Haswell E De-Lidded

[CrazyElf]

Someone delidded an ES chip of Haswell E.

See here:
http://www.ocdrift.com/intel-core-i...uses-soldered-thermal-interface-material-tim/


That looks like a solder to me, probably epoxy on the sides?:

 

[Techhog]

Yeah, this definitely needed it's own thread, sine so many people are confused about the 2011 CPUs. FYI guys, IB-E was soldered. There was never a risk of cheap TIM being used in the first place. Remember that these aren't consumer CPU designs; they were designed for high-performance servers. Intal can't get away with hot chips in a server blade.

 

[Maximilian]

Has that beast got 12 cores?

 

[NostaSeronx]

Has that beast got 12 cores?

The chip is the Haswell-E 8-core which uses the Haswell-EP 12-core die. It will be interesting if some motherboard vendors do a core unlocker scheme.

Buy 8 core SKU for $500-1000 less than 12-core SKU then unlock the disabled cores. Get a nice DIY water cooling rig then overclock to 5 GHz. Set for life till the Icelake-E 24 core SKU.

 

[CrazyElf]

Has that beast got 12 cores?

The top Ivy Bridge Xeons had 15 cores.

http://ark.intel.com/products/79209

 

[DrMrLordX]

Nice delid pic.

Intel will likely do anything they can to prevent core unlocks. It would pretty sweet if someone could pull that off, though.

 

[Fjodor2001]

Can't they fuse off the extra 4 cores on Haswell-E to prevent that?

Or use Hawell-EP dies where some of the cores failed test in factory? Isn't that the whole idea in the first pace by the way, to make use of failed Haswell-EP dies and sell them for less...

 

[crashtech]

I didn't think there was any doubt that Has-E would be soldered.

 

[WhoBeDaPlaya]

Solder couldn't be done my a*s!

 

[Lepton87]

The chip is the Haswell-E 8-core which uses the Haswell-EP 12-core die. It will be interesting if some motherboard vendors do a core unlocker scheme.

Buy 8 core SKU for $500-1000 less than 12-core SKU then unlock the disabled cores. Get a nice DIY water cooling rig then overclock to 5 GHz. Set for life till the Icelake-E 24 core SKU.

That core unlocking thing worked great for SandyBridge-E didn't it? Desktop IVY-E was a native 6 core chip while Sandy-E had 2 dice one with 1.27B transistors and 4 cores and 8 cores one with 2.27B transistors but only 6 cores version was available for desktops. Those 2 disabled cores are either laser fused or disabled in microcode but it doesn't really matter since no one was able to unlock either the disabled cores or the cache in the non-extreme versions. No to mention that 5GHz even on 4-core HW is wishful thinking, what makes you think that a CPU with double or triple(as you naively believe) the core count would overclock better? If it overclocks in the same ballpark as 4790 that would be great and better then expected.

Can't they fuse off the extra 4 cores on Haswell-E to prevent that?

Or use Hawell-EP dies where some of the cores failed test in factory? Isn't that the whole idea in the first pace by the way, to make use of failed Haswell-EP dies and sell them for less...

Those chips can be fully functional but have the cores disabled simply for market segmentation, that becomes more common as process matures and less and less cores and faulty in some way. But for us, consumers it doesn't matter if the silicon was faulty or disabled for market segmentation to meet demand for lower-end part, we can't unlock anything either way. I can't remember any Intel CPU where you could unlock anything barring upgrades that you had to pay for unlike AMD where core unlocking were common but not guaranteed. I had two single core Semprons and while both unlocked to dual core CPUs they became unstable with time so even if its technically possible to unlock you still face the risk of something going wrong. On the other hand I had X800PRO VIVO and it unlocked to 16 pipelines with 100% success rate(X800PRO had 12), I bought X800PRO VIVO not X800XT simply because the other one was unavailable and those card were identical except for BIOS.

Thanks for that. I did a bit of digging myself and couldn't find anything on IB for some reason- however I found a thread here that has a comparison of a ton of cores. If I am not mistaken, it looks like Haswell is almost 5 times the size of Jaguar to get that 3x improvement in ipc.

Keep in mind that those figures are from different nodes. Intel's 22nm is clearly denser than GF's/TSMC's 28nm. IDC posted a graph about CPU architecture that says that if you want to increase IPC by 25% you need to disproportionately increase the die size, I can't find the graph, maybe someone wants to help finding it? I'm going to look for it anyway since I think it's interesting. I don't remember the exact numbers but it was something like IPC +25% die size +50% or more.

 

[Techhog]

I didn't think there was any doubt that Has-E would be soldered.

There was doubt, though the doubters were just less knowledgeable people, many of whom thought IB-E used TIM.

 

[2is]

That's a huge ass die

 

[Lepton87]

There was doubt, though the doubters were just less knowledgeable people, many of whom thought IB-E used TIM.

AFAIR Intel said that all CPUs with a TDP of 125W and up will use solder, the same with AMD. AM3 parts get solder but APUs do not. Solder is of course better then TIM, but the whole outrage wasn't so much about TIM instead of solder but about that gap between IHS and the die, that's why simply re-lidding can lower the temperatures by a significant amount. Properly done TIM is fine, a couple of degrees worse then solder but livable, it's that gap that increases temperature upwards of 20C in some cases. Was there a mass hysteria about AMD's use of TIM instead of solder? I can't remember. I guess that gap was put there to further protect the CPU die from damage dealt by ussing excessive mounting force. At stock it still worked, 60C or 80C Intel didn't care, the CPU would break down long after the warranty expired. A chipped die or a whole die on the other hand did matter. That extra gap acted like a cushion. Intel at the time considered enthusiast less important then those few saved RMA procedures. I don't know if DC still has the gap or not. The severity of the gap varied greatly between HW CPUs. The new TIM and smoother power delivery seems cool and dandy but if they didn't remedy that gap lottery IMHO it's still MEH.

 

[Techhog]

AFAIR Intel said that all CPUs with a TDP of 125W and up will use solder, the same with AMD. AM3 parts get solder but APUs do not. Solder is of course better then TIM, but the whole outrage wasn't so much about TIM instead of solder but about that gap between IHS and the die, that's why simply re-lidding can lower the temperatures by a significant amount. Properly done TIM is fine, a couple of degrees worse then solder but livable, it's that gap that increases temperature upwards of 20C in some cases. Was there a mass hysteria about AMD's use of TIM instead of solder? I can't remember. I guess that gap was put there to further protect the CPU die from damage dealt by ussing excessive mounting force. At stock it still worked, 60C or 80C Intel didn't care, the CPU would break down long after the warranty expired. A chipped die or a whole die on the other hand did matter. That extra gap acted like a cushion. Intel at the time considered enthusiast less important then those few saved RMA procedures. I don't know if DC still has the gap or not. The severity of the gap varied greatly between HW CPUs. The new TIM and smoother power delivery seems cool and dandy but if they didn't remedy that gap lottery IMHO it's still MEH.

Yeah, I know. It's harder to prevent the gap when using TIM though.

 

[ViRGE]

Holy moly! It's one thing to delid a CPU, but to rip off the top layer... That's some good solder!

 

[CrazyElf]

AFAIR Intel said that all CPUs with a TDP of 125W and up will use solder, the same with AMD. AM3 parts get solder but APUs do not. Solder is of course better then TIM, but the whole outrage wasn't so much about TIM instead of solder but about that gap between IHS and the die, that's why simply re-lidding can lower the temperatures by a significant amount. Properly done TIM is fine, a couple of degrees worse then solder but livable, it's that gap that increases temperature upwards of 20C in some cases. Was there a mass hysteria about AMD's use of TIM instead of solder? I can't remember. I guess that gap was put there to further protect the CPU die from damage dealt by ussing excessive mounting force. At stock it still worked, 60C or 80C Intel didn't care, the CPU would break down long after the warranty expired. A chipped die or a whole die on the other hand did matter. That extra gap acted like a cushion. Intel at the time considered enthusiast less important then those few saved RMA procedures. I don't know if DC still has the gap or not. The severity of the gap varied greatly between HW CPUs. The new TIM and smoother power delivery seems cool and dandy but if they didn't remedy that gap lottery IMHO it's still MEH.

Devil's Canyon I think has (mostly) addressed the issue, although there is still some grumbling around about having to delid. The gap is not as big I think with the 4790Ks, they use somewhat less power, and they run mostly cooler.

The issue I see is that even with the gap addressed or if you delidded, you still did not get much more than say, 4.7-4.8 GHz on air no matter what on air. And getting the last 100-200 MHz saw power consumption skyrocket.

What it suggests is that even with the solder, I think many of the E versions will struggle to make 4.5 GHz.

Holy moly! It's one thing to delid a CPU, but to rip off the top layer... That's some good solder!

With the power consumption of Haswell-E expected with 8 cores, I get the feeling that we are going to need it.

That epoxy seems to be keeping the core attached pretty solidly to the top of the CPU.

 

[Grooveriding]

I cannot wait to get my hands on one of these beasts. Release them already!

 

[know of fence]

Solder couldn't be done my a*s!

For insurance reasons, you understand. It may also not been up to fire codes and was possibly blocking fire truck access.

Actually, I figure, the real reason is the price hike of solder, Indium now costs about one dollar a gram. For a high margin company like Itnel it would be just very annoying to pay a price that due to speculation and supply shortages was artificially 10x inflated.

 

[NTMBK]

Oh my goodness, that is just painful to see. I WOULD HAVE TAKEN GOOD CARE OF YOU LITTLE CPU.

 

[PliotronX]

Oh my goodness, that is just painful to see. I WOULD HAVE TAKEN GOOD CARE OF YOU LITTLE CPU.

I know

 

[UaVaj]

+1 for it being solder



in regard to a gamer's cpu.

i7-5960x - 8 core 16 thread - and only 3.0GHz to 3.3GHz Turbo
unless new software/games can honestly take advantage of more core and slower speed. this is definitely not a gamer's chip.

vs

i7-4960x - 6 core 12 thread - 3.6GHz to 4.0GHz Turbo
this is a better balance for a gaming chip.

 

[Idontcare]

Holy moly! It's one thing to delid a CPU, but to rip off the top layer... That's some good solder!

No kidding! My first thought when I saw the delayered CPU was one of sympathy...dat solder be good!

Oh my goodness, that is just painful to see. I WOULD HAVE TAKEN GOOD CARE OF YOU LITTLE CPU.

Truer words are rarely spoken here. Well said. And RIP you blessed little Haswell E ES. You passed too soon. :rose:/

 

[CrazyElf]

Oh my goodness, that is just painful to see. I WOULD HAVE TAKEN GOOD CARE OF YOU LITTLE CPU.

Considering what they do to engineering samples, I would not be surprised if that ES had been damaged already by the massive stress tests they usually put them through.

+1 for it not being solder



in regard to a gamer's cpu.

i7-5960x - 8 core 16 thread - and only 3.0GHz to 3.3GHz Turbo
unless new software/games can honestly take advantage of more core and slower speed. this is definitely not a gamer's chip.

vs

i7-4960x - 6 core 12 thread - 3.6GHz to 4.0GHz Turbo
this is a better balance for a gaming chip.

What matters at this point is, can the 8 core version overclock to 4.5 GHz, assuming enough cooling is available? If it can do at least 4 GHz, maybe. If not then it's not useful as a gaming chip and the 6 core is a better buy.

 

[Mand]

That's not a delid...that's a vivisection

 

[Lepton87]

What's very concerning to me is the very low turbo speed of 3.3GHz, even with IPC improvements HW brings to the table over IB it's going to be slower then 4960X in ST workloads and low-threaded workloads like games. Compared to 6-cores HW it's going to look even worse. On a positive note I think that the low stock clocks are attributable to the abysmal stock cooler which I guess haven't changed at all compared to previous CPUs with that TDP. With a proper cooling it should pass 4GHz with ease but anything over 4.5GHz might require a custom loop. I don't know why enthusiast whose most stressful workloads are games flock to this CPU. 6c/12t is going to be plenty for games for the foreseeable future. I think the only reason is e-peen. For the wast majority of users Core i7-5930K clocked at 3.5GHz base/ 4GHz boost is going to be faster, especially in games.
That's 16% higher base frequency and that's a lot. 16% higher frequency should be better then 33% more cores in a majority of contemporary workloads. As for overclocking crowd, for those with custom loop coolers that would allow the 8 core monster to stretch its legs it should makes sense as that frequency gap should erode. IMHO both CPU should hit a wall well before 5GHz, the 6-core version maybe getting 100-200MHz more.