Confirmed - i9 9900k will have soldered IHS, no more toothpaste TIM

[DigDog]

And 4C is not what I'd consider slight if you're pushing things to the very edge! Anyone who is delidding chips isn't your typical OC for 24/7 100% load person that's looking to maximize throughput and have longevity and stability suitable for heavy, continuous workloads.

Eh, not really. Delidding chips became a thing in these last few years *because* of the intel tim situation. Earlier you might have had LN overclockers delidding, the same people who push 1.8v and that value 4c. temp, but since, there's been plenty people that casually delid because A) they have a bad chip and delidding gains them 20c., and B) they saw it on youtube and they think it's 1337 - these are the people that made a market for delid tools. Neither of these two groups is "pushing the envelope".

 

[Fir]

I delidded my 7980XE because it had extreme deltas and the result was excellent! Full AVX512 load temps in the mid 60s with 1.275V 48x100 on all cores. Temps before that were 30C higher on some cores. Using the EK monoblock with a chilled water system (20C fluid temp at block, regulated) Interestingly enough at stock the delta was nowhere near as severe. I had a good chip that was "benchmark stable" at 5GHz so figured delidding would be worth the trouble.

For 4C, I certainly would not have bothered I can tell you that!

I do miss bare die chips though. Never cracked a P3 or Athlon and wasn't particularly careful either. Guess I was lucky or people that do have Neanderthal hands.

 

[aigomorla]

In the theory, having a soldered IHS should mean we're no longer limited by thermals when overclocking (throttling) due to die / IHS heat transfer.

no your always thermal limited.... why else do you think people go extreme and go on sub ambient / phase cooling, or drop ln2?


But honestly, every 10C you drop your overall temps, you not only effectively double the life of said silicon, but also make it run a bit more efficient. This is clearly shown more on Nvidia GPU's more then intel cpu's tho.

If you want the best results, don't care about cost why even deal with a heatspreader and just engineer your own direct die contact block and be done with it?

You can just buy a direct die mount like the ones i showed in those pictures... although MSI stopped making there's back in the 4700 series. But yes i agree with you if i am going to delid the chip, i would not liquid metal it and put the IHS back... run the baby straight up naked under a waterblock, because a heavy air sink will most likely crack that puppy.

Delidding chips became a thing in these last few years *because* of the intel tim situation.

no... deliding was also a thing back in the athlon 64 /opteron 165 days with the manchester die.
You remember back when server chips actually overclocked better then consumer ones.

The chips ran fast as heck when you delidded them, and getting that 3.2ghz out of the chip 3800+ chip was easy as pie....

My first Delid was actually on my opteron back when people thought Intel had no chance again AMD.

 

[DigDog]

Again, the gist of the argument is that it was a rare occurrence that became much more common now due to that tim issue. Nobody even mentioned delidding 10 years ago, dont give me the "i did it so we all did it" argument.

 

[RichUK]

no your always thermal limited.... why else do you think people go extreme and go on sub ambient / phase cooling, or drop ln2?


But honestly, every 10C you drop your overall temps, you not only effectively double the life of said silicon, but also make it run a bit more efficient. This is clearly shown more on Nvidia GPU's more then intel cpu's tho.



You can just buy a direct die mount like the ones i showed in those pictures... although MSI stopped making there's back in the 4700 series. But yes i agree with you if i am going to delid the chip, i would not liquid metal it and put the IHS back... run the baby straight up naked under a waterblock, because a heavy air sink will most likely crack that puppy.



no... deliding was also a thing back in the athlon 64 /opteron 165 days with the manchester die.
You remember back when server chips actually overclocked better then consumer ones.

The chips ran fast as heck when you delidded them, and getting that 3.2ghz out of the chip 3800+ chip was easy as pie....

My first Delid was actually on my opteron back when people thought Intel had no chance again AMD.

Extreme overclocking aside.

 

[TheELF]

Also regarding RMA rates,surely making your chips run hotter through having a poor thermal interface between the bare chip and the cooler,will also increase potential failure rates?

No because modern systems do thermal throttling,they lower the clocks to keep the CPU save.
Also CPUs shut down before they reach really dangerous temps,that's a command from within the CPU and nobody can bypass it,it's something like 5° below the max temp they could handle.
Even on the toothpaste K CPUs even when overclocked (not extreme) temps where always safe because the CPU could handle up to 100°C without shutting down (page 102 here) it's just that enthusiasts can't see such a number without getting pimples although it is safe.

 

[aigomorla]

Again, the gist of the argument is that it was a rare occurrence that became much more common now due to that tim issue. Nobody even mentioned delidding 10 years ago, dont give me the "i did it so we all did it" argument.

if your making this comment i can pretty much bet you came into overclocking late....
Everyone and anyone whose had the AMD X2 3800+ / Opteron 165 and overclocked it on a DFI Lanparty board has at least once considered to delid the cpu.

Back then overclockers were limited in number, however a much larger percentage of them were extremist willing to easily goto the next extreme... so it wasnt a rare occurrence, unless you take overclocking in general a rare occurrence...

 

[Markfw]

if your making this comment i can pretty much bet you came into overclocking late....
Everyone and anyone whose had the AMD X2 3800+ / Opteron 165 and overclocked it on a DFI Lanparty board has at least once considered to delid the cpu.

Back then overclockers were limited in number, however a much larger percentage of them were extremist willing to easily goto the next extreme... so it wasnt a rare occurrence, unless you take overclocking in general a rare occurrence...

I beg to differ. I had (actually still have a 3800 X2) those chips, and overclocked them, and never considered delidding, and I was very active on this board at that time, 2002-2006. It was only 2 years later I was elected the CPU moderator due to my extensive experience in overclocking.

 

[IEC]

Well if you want to go back far enough to the Athlon/T-Bird days, there was no heatspreader at all and it was naked die for everyone.

Hell, my Celeron 300A @ 450MHz went into a slot, not a socket!

Delidding only became somewhat common in the enthusiast community thanks to inexpensive tools and the gains you could get on Skylake/Kaby Lake/Coffee Lake CPUs. Now that the top-end desktop CPUs will be soldered again, I expect that market to dwindle. There might be a few die-hards who will go naked die or delid for minuscule gains, but there won't be many.

 

[dsplover]

Well delidding lowers temps on certain CPUs and it’s why I bought a high binned 8086k.
Needed to run @ 4GHz but mid 50s since I work outdoors in the summer.
Using 1U mobile PCs I’m not digging the water cooling options, but triple barrel led 22k fans and a delidded CPU work great.

I’m no genius CPU Guru, but hopefully I’m not posting wrong.
But any advice appreciated.
My question about 8 Core 9000 is can I disable 4 Cores and make the shared L3 Cache work with 4 Cores?

My apps benefit from larger cache. Even an i7 5775C using slow L4 Cache whooped my Devils Canyon.

Thanks, great thread.
I love delidding.
I’m going to love not having to as well.

 

[aigomorla]

Well if you want to go back far enough to the Athlon/T-Bird days, there was no heatspreader at all and it was naked die for everyone.

rofl... and they could melt though the board if you didnt put a heat sink on...
i think i cracked one of those guys by over tightening a zalman heatsink on it....

 

[Fir]

rofl... and they could melt though the board if you didnt put a heat sink on...
i think i cracked one of those guys by over tightening a zalman heatsink on it....

Obligatory

 

[darkswordsman17]

Again, the gist of the argument is that it was a rare occurrence that became much more common now due to that tim issue. Nobody even mentioned delidding 10 years ago, dont give me the "i did it so we all did it" argument.

I was hardly an LN2 max OCer and I delidded over 10 years ago (back in like 2005, had an Athlox 64 X2). It was because it was incredibly easy to do, and I was far from the only one doing it (I did it because plenty of other people had and reported on how easy it was to do; same thing with people using conductive pens to connect traces on 7900GT video cards I seem to recall). And there were quite a lot of people doing "pin-mods" and stuff like that (which they were doing some things like that in the 90s even).

The difference between now and then is that it got commercialized to an extent. There was the delidding tool made and then there was the site (siliconlotto?) that bought and sold chips that were delidded or had the TIM redone, and were rated for certain clocks.

I think the other thing that changed is that a lot of people adopted huge HSFs and/or went watercooling, so many of them did that instead of delidding. But then after Intel arguably started losing top clocks because of the TIM, people started resorting to changing the TIM.

Might be interesting to see based on rates/ratio/percentage. I kinda think it might actually be similar throughout, but that the numbers just being higher now, makes it seem like the ratio is higher when it isn't. Mods like that though were hardly super rare years and years ago. I seem to recall some interesting "hacks" for chips in the 80s (not that I was doing them, but through vintage YouTube channels found out about them). And then seeing people talk about modchips and things that were "common" in the PS1 era that I didn't hear about much if at all (so the original Xbox seemed like it was really the first console where modchips became a real thing, when that wasn't the case at all).

 

[darkswordsman17]

Do mobile chips still go without the heatspreaders? I always wondered why they didn't figure out a way to protect the die but ditch the heatspreader (I noticed on the chips with HBM that they put a material between the HBM modules and die that gives it a unified flat face; and I remember the old Athlons and stuff having those spacers on the corner for heatsinks to try and keep them from pushing on the die too hard).

I've actually wondered why they haven't tried to go to a dual faced chip where the interface between the die and the package would be on the sides/edges of the die instead of the bottom, leaving a face on both sides as avenues for applying a heat transfer system (which to me would help). Especially on video cards (CPUs would need to change the motherboard setup).

 

[coercitiv]

Do mobile chips still go without the heatspreaders?

No heatspreader on mobile chips, the die is naked and whatever cooler gets mounted on top applies very limited pressure.

 

[epsilon84]

Obligatory

Wow I remember watching this video about... 18 years ago?! Cool techno music too hahaha

I was especially careful installing HSFs (Thermaltake Volcano 5+, why do I even remember this?!) on my Athlon XP machines after watching this. Thankfully none ever went up in smoke like the video.

 

[TheGiant]

such na Intel Bias.....they had to find something because Athlon was just a miles better, the p4 was a hand calculator comparing to it, except for the benchmarking paid by Intel...

but now its different story..

Imo there are 2 things that will help 9900K- first is solder, secont is heat flow density (heat flow from 8C doesn't increase linear with surface and thus the most intensive heat flow area is greater- as we see with series 7700K (4C), 8700K(6C), 9900K(8C)

 

[TheELF]

such na Intel Bias.....they had to find something because Athlon was just a miles better, the p4 was a hand calculator comparing to it, except for the benchmarking paid by Intel...

but now its different story..

How is it a different story now?
AMD still needs about *33% more CPU to be competitive in overall CPU power.
Except for the benchmarking paid by AMD,or in better,corecterer wording,except for the workloads AMD CPUs are designed for.

*i3-7100 vs r3 1200 =2c/4t vs 4c
i3-8350k vs 1500x =4c vs 4c/8t
i5-8400 vs 1600x = 6c vs 6c/12t
i7-8700 vs 2700 =6c/12t vs 8c/16t
(And AMD's smt is supposed to give a 50% boost)
https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700/19.html

 

[maddie]

How is it a different story now?
AMD still needs about *33% more CPU to be competitive in overall CPU power.
Except for the benchmarking paid by AMD,or in better,corecterer wording,except for the workloads AMD CPUs are designed for.

*i3-7100 vs r3 1200 =2c/4t vs 4c
i3-8350k vs 1500x =4c vs 4c/8t
i5-8400 vs 1600x = 6c vs 6c/12t
i7-8700 vs 2700 =6c/12t vs 8c/16t
(And AMD's smt is supposed to give a 50% boost)
https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700/19.html

Quote from above:
"(And AMD's smt is supposed to give a 50% boost)"

Love this line. Try not to sound so desperate.

 

[Zucker2k]

Well delidding lowers temps on certain CPUs and it’s why I bought a high binned 8086k.
Needed to run @ 4GHz but mid 50s since I work outdoors in the summer.
Using 1U mobile PCs I’m not digging the water cooling options, but triple barrel led 22k fans and a delidded CPU work great.

I’m no genius CPU Guru, but hopefully I’m not posting wrong.
But any advice appreciated.
My question about 8 Core 9000 is can I disable 4 Cores and make the shared L3 Cache work with 4 Cores?

My apps benefit from larger cache. Even an i7 5775C using slow L4 Cache whooped my Devils Canyon.

Thanks, great thread.
I love delidding.
I’m going to love not having to as well.

If you're going to disable four cores and leave HT on, it might be better to simply disable HT and leave the logical cores alone.

Edit: With regard to delidding, I'm no extreme overclocker but I delidded a few Core 2 Duo/Quads back in the 2007/2008 timeframe. It was dangerous and needed a keen eye, judgment and timing to avoid frying the die underneath the IHS. There were different methods, of course, but I learnt the hard way. Killed a couple of chips; one had the die ripped off from the substrate prematurely, the other cracked under my makeshift socket adapter.
Delidding was the thing a little over ten years ago. It baffles me to see enthusiasts complaining about non-soldered cpus. That should be a passport to wringing the last bit from your processor, regardless of cooling type. Yes, it has it's risk but if you want that last efficiency, that's the way to go, as demonstrated in laptops where bare die cooling is the norm because of design limitations due to size and form-factor.

 

[BigDaveX]

I don't think delidding was something that enthusiasts did in any great numbers back in the early-mid 2000s, but there were enough who did do it to lead to complaints when Intel first started soldering the IHS with Prescott.

 

[TheGiant]

How is it a different story now?
AMD still needs about *33% more CPU to be competitive in overall CPU power.
Except for the benchmarking paid by AMD,or in better,corecterer wording,except for the workloads AMD CPUs are designed for.

well its totally different

the p4 general performance, especially typical x87 code performance was catastrophic, way worse than Athlon or P3 (and the f...up with RD RAM cost.....)

now the only problem with ryzen are latency based programs, everything else runs comparable to CFL/KBL/WTFLake

AMD will start to lose much when I9-9900K reaches 5GHz with left hand in ass....today 2700X performance is very acceptable and at least tradeoff (6C vs 8C)

as for the solder, anyone has die pictures or guess of heat distribution diagram?

I like intel's 14_more_+

 

[TheELF]

Quote from above:
"(And AMD's smt is supposed to give a 50% boost)"

Love this line. Try not to sound so desperate.

Tell that to everybody that claims that amd's smt is superior to intel's HTT.

 

[Markfw]

Tell that to everybody that claims that amd's smt is superior to intel's HTT.

Well, it is superior, and has been proven many times. You are trying to say its not ?

Edit, its not 50% boost, maybe 40%, I don't remember the finding

 

[Zucker2k]

Well, it is superior, and has been proven many times. You are trying to say its not ?

Edit, its not 50% boost, maybe 40%, I don't remember the finding

It is not superior, To say it is superior is to imply Intel's implementation is not efficient; which is to say, don't talk about SMT in a vacuum.