Delidding an unreleased 56 Core Intel CPU - Sapphire Rapids Xeon

[@OOIEatte]

Maximilien der8auerspierre immediately guillotining any new CPUs he gets his hands on

[@lordjaashin]

are you not entertained??!

[@landonluebke7627]

The fact that I’m learning about this now in AP world makes it even better

[@ReaperX7]

Off with its IHS!!!

[@QuietStorm4964]

The king of overclocking has executed another IHS for the crime of being in the way of cooling

[@TechTechPotato]

Looks tasty

[@Romnipotent]

Knew I'd find you in here somewhere

[@peterzeboroff7448]

Its hard to wrap my head around that many individual extremely tiny balls for connection and have it all connect perfectly with no lost connections. That just shows the extreme level of technology involved, absolutely incredible.

[@AlpineTheHusky]

Well some may not be connected as alot of those solder balls is for power

[@mianderson86]

I want to learn how to "clean" the die and get photos like that. I have a ton of old crappy intel cpus I could practice on, it just looks so awesome!

[@volvo09]

If you find an old ceramic Intel CPU (edit, like a pentium or 486) you can pop the cap off the center of the bottom where all the pins are and see the bottom of the die. You have to take the die off the substructure pcb to see the "cool stuff" so you aren't left with much of anything to hold or look at afterwards.

[@MrMcGreed]

I think der8auer has a video on how to - though it's something with a lot of acids, so be carefull... also, it might be on his German channel, because it's from before he split them...

[@forbiddenera]

@@volvo09 the old pentium pros have a sh*t load of gold

[@Altirix_]

@@MrMcGreed the guy that does the ones on flikr most people have seen i think has an old tutorial on how to do it too, but the process of what to use varies a little by the fab process iirc

[@AtaGunZ]

@@MrMcGreed thanks for the heads up, do you know the video name?

[@neurokinetik64ES]

With all those fine connections underneath, it is amazing that they can attach these chiplets successfully without any of the connections being bridged.

[@TheBackyardChemist]

some chips are indeed lost to defects in those connections, yield of packaging is not 100%

[@suntzu1409]

What "connections being bridged"?

[@osirisgolad]

@@suntzu1409 If one of the chips is less than the width of a hair out of alignment, the solder bump will make contact with two of the connections at the same time instead of just being connected to the one it should be connected to.

[@andytroo]

@@osirisgolad miss-alignment of even less than 1 hair width would create high resistance connections, potentially limiting frequency

[@darknez09240]

agreed it just looks like a technician's nightmare if this cpu chip is on a laptop and they have to solder it back lol

[@Hydrazine1000]

@10:17 Next time, if you want to measure pitch more accurately, measure the distance between 10 or even 20 bumps and divide by the correct amount. The measurement error for inaccurate positioning on the exact center of the bump will be greatly reduced.

[@Ang3lUki]

I'd imagine yields will be improved by the chiplets being smaller than their monolithic designs, however I imagine there are some losses to bad EMIB connections. I remember them talking about interconnect health when discussing Ponte Vecchio.

[@MrMcGreed]

I love how Roman will just be too interested in what's inside to actually restrain from killing a top-of-the-line Xeon

[@TheDoomerBlox]

Sapphire Rapids engineering samples are locked down like never before, so this CPU would not have served much of a purpose besides being a pretty paperweight. Now it's a grilled paperweight!

[@MrMcGreed]

@@TheDoomerBlox Well... as far as I understood, it was functional, right? He says when he wrecks it in the de-lid vice, that it's a good think he didn't need it... so sure it might be limited, but it's still a 56c Xeon...

[@TheDoomerBlox]

@@MrMcGreed It is functional, in so far that you if you were to trick it to think it is in the past - while running the special "key" BIOS that allows it to function - it would run. Besides that specific set of circumstances, it's a paperweight.

[@JETWTF]

It was just an EBay purchase, it's worth is just what was paid for and top of the line does not exist for it considering there are no motherboards yet. A Very early engineering sample that may be worth as much a an I7 7700k in performance on release of motherboards at best is what he bought. It could have been a golden sample but less likely than your pet frog from childhood becoming a sentient being who is hell bent on seducing your significant other as a revenge for letting his little frog life end.

[@fy7589]

Dude you probably just killed a $10k cpu just out of curiosity and education. Assuming it wasn't an already dead engineering sample. Not many people could have done that. You definitely deserve a medal of honor...

[@der8auer-en]

Haha thanks

[@leyasep5919]

Medal of horror ? :-D

[@benNdaKen]

more a medal of stupidity

[@leyasep5919]

@@benNdaKen Tell us what he should have done instead with a chip he can't use/exploit before months ? He got us prime news, he showed us the immediate future and made a lot of views... What's your options ?

[@vukzlatkovic1183]

@@leyasep5919 i mean theorethically he could have spent a month or two to try to come up with a less destructive deliding method but intel could just as easily make engineering samples not work on any non engineering board via bios so it doesn't really matter

[@-MaXuS-]

This is why I genuinely love what Der8auer does when he takes his time and freely shares these videos with us! No one and I mean no one does this and does it out of the same curiosity as at I myself have for things like this! Thank you Roman!

[@der8auer-en]

thank you very much :)

[@nneeerrrd]

You love him because he buys stolen Intel confidential material? This tells a lot about you.

[@kuyans3889]

@@nneeerrrd This will literally ruin intel's business

[@-afyx-]

@@kuyans3889 didn't you think for a moment that AMD hadn't got such a CPU and analyzed it?

[@kuyans3889]

@@-afyx- how does roman buying one make a difference to that? also if AMD bought one and copied it, they would be sued to oblivion. these aren't little 486 clones any more. AMD wouldn't be stupid enough to, at every step they make their own technologies because they don't use intel obviously (for example DOCP instead of XMP). If AMD's business relied on copying intel designs, they would probably be getting samples from sources other than random Ali Express listing or whatever. I genuinely do not understand how it would be of any value for AMD to acquire pre release intel CPU's and analyse them.

[@JohnWilliams-gy5yc]

Intel IHS team: Let's make it impossible to be delidded safely. Roman: You just give me a video plot and a new product. Thanks.

[@SyahrulVEVO]

wait, thats me

[@mycosys]

I gotta admit i'm really surprised you dont have a reflow gun and even a PCB heater, they arent particularly expensive and seem like they would make your life easier often enough. You can pick up a really nice Quick reflow station for a couple hundred US.

[@mycosys]

(example use - put your entire delidding tool on the PCB warmer, heat to 90C and use the reflow gun to bring the IHS to 155 without cooking the entire PCboard/die that far for ages.) FWIW a double wedge delidder might still work with those very close caps

[@herpderp_6488]

this guy deserves so many more subs like he is really interesting and he does the things that most tec youtubers don't dare to do.

[@paulrichalland]

Well he has 450k subscribers on the main channel which is now only german

[@ziggo0]

@@paulrichalland yep. Op is so dedicated to message of needing more subs they literally don't even know who the KZbinr is lmao

[@herpderp_6488]

@@ziggo0 will admit I didn't know he had a main channel. Have been watching him for a while tho and found it weird how he has so little videos and subs. Makes sense now that you say it

[@suntzu1409]

@@ziggo0 Derbaeur still deserves more subs

[@TobyIKanoby]

This guy is already an OG at this point. He achieved quite a lot most tec youtubers can only dream of.

[@Techy93]

this needs an R rating for tech gore or something 😂😂

[@starwolf4109]

I thought this CPU was useless after the failed delidding attempt, but you did an awesome job of making the most of the situation!

[@NeroKoso]

The chip looks amazing! Really cool to see.

[@lhxperimental]

Intermediate core numbers like 56, and 112 PCIe signal that Intel is struggling with pushing performance.

[@TheTechDragonChannel]

you could try to do some image stacking to see if you can pull more details out if you overlap all of the states and colors, some details may stand out a bit more.

[@justinbouchard]

Can you design a delidder that would push the lid on a pivot point basically to avoid damaging the caps? The top left corner with the writing being the pivot point and pushing the bottom right corner towards the bottom left corner? It would only take a very minimal amount of movement before you could just remove it from the delidding tool and manually "wiggle" it looser. Just my backyard north of lake superior thoughts ;) Love from Canada :)

[@mrawesomelemons]

Or maybe create some sort of "stopper" for one side and push up the other side using some sort of wedge shape.

[@justinbouchard]

@@mrawesomelemons wedge is also a great idea that might work even better! I mean taking it out when it's 150 C and "wiggling" the lid is somewhat laughable to like argue against my own idea hahaha. The wedge would eliminate that for sure!

[@CaseyHancocki3luefire]

Desoldering the caps before delidding should be pretty easy on capacitors that size. they look like they are at least 0402, 0603, or maybe even 0805 size. 0402 size caps are pretty hard to do by hand but they are definitely possible and i think those caps are probably bigger than 0402

[@BenQuigley]

@@justinbouchard you could create a delidder that pushes teflon wedges into it from four sides, maybe the corners, maybe the edges, that way you would lift the lid vertically

[@McFlyOrPie]

You killed it! This made me cry a little. lol

[@simulationbros]

Im sorry bro but the way you delidded that made me want to cry. The proper way to do it like that is to take 4 thin razor blades, cut the adhesive holding the IHS, then put each razor blade on each side on the cpu between the IHS and the CPU, causing slight pressure. You then take a Heat gun that can melt solder and heat the cpu (make sure the cpu is upside down and elevated on something) and wait till the IHS pops off.

[@ToTheGAMES]

You clearly don't know who Der8auer is. Pathetic comment.

[@hohaii]

Der8auer HBM2 is not to be compared to AMD's 3D cache. It's SRAM vs DRAM... second, it's 16cores inside that Intel chip with some deactivated because of yield problems.

[@floodo1]

d00d thanks for always showing your mistakes, makes the successes that much better!!!

[@sublime2craig]

You dont marinate your CPU's before slapping them on the grill? Man your missing out on some FLAVOR! Great vid and super interesting, also a little sad to see that CPU get crushed vaporized!!!

[@Nobe_Oddy]

WHOA!!! @ 9:48 - if you pause it and pay attention to all of the 20 rectangles (inside one of the 4 quarters of the whole chip) you can see the ones that are connected to the 'inter-connect' (i forget what it's called) to the cores next to it are the ones2 doing the processing. There are only 14 in each of the quarters, and you can see that only the 14 that have TWO CONNECTIONS (or 1 double wide) of that 'inter-connect" OR to another processor.... In otherwords, while looking at the quarter that is in the center of the screen @ 9:48 the 6 cores that are on the very left and very bottom (the top core on the left side and the far right core on the bottom side are part of the 14 processors enabled, so they are not what I talking about here) must be used for something other than processing, so maybe cache or IO or just disabled???? Or is that all done in another area? And if a core is disabled is it then just taking up room and becomes a useless area on the chiplet. (I really don't have much knowledge on what all these pieces are, I'm just guessing really lol) ..... So this is why it is such a STRANGE number of CORES in the chip (56 in this case, so 14 in each quarter) BUT I would like to know if it is the same cores in each CPU or can one of the quarters have ALL 20 cores enabled while another has one 8 and the other 2 both have 14??? Is it possible for Intel to have ALL 80 CORES enabled if they all just happened to work perfectly??? - There is so much I would love to learn about, but I have no clues where to get the answers...

[@Nobe_Oddy]

ohhhh.... hrmmmmm..... so @ 10:57 you show the chiplet etched a bit and we can see that there are 16 CORES in each chiplet.... (so 2 of them mush get disabled, but I dont think they can be any of the ones with the 'inter-connect' - I think it's called an E-MIP??? 🤷‍♂) so there goes my whole theory lol - I REALLY think I should start watching the WHOLE ENTIRE VIDEO BEFORE I MAKE AN ASS OF MY SELF LEAVING CO

[@gudenau]

Might be fun to get a 8 socket board for this, dream workstation I suppose.

[@chubbysumo2230]

At 350 W TDP each, thats 2800 watts of cooling needed. I can only imagine the power consumption not being much lower than that, would be hilarious to see him try and overclock all of them at once. 448c/896t at 4ghz. I would bet these 56 Court chips are probably clocked at 2 or 2.1 GHz max.

[@ShaneBuchman]

This particular chip will need a dumpster. A new one of these would be great to use in a multiple cpu workstation.

[@hariranormal5584]

Pretty sure only dual exists for this

[@DGao-zz5vq]

Ice Lake SP and newer only support 2S I believe.

[@suntzu1409]

@@chubbysumo2230 well, if 2800W is just too much, how tf HPE is making >=30MW supercomputers? Why and how did Microsoft and Amazon make data centers that draw several kilowatts of power?

[@Maisonier]

Ohh Derbauer, your videos are like gold in my chip life, although I will never could buy one of those ... like as usual, greetings from Argentina.

[@thephoenixking1086]

It is amazing just how small these component's are, and what we can see here is not even the smallest part, imagine being able to see all of the Transistors, BILLIONS of them. It amazes me how something so incredibly small (like EXTREMELY small) can be so incredibly powerful, BILLIONS of these transistors can fit into an area smaller than a British 10 Pence Coin...

[@BARCH-wp5vl]

Great video, incredible work

[@KRAVER_]

Damn Roman, You F* that CPU up. PCB was thick. Fun Deliding tho thanks !!

[@DanielGT_93]

Great video as always! Very nice work on showing the cores

[@martinpeel1820]

When removing the IHS you should have made a small spacer to put between the cpu and de-liding tool and then shaped the spacer so that there's a cutaway on the bottom half of the side that presses on the cpu so that you could be certain that the pressure is against the IHS only and the spacer will clear and go over the PCB and not into it. Additionally you could have made an additional cut away in the section of the spacer edge that would have contacted the SMD's so that it wouldn't have damaged them either and only made contact to the IHS either side of the SMD's. Just thought I'd mention the idea in case your in a similar situation in the future and can hopefully avoid damaging another chip. Love the videos always really interesting and massive respect for the guts to do something like this knowing you'll almost definitely ruin this cpu using that de-lidding tool and still going ahead with it.

[@fVNzO]

Is this based on the same 10nm from Alder Lake? Edit: Rest in pieces. Hardwaregore.

[@17Patrickkk]

6:45 Me and many others would give you a sub, if you would say: "The imild technologie is basicly glueing together the dies with hot glue" Geiles Video wie immer ;)

[@ionottoi]

4:36 - That hurts. A lot. No words to describe it.

[@mauriceo730]

i cried a little when it broke. Dmn that hurts. But those die-shots wow

[@BlackSkorpionLP]

3:45 but these are big enough to be desoldered before delidding and resoldered after.

[@rellaxshala4189]

I wanted to say the same thing. You can desolder them, delid and after that resolder when you done your work

[@Dreams_Of_Lavender]

The socket for Skylake and Cascade Lake is 3647, but currently Ice Lake is on socket 4196.

[@D.u.d.e.r]

Beautiful! Fritzchens Fritz took it to next level😉

[@BreakingTaps]

Great work as always, really enjoy your decapping videos! If you ever want or need some SEM shots of the die, hit me up. Happy to help out 🙂I can do automated stitching across the whole die as well if there's interest, although it could take a while depending on magnification.

[@indridcold2840]

F in the chat to pay respects to Roman's wallet.

[@Radovanslav]

type your credit card info in the chat for Roman to pay to Roman's wallet

[@dominic.h.3363]

@@Radovanslav This would be funny, if some people wouldn't be stupid enough to do it...

[@Radovanslav]

@@dominic.h.3363 yeah.................

[@thatautogarage3644]

Incredible work!

[@MrKentaroMotoPI]

Reminds me of my Pentium Pro (500 nm) from 1996. Overclocked from 200 to 233 Mhz. Still running 😎

[@lordjaashin]

what are you able to even run on it?

[@MrKentaroMotoPI]

@@lordjaashinIt was used as a file server until about five years ago. It has Windows 2000, Office 2003, and Photoshop 6. We use it to convert old-format Office files, run old Photoshop plug-ins, and play an occasional Age of Empires scenario. It has a SuperMicro AT mobo made in a place called "USA". The power connector runs hot when under load, and the contact clips eventually oxidize causing crashes. I have surgically replaced clips (take from a dead Seasonic) and put conductive thermal paste on them. That experience has made me very sensitive to power delivery - both number and quality of conductors.

[@heyarno]

would polishing with cerium oxide be more controlable?

[@Luscious3174]

Those are 16 core chiplets indeed, so a 4x16 or 64 core part was definitely the design. Knowing Intel they probably do have 64 core working in their lab already - it's the yield rates, lithography, complexity of the design and packaging that's holding it all back. These guys are at the bleeding edge of physics and materials. A 350W TDP is an incredible amount of heat for a CPU to push let alone 2x or 4x - those 1U designs will be interesting to see how they stay cool if not quiet. I do wonder what form factor those 8 CPU systems are because all I can seem to find are 4 CPU XCC servers from Supermicro/Tyan. Maybe they are referring to the systems that cram four nodes in a 2U chassis, that would give you 8 CPU's in one box, but those top out at 185W TDP parts due to thermal constraints. Even when liquid cooled 8x350W isn't something you will be doing with a case mounted radiator setup, and a server rack full of multiples of those will require a special room/facility to handle the tubing, noise and cooling. Gotta appreciate AMD for their 64 core CPU's though - those things rock!!!

[@張歐-g9m]

16 sections, 1 reserved for integrated memory controller. So 15 cores each die max.

[@Luscious3174]

@@張歐-g9m Nope - the IMC has it's own dedicated real estate on the die. You can see it clearly in the shot. It's a 16 core part with two of the cores nerfed here.

[@JumpingJoseph]

Nice.. :)) hopefully you can etch another core cleanly :)

[@dabombinablemi6188]

I'm wondering if there is more consistent silicon quality in Intel's new multi die CPU. Some of their Core 2 Quad and matching Xeons used to have vastly different binning between them (eg. 1 of the die would run significantly hotter).

[@pino_de_vogel]

Those solder dots are so small im surprised the failure rate isnt a lot higher on cpu's these days .

[@LalaFafa]

That is not a BGA anymore ... that is an AGA -- Atom Grid Array. Mindblowing seeing such a big amount of contacts underneath the chiplet

[@longpham-sj5sv]

ok

[@longpham-sj5sv]

ok

[@Alan_Skywalker]

It actually has 64 physical cores. Currently my info suggests that 8480 will have 60 cores activated.

[@pamdemonia]

So cool and tiny! Thanks for sharing.

[@hydrocarbon82]

I wonder if each chiplet indeed has 15 cores...but 1 is always disabled. That would drastically reduce 'duds' since it could survie with a blemish on 80% of the die.

[@potatopotatopotatopotatopo8746]

I died on the inside after that delid fall. Never witnessed anything like this holy shit

[@MrSalty-eg4zb]

I would love to hire this guy to work on my pc

[@ColonelFrosting]

I'm still somewhat new to this channel and some of Roman's methods and was only half-watching when I heard "Grilling time" and I was like "you what-" and then when it cut to a charred Xeon I think my soul left my body

[@der8auer-en]

😹😹 sorry for that

[@deagt3388]

Be careful not to let that processor fall out of your trembling hand again! ;-)

[@triangle964]

best vid u made wow i am amazed

[@WiztotheIzzard]

Looks like delidding that one sideways will be the preferred method, in the future. More PCB frontage, and more heatspreader wiggle room.

[@Ang3lUki]

I can't wait to see how these perform

[@derohneusername]

Worse than Genoa. Perhaps even worse than Milan.

[@Ang3lUki]

@@derohneusername Doubt.

[@EvL--_--]

Would be curious about a 12900K direct-die oc test....

[@HypnoticSuggestion]

I want to know just how difficult it is to attach everything to the substrate, with these and Zen as well, must have required a lot of fresh engineering.

[@uhohwhy]

lol just glued 4 CPU in one, classical Intel

[@JasperLTZ]

11:50 now, that's the rgb that can improve your performance

[@ed.puckett]

Great closing music!

[@unh0lyav3ng3r8]

Couldn't you use a laser like the ones they remove rust with, or to make a logo.. and get it just do the whole surface in sweeps? Maybe there's a high tech machine shop near you could approach for advice. The newer manufacturing process has just made die shots using your current process very hard.

[@benjamintrathen6119]

I counted 16 cores per die, maybe a max of 64 cores?

[@ravenclawgamer6367]

Considering AMD's EPYC 7773X and Threadripper Pro 5995WX takes 280W and Sapphire Rapids take 350W (according to official specs, at least. I expect real power usage to be around 310W and 450W respectively), it's nuts to see that Intel's 56 cores take more power than 64 cores.

[@weirdsilence1914]

"human hair is 60 micrometers" der8auer literally splitting hairs at this scale : D

[@InfraredVisuals]

That's some impressive shots.

[@TheDukeOfZill]

Is hedt still useful? I remember starting with the i7 920 back when it launched, but really wanted a 980 extreme.. only to find it was ousted by the 2600k shortly after, which was technically a budget chip. Intel had been shooting itself in the foot each release where these cheaper K series chips were out performing their unreasonably expensive X series (hedt) chips. 9900k was my jump from the i7 920 and find zero reason to get a 13900k. Out of curiosity, I checked to see if a 13th gen X series would exist (last I saw, 10980xe was the last hedt) but we now have confirmation of a 13th series X chip, will it be slower than the K series for gaming?

[@NameLessSultan]

Love the Stargate symbols tattooed on your forearm

[@nnm711]

Congratulations, you probably killed a more expensive CPU than Linus! Each using your signature move, Linus by dropping and you by delidding. :D

[@nooneisback]

I how much the additional layer affects the chance of the solder balls cracking. It's already not that rare for CPUs not using this method.

[@Ryan-The-Grifter]

Do you upload your die shots somewhere? I'd like to use them for my desktop wallpaper

[@samiulislamsharan]

can you post a link to some of the Hi Res images of the die shot?

[@dorkasaurus_rex]

The delidding fail hurt me physically.

[@Patrick73787]

The first DDR5/PCIe 5 HEDT platform. Beautiful.

[@runninginthe90s75]

Also it glued better because Intel using EMIB, unlike AMD.

[@MissMan666]

So where is it ? Intel needs to get it's act together... here at my work (Government IT) they just swapped from Intel to AMD in a deal for the next 3 years cause AMD's offering was higher performance, lower power usage and lower TCO. The IT staff said the decision had to be made since there was no other logical choice as Intel has fallen behind for the last 3 years, they could not longer justify buying Xeons.

[@rkan2]

Could the tool not push from the "side" of the CPU? Then there would be a bit more space between the IHS and the caps.

[@unh0lyav3ng3r8]

He would need to make a new die mate as the cpu is rectangular.. he's got 3 left, hopefully we will see what he engineers in the next video

[@rkan2]

@@unh0lyav3ng3r8 Well yes, but he has a CNC :D

[@OTechnology]

It looks like it has 16-core per chip no? Wouldn't that be a 64-core chip? Maybe intel is disabling cores for yield reasons.

[@ChopperPBM]

Yep, exactly that, by allowing 14 cores per chiplet that allows for better yields, then when they start getting better and they get some chips with all 64 cores they can sell them for a premium as a new part

[@vukzlatkovic1183]

Hey, how did you go about eching those dies, I recently dug up some dead ddr2 ram and wanted to expose dies

[@TheReverendPlays]

I've got a few of those prototype boards behind me :)

[@georgiedagreek2734]

My man, I felt your pain.

[@Psychx_]

What's the on-package FPGA used for?

[@Hamlock_Maneuver]

Why not simply de-solder the components before doing the delid? seems pointless damaging the cpu for no real reason...

[@yetzt]

why not desolder the smd caps before delidding?

[@rahimdamji9602]

more alderlake delidding!! im wanting to do mine bad! more vids like this please

[@Zeno-]

...why would you need to delid yours?

[@rahimdamji9602]

@@Zeno- because I enjoy pushing things to max. Can lower temps 10 to 20c… basically 5.4 5.5 all core oc highly possible. Why wouldn’t you do it for free performance? I’ve done it on every Intel chip

[@ddnguyen278]

These modern chips are becoming incredibly complex and dense, amazing.

[@supercool_saiyan5670]

they already have been for the past 10 years :)

[@razter6678]

How to BBQ a CPU. Just needs some sauce and a biscuit now.

[@bubbafats6246]

4 E-cores in the space of 1 P-core, i wonder how strong a 256 E-core xeon cpu would be.

[@volodumurkalunyak4651]

Significantly weaker than P-core only design in AVX-512 - accelerated workloads (NAMD), should be around

[@Leoappeared]

What

[@Lisa_Minci96]

Latency would be pretty bad

[@volodumurkalunyak4651]

@@Leoappeared 4 E-cores take the 1,25 as much space as 1 P-core. 256/4×1.25=80 (P cores to take the same amount of space as 256 E - cores). 256 E cores vs 80 P cores is a valid comparison, not a 256 E cores vs 56 P-cores. This (56 functional P cores out 64 physically present) new CPU will fit around 208 E-cores within die area, some of those will have to be disabled for yield purposes.

[@劉奕彤-q6g]

@@volodumurkalunyak4651 10nm is still not dense enough to play this, we need 512 E-core in one of the sub-dies

[@floatingpointerror55]

Watching you kill this cpu just killed my Soul... Subscribed!

[@ralmslb]

11:44 If we consider each of the middle square in the middle of the die which has that bar on one of the sides a core, wouldn't it make each die have 16 cores? For a total of 64 cores? That would make more sense in order to compete against AMD.

[@Harish2923]

The right most square on the 3rd row is not a core, it's the IMC, so it has 60 cores max and AFAIK, the top model has only 56 cores(1 core disabled on each for yields)

[@suntzu1409]

Considering how much better alder lake is compared to Ryzen 5000 desktop in single thread, i doubt intel needs 64 core to at the very least seriously challenge 64 core EPYC.

[@suntzu1409]

@@Harish2923 "Its the IMC" Source?

[@Harish2923]

You can always find the IMC on the mesh network, they are most of the time identical size with the cores, you can also find it on icelake SP die shot

[@ralmslb]

@@suntzu1409 Loool, that is not true, not sure what reviews you have been watching. Most of the consumer Alder lake products are just highly overclocked cpus from factory, the 12900k with the power limit disabled running stock, is half as efficient as a 5950x. On server hardware, this type of behaviors won't be accepted, as having a cpu consume twice the power as the previous generation, will mean double to cost of operation. At this time, Intel has no response to AMD server lineup

[@AndreiNeacsu]

Is Intel using glued-together desktop chips in their server CPUs?

[@stanislavtihohod]

Microelectronics is starting to resemble alien technologies

[@averyoldYoutubeuser]

Absolutely incredible 🙂

[@CaseyHancocki3luefire]

Those caps look large enough to manually solder/de-solder before delidding the CPU.

[@der8auer-en]

that's a good point :)

[@XantheFIN]

Or just add some adapter block in "delidder" to avoid them... (?)

[@user-dn5bx2iu3e]

Bro, lmfao. This is why you're the best.

[@UnknownProductions0]

are you going to sell the delid tool for these?

[@fVNzO]

Is EMIB a passive interposer?

[@suntzu1409]

As far as i have seen, yes

[@92trdman]

Can do the E3 or Phenom II ? Danke schön

[@12100F]

crazy to me that this was the best Intel had to offer just 2 short years ago.