" clocks, it just depends on how much voltage you are willing to run through it. Intel reserves clock speed so that they can offer some "value" in the form a refresh (see the 2600K to 2700K, 4770K to 4790K etc)..

Lastly, since they are K SKUs, it is quite easy to go into the BIOS and adjust 1 setting to increase the clockspeed via the multiplier, even if you want to keep the stock Vcore..You will most likely be able to squeeze out at least a few hundred Mhz more unless these chips are complete dogs..

I was 99% certain I was going to move to X99 and get the 8 core flagship part (never owned an "EE" part and while I don't want to drop 1K that is the only way to play these days) when they were announced, the closer we get to launch the more I am worried. If the 8 core parts do come in @ 3Ghz stock, and we assume the best parts are able to reach a 50% OC (which is a lot), then we are looking at ~4.5Ghz for the platinum center wafer cut chips..

While I have the cooling to take said CPU to that level, finding one is going to be the challenge..And then I have to look at losing 200~300Mhz of single thread performance (vs my sig rig) since the IPC jump of Haswell isn't much over Ivy..I think the 8 core parts are going to reintroduce the old "batch hunting" we used to do when trying to source the best OC'ing parts, since we know the 8 core SKUs are being harvested from 12 core dies that don't make the cut..

A 12-core chip that has been harvested into an 8-core chip may clock just fine. The reason why the cores were disabled may just be due to a defect in the cache, one of the largest areas of the die and the most likely where a defect will occur. Even if a core is defective and is deactivated as a result, it may not be indicative of an inability of the other functional cores to overclock well and may actually overclock well (it may have been a high clocking die, except for a defective core or two). It's really tough to say for sure, though, as the reasons why a core is disabled are very numerous and more than one may apply to a given die.

 

[dr/owned]

Retail versions of the 5960X are going to be native 8 cores. 6 core versions are going to be 8 cores with 2 cores disabled. The article surmises that we might see 10 or 12 core retail parts. This is a flawed assumption based on the results of delidding 1 engineering sample part.

Financially there's no reason to sell a 12 core with 4 cores disabled. The 12 core die with defeatured cores can still be sold as 10/8/6/4 core Xeons which would likely cost more than a 5960X sells for anyways. 2643v2 for example...6 cores for $1500.

And also, that guy didn't really "delid" the part. He ripped the die off the substrate.

 

[lutjens]

Retail versions of the 5960X are going to be native 8 cores. 6 core versions are going to be 8 cores with 2 cores disabled. The article surmises that we might see 10 or 12 core retail parts. This is a flawed assumption based on the results of delidding 1 engineering sample part.

Financially there's no reason to sell a 12 core with 4 cores disabled. The 12 core die with defeatured cores can still be sold as 10/8/6/4 core Xeons which would likely cost more than a 5960X sells for anyways. 2643v2 for example...6 cores for $1500.

And also, that guy didn't really "delid" the part. He ripped the die off the substrate.

I think that this all depends on what Intel's yield is and how much demand they have for Xeon 12-core chips. A high 12-core demand would require the running of more 12-core dies, the quantity of inevitable defects that would result from running that number of 12-core wafers could potentially outstrip the demand for the 10-core SKUs (unlikely, but possible). Also, the defects in the cores could be so severe that the cores are unsuitable for even 10-core use, and need to be further neutered to be useful. In this unlikely case, it's possible some of these dies could theoretically become 8-cores. But you're right, they'd likely be labeled as 4/6/8-core Xeons, and perhaps marketed with a higher cache amount (which is a partial giveaway as to what die is used), and again this depends on yields.

The important grain of salt to retain of the photo of the IHS removed is the fact that there absolutely no evidence that the CPU in question was actually an i7-5960X. It's more likely, in fact, an early (and nearly useless) Haswell-EP 12-core chip that someone obtained and decided to remove the IHS, probably mostly for click bait.

 

[KazeoHin]

But will it clock? I'd like to see how far it can go under my Monsta 280 rad...

 

[lutjens]

But will it clock? I'd like to see how far it can go under my Monsta 280 rad...

I think they'll clock quite well, on a par with the 4 and 6-core chips (as long as your cooling is capable of keeping the extra heat under control)

 

[dr/owned]

I think they'll clock quite well, on a par with the 4 and 6-core chips (as long as your cooling is capable of keeping the extra heat under control)

This is the part that scares me. Back when I had a 3770K at 5 Ghz, by itself it was pulling 300W because I had to throw a ton of voltage at it. If we're talking 8 cores now....that would be 600W. That's pushing the limit of 2x 8 pin CPU power connectors. I haven't yet found the limit of my cooling setup....but I think the 5960X is going to probably find the limit of the waterblock.

 

[ccityinstaller]

Retail versions of the 5960X are going to be native 8 cores. 6 core versions are going to be 8 cores with 2 cores disabled. The article surmises that we might see 10 or 12 core retail parts. This is a flawed assumption based on the results of delidding 1 engineering sample part.

Financially there's no reason to sell a 12 core with 4 cores disabled. The 12 core die with defeatured cores can still be sold as 10/8/6/4 core Xeons which would likely cost more than a 5960X sells for anyways. 2643v2 for example...6 cores for $1500.

And also, that guy didn't really "delid" the part. He ripped the die off the substrate.

I highly doubt we are going to see a native 8 core die this time around..Just like with the move to the 22nm Ivy-EP SKUs, Intel has harvested the 8 core and many of the early 6 core parts from larger dies that have fused cores. Even with Sandy intel released some 6 core parts that were cut down dies. Anandtech published a small piece on this back when Ivy-EP hit IIRC...

I think they'll clock quite well, on a par with the 4 and 6-core chips (as long as your cooling is capable of keeping the extra heat under control)

While I want to believe, I am not sure we are going to be seeing the majority of 8 core chips doing 4.5Ghz (let alone higher) unless you are willing to feed them a Super Sized doze of Voltage..That of course then bring cooling into question, but if you can afford a 1K EE CPU, then you should be able to afford a custom loop to push it...

This is the part that scares me. Back when I had a 3770K at 5 Ghz, by itself it was pulling 300W because I had to throw a ton of voltage at it. If we're talking 8 cores now....that would be 600W. That's pushing the limit of 2x 8 pin CPU power connectors. I haven't yet found the limit of my cooling setup....but I think the 5960X is going to probably find the limit of the waterblock.

Jesus man, how much voltage were you putting through that poor 3770k? My sig rig is 24/7 stable under any load (3D, full synethic, Prime95 (with AVX) and F@H BA WorkUnits) and with my cpu loaded 100% and all the fans and my pump I don't pull more 230W (total system power @ my UPS)..

That is why I laughed at all of the AMD haters for telling the truth when telling they had 8 core silicon that could run @ 5 Ghz had a 220W TDP..They could have lied and used some typcial marketing BS but told the truth..This is what we have, this is the speed, and this is how much power you need for it.

Back on topic, why the hell do people here keep posting links to this sham of a tech site that rip stories/rumors/facts from every other site and forums (like here, OC.net, and Xtreme)..I refuse to even go there and give them a single page view..It makes me sick to think I am helping them make money in an industry that Kyle, and Anandtech created while doing .000001% of the work.

 

[dr/owned]

I highly doubt we are going to see a native 8 core die this time around..Just like with the move to the 22nm Ivy-EP SKUs, Intel has harvested the 8 core and many of the early 6 core parts from larger dies that have fused cores. Even with Sandy intel released some 6 core parts that were cut down dies. Anandtech published a small piece on this back when Ivy-EP hit IIRC...



While I want to believe, I am not sure we are going to be seeing the majority of 8 core chips doing 4.5Ghz (let alone higher) unless you are willing to feed them a Super Sized doze of Voltage..That of course then bring cooling into question, but if you can afford a 1K EE CPU, then you should be able to afford a custom loop to push it...



Jesus man, how much voltage were you putting through that poor 3770k? My sig rig is 24/7 stable under any load (3D, full synethic, Prime95 (with AVX) and F@H BA WorkUnits) and with my cpu loaded 100% and all the fans and my pump I don't pull more 230W (total system power @ my UPS)..

That is why I laughed at all of the AMD haters for telling the truth when telling they had 8 core silicon that could run @ 5 Ghz had a 220W TDP..They could have lied and used some typcial marketing BS but told the truth..This is what we have, this is the speed, and this is how much power you need for it.

I can assure you it's native 8 core

My 3770K was "only" 1.585V...I laugh at people who get scared about going > 1.3V. Haswell is more silicon/design limited than it is voltage / temperature limited (at least with a custom loop), so I doubt it will be necessary to use those kinds of voltages to get the max overclock the chip will take. Ivy I could throw 1.7V at it and get 5.2Ghz. Haswell I threw that same voltage and couldn't even get 4.8 on one chip.

 

[geok1ng]

I can assure you it's native 8 core

nothing like lifting an NDA to make such statements true. nobody can assure us of anything

 

[xorbe]

Isn't it beneficial to have dead silicon for distributing heat? (ie, 8 of 12 cores active, versus native 8 core package.)

 

[chrcoluk]

yeah ccityinstaller if they overclock well to 4.3ghz+ then my concern isnt a big deal, time will tell.

Of course I will never be buying one I only just got haswell and am not rich enough, so I shouldnt be concerned I guess

 

[Riccochet]

Considering moving from my 4670k to a 4690k. Probably won't see much improvement besides heat.

 

[schmuckley]

Have you ever actually compared how much people use when they say they use a "pea size drop" of TIM? I can pretty much guarantee that if yu compared with an actual pea, it would be a lot less.

The "pea" method has not been very good advice since the mid-90's. You generally don't get good coverage.

What I do it a drop in the middle and then smaller drops nearer the corners. I then install the heatsink, squish the paste around, remove the heatsink and add a bit anywhere I am not seeing full coverage. This tends to allow me to get a nice thin layer and also not use too much.

Since the IHS is generally not really flat unless you lap it, the whole spread with a gloved finger is not necessarily going to get you good coverage either since some places will actually need more than other places.

http://www.youtube.com/watch?v=e3rFigexx3c&feature=player_embedded
2:56
this is a joke

 

[drescherjm]

http://www.youtube.com/watch?v=e3rFigexx3c&feature=player_embedded
2:56
this is a joke

Oh my. I hope that was not electrically conductive compound. Although it looked like hand cream..

 

[lutjens]

To all the naysayers who said that the 8-core wouldn't overclock well...seems that 4.5GHz isn't too hard...and this was on a buggy preproduction BIOS with 1.3V.

http://www.overclock.net/t/1506847/intel-5960x-haswell-e-8-core-quick-observations


<returns to praying for unlocked 18-core CPU>

 

[geok1ng]

To all the naysayers who said that the 8-core wouldn't overclock well...seems that 4.5GHz isn't too hard...and this was on a buggy preproduction BIOS with 1.3V.

http://www.overclock.net/t/1506847/intel-5960x-haswell-e-8-core-quick-observations

<returns to praying for unlocked 18-core CPU>

Pray harder:

"CPU always seemed to be on a 60% CPU load...could not figure out why even thought when looking at processes it showed no processor load.....again more bugs..
Realtemp would only show 6 out of the 8 cores..."

No f way that 60% CPU load at 6 out of 8 cores is a hint of good overclockability. we still have the die issue: if it a 12 cores die, things look grim for OCers, because dual bus rings menas trouble to sync the signals paths between cores.

 

[lutjens]

Pray harder:

"CPU always seemed to be on a 60% CPU load...could not figure out why even thought when looking at processes it showed no processor load.....again more bugs..
Realtemp would only show 6 out of the 8 cores..."

No f way that 60% CPU load at 6 out of 8 cores is a hint of good overclockability. we still have the die issue: if it a 12 cores die, things look grim for OCers, because dual bus rings menas trouble to sync the signals paths between cores.

Realtemp would only show 6...all 8 cores were enabled....clearly a bug in Realtemp.

If it's a 12-core die (which I doubt), the dual rings function at a set clock speed which is the same between the two of them. They are both synchronized with that clock speed...what you are describing is a non-issue.

 

[R3MF]

It's interesting to me that the 8 core /16 threads is really a 12 core 24 threaded CPU with 4 cores disabled. Will they disable 1/2 the cores for the 6 core?

i believe that the six core is a native hex.

 

[sblantipodi]

I think that this is the minimum considering the fact that they are releasing a 3GHz CPU at the end of 2014 while there are CPUs running at 1GHz more with the same architecture since months.

 

[Zomoa]

Pray harder:

"CPU always seemed to be on a 60% CPU load...could not figure out why even thought when looking at processes it showed no processor load.....again more bugs..
Realtemp would only show 6 out of the 8 cores..."

No f way that 60% CPU load at 6 out of 8 cores is a hint of good overclockability. we still have the die issue: if it a 12 cores die, things look grim for OCers, because dual bus rings menas trouble to sync the signals paths between cores.

He figured it out what the problem was. Disabling C-states was causing that bug. He enabled C-states and the problem went away.

 

[lutjens]

This is the part that scares me. Back when I had a 3770K at 5 Ghz, by itself it was pulling 300W because I had to throw a ton of voltage at it. If we're talking 8 cores now....that would be 600W. That's pushing the limit of 2x 8 pin CPU power connectors. I haven't yet found the limit of my cooling setup....but I think the 5960X is going to probably find the limit of the waterblock.

Keep in mind that a good portion of your heat production is coming from the onboard video...these transistors do nothing toward helping the mainstream chips clock higher. Fortunately, the HEDT platform isn't saddled with such a thing, but is becoming more and more burdened with items that are a hindrance to high CPU clocks as Intel tries to integrate as much stuff onto the CPU as they can (like the FIVR).

To me this is the exact opposite of what should be done with the HEDT platform...everything that doesn't assist in increasing CPU performance should be de-integrated. This keeps transistor counts and the heat load down, allowing more headroom for more voltage and more clock speed. I understand the economic realities with integrating things on a CPU when it comes to the mainstream chips, but on the HEDT platform, integration should be minimized.

 

[sblantipodi]

Keep in mind that a good portion of your heat production is coming from the onboard video...these transistors do nothing toward helping the mainstream chips clock higher. Fortunately, the HEDT platform isn't saddled with such a thing, but is becoming more and more burdened with items that are a hindrance to high CPU clocks as Intel tries to integrate as much stuff onto the CPU as they can (like the FIVR).

To me this is the exact opposite of what should be done with the HEDT platform...everything that doesn't assist in increasing CPU performance should be de-integrated. This keeps transistor counts and the heat load down, allowing more headroom for more voltage and more clock speed. I understand the economic realities with integrating things on a CPU when it comes to the mainstream chips, but on the HEDT platform, integration should be minimized.

Hi, I see on your signature that you use Xeon processor on a gamer rig.
Do you see any performance improvement using xeon instead of using a quad core CPU?

 

[wabbitseason]

Keep in mind that a good portion of your heat production is coming from the onboard video...these transistors do nothing toward helping the mainstream chips clock higher. Fortunately, the HEDT platform isn't saddled with such a thing, but is becoming more and more burdened with items that are a hindrance to high CPU clocks as Intel tries to integrate as much stuff onto the CPU as they can (like the FIVR).

To me this is the exact opposite of what should be done with the HEDT platform...everything that doesn't assist in increasing CPU performance should be de-integrated. This keeps transistor counts and the heat load down, allowing more headroom for more voltage and more clock speed. I understand the economic realities with integrating things on a CPU when it comes to the mainstream chips, but on the HEDT platform, integration should be minimized.

Solid point and one most enthusiast users would agree with. The exception is when integration can lead to a faster system overall, even if the max overclock headroom drops somewhat. Most of these integration thrusts, though, won't lead to a faster system.

 

[Zomoa]

Solid point and one most enthusiast users would agree with. The exception is when integration can lead to a faster system overall, even if the max overclock headroom drops somewhat. Most of these integration thrusts, though, won't lead to a faster system.

I've done some extensive testing with the integrated HD4000 GPU on my 3770k. We're talking a 0~3c at most temp increase with the iGPU enabled. The iGPU really adds minimal heat to the die if you have sufficient cooling. It's a seperate part of the chip that has it's own connection to the internal heatspreader which is contacting the CPU cooler. Our CPU coolers are way more powerful than our chips require - a little more wattage overall isn't suddenly going to spike temps up. Not to mention you can completely disable it if you are that concerned about temps. Intel iGPUs pull VERY MINIMAL power, people acting like enabling the iGPU will significantly increase temps frankly have no experience using a modern intel iGPU.

Personally I'd love if an iGPU was integrated into the HEDT chips, or at least the quicksync functionality. The encoding speed it provides is beastly.

 

[wabbitseason]

I was talking about added complexity, wasting TDP on non-essential crap, and limiting overclock headroom. The temps is a secondary issue.

 

[wtourist]

Out of billions of transistors, those used for iGPU seem minor issue. Every CPU is used for multiple purposes, there is no reduction wanted or desired in the ability to run various instruction sets.

A friends wife bought a new mini van. Got a flat in heavy traffic with a full car far from home. No spare.
Not only not supplied but no one told her there was no spare.
Not only no spare, no space saver spare, no spare mounting kit, or tools. No informative note.

Some complexity is a good thing. Especially when unexpectedly needed.

 

[aphexcoil]

Out of billions of transistors, those used for iGPU seem minor issue. Every CPU is used for multiple purposes, there is no reduction wanted or desired in the ability to run various instruction sets.

A friends wife bought a new mini van. Got a flat in heavy traffic with a full car far from home. No spare.
Not only not supplied but no one told her there was no spare.
Not only no spare, no space saver spare, no spare mounting kit, or tools. No informative note.

Some complexity is a good thing. Especially when unexpectedly needed.

I disagree. It really depends on the end-user. If it's a simple mom and pop computer, than a fully integrated system works well for them.

However, I believe most of the people who come to these boards (myself included) want the ability to pick the best parts for our system. I don't want my CPU wasting any space for GPU because I have an AMD R9 290 to handle that. I want the best of the best.

 

[chrcoluk]

I got no issue with the integrated GPU, and I am willing to bet that the GPU causes less extra heat than an extra 2/4 cores does. I prefer 4 core with iGPU than 8 core. Since desktop software that needs that many cores is rare, I do occasional encoding but very infrequently, maybe once every 3-6 months.

 

[wabbitseason]

Out of billions of transistors, those used for iGPU seem minor issue. Every CPU is used for multiple purposes, there is no reduction wanted or desired in the ability to run various instruction sets.

A friends wife bought a new mini van. Got a flat in heavy traffic with a full car far from home. No spare.
Not only not supplied but no one told her there was no spare.
Not only no spare, no space saver spare, no spare mounting kit, or tools. No informative note.

Some complexity is a good thing. Especially when unexpectedly needed.

Your analogy is nonsense. Enthusiast users will be using a dedicated GPU and a maxed-out mobo. They don't need more and more crucial system hardware consolidated onto the CPU, wasting TDP, adding complexity, and limiting the maximum stable overclock.

Consolidation onto the CPU makes sense for the enthusiast only when it substantially increases performance in desired tasks, without limiting the overclock significantly.

 

[sblantipodi]

Your analogy is nonsense. Enthusiast users will be using a dedicated GPU and a maxed-out mobo. They don't need more and more crucial system hardware consolidated onto the CPU, wasting TDP, adding complexity, and limiting the maximum stable overclock.

Consolidation onto the CPU makes sense for the enthusiast only when it substantially increases performance in desired tasks, without limiting the overclock significantly.

Enthusiast platform should support newer standards like usb3.1
Is ridiculous to see z97 mobos with USB 3.1 and X99 mobos without USB 3.1

 

[wabbitseason]

Enthusiast platform should support newer standards like usb3.1
Is ridiculous to see z97 mobos with USB 3.1 and X99 mobos without USB 3.1

How does this have anything to do with my post? I agree, btw, that the newest mobos should have the newest features and standards.