IBM company: 30 billion transistors squeezed into a fingernail-sized

I'm fucking hyped
This is the first we've heard from IBM in the tangible computing market in a long time and I'm excited to see development on it.
Will buy stock.

>>60789865
>I thought IBM was a collapsing company
what gave you that idea ?
they just stopped giving a shit about the consumer market

Hold the fuck there cowboy, prove it's actual logic and not easily shirnkable DRAM

>>60789973
IBM's test chip on their 5nm node almost certainly was a POWER9 design of some sort. They did the same thing on their 7nm node. Their test chip was in fact a fully functional POWER part.
As such a large amount of transistors are spent on DRAM.

Yeah, not buying it, Nvidia's GV100 is 21 billion transistors in 820mm^2

That IBM process would need to be 15 times smaller.

>>60790008
And 10 times slower :))

>>60790008
>GV100
TSMC's 12nm node is an incremental update to their 16nm FF+.
It is in essence no different from Samsung's 14nm LPC, TSMC just changed the number for some inane reason.
The area scaling from 14/16nm down to 10nm is quite substantial. We're talking about a 5nm class node here.

>>60790045
Yeah, it's substantial, but not FIFTEEN TIMES substantial, even Intel admitted their 14nm > 10nm density is close to double.

>>60790008
>>60790045
Also IBM cares a lot less about yields so they layer shit THICC.

>>60790068
Whats their 14nm POWER9 at? 17 or 19 metal layers?
AMD's Zen parts are 12.

File: ZCuzgCl.png (273KB, 475x473px) Image search: [Google]

273KB, 475x473px

>>60790008
3D>2D

>>60789865
>IBM
>collapsing company
They left consumer market.

>>60790166
what do they even make besides super computers?

>>60790088
It's 19 metal layers. The thing is so fucking THICC.

>>60790185
Mainframes I guess. And they invest fuckton of money into R&D.

>>60790185
Top secret Nazi science for DARPA. Not joking.
If a war were to break out IBM is considered vital infrastructure to the US. IBM employees are some of the people who would get priority access to doomsday shelters so the company could keep running.

>>60789865
>Nintendo
I don't think it'll collapse anytime soon. Probably they have narrowed down their playing field.

>>60790214
>Nazi science
Come on Brian, no matter how many NSA backdoors you will put into Intel chips you'll never be more valuable than IBM.

>>60790185
lots of research
experimental technologies (e.g. that fuzzy chip for AI)
Watson

>>60790231
You're mistaken. I am not using the term Nazi science in a negative way at all.

It's so fucking hilarious when IBM's fab advantage can be used by AMD.

File: typical.png (177KB, 716x429px) Image search: [Google]

177KB, 716x429px

>>60790242

>>60790250
what?

>>60790236
something besides government has to fund that

>>60790265
IBM merged it's fabs with GloFo, anon. GloFo and Samsung are IBM's partners in IBM research alliance.

>>60790265
Global Foundries acquired IBM's foundry business, engineers, and IP.
In turn they're offering IBM's processes to AMD.

AMD will be bringing 7nm FinFET parts to market next year utilizing IBM's 7nm process node employed by GloFo Fab8.

>>60790279
Wow and Intel is having issues breaking 10nm. Is this the end?

>>60791896
Everyone is measuring this shit differently.

>>60791896
Intel has quintessentially intel problems. Same issue they had on 14nm, though its alleged to be worse.
Management made a decision to pursue industry leading BEOL density. The foundry business however had so much trouble actually realizing it that yields were poor for well over a year. A bean counter treating engineers like a cog in the system who mindlessly obey instead of listening to their input.


>>60791913
They're not, you're just reading far too much into a simplistic marketing name. Back end area scaling metrics alone aren't the end all be all. Far from it.

>>60791949
So you're saying Intel has been fruitlessly beating their heads against 10nm for years because .. they're stupid?

>>60791981
You'd be surprised how many CEOs issue impossible directives to their charges. Same for every area of management. The only thing that saves a company like intel is having enough money to hire enough of the right people to pull it off by the skin of their teeth. In this case however intel tried to bite off more than the engineers could chew.

There are solid business reasons for trying to do what they did. Running their own foundry has some implications. The foundry is technically a separate business, the X86 design part of the company is a separate business. The X86 business is the foundry's customer and they have to pass purchase orders back and forth. The X86 business wants to get the best deal possible, meaning the lowest price, since at the end of the year they have to write it all off.

Management decided that more candidates per wafer was the viable option for lowering costs, believing yields could be brought up rapidly, and the work done in tackling such a lofty goal would pay off when it came to ramping up future nodes.
Management was simply wrong, and the foundry business didn't deliver on what was asked.

>>60789908
Great choice IMO

>>60790185
Military and aerospace i guess.

Is 5nm the last node? ;___:

>>60792180
Of course not.

>>60790250
IBM has no fabs anymore :)

>>60792041
Wasn't their 22nd trigate also quite a Conservative node? So Intel had their 22nm go online whilst no one else managed a high performance 20nm node. (although I do like what tsmc 28nm did with excavator and maxwell)

Conversely the 14nm Intel too dense and now the competition has caught up with less aggressive finfet nodes.

>>60792213
22nm Trigate was on paper very substantial when looking at scaling metrics, sram cell size, and things like that when compared to their prior 32nm planar node. Though it wasn't radically more dense than upcoming 28nm planar processes offered by competing foundries at the time.
At the time though intel was quite a bit ahead of everyone else on the market when it came to feature size. I believe only IBM was keeping up. Global Foundries, TSMC, Samsung, and Chink fabs were a couple years behind.

File: 1488989916520.jpg (96KB, 786x509px) Image search: [Google]

96KB, 786x509px

THANK YOU BASED IBM!

INTELJEWS AND AMDPAJEETS ON SUICIDE WATCH!

>>60792353
The RAM issue was pretty much fixed last AGESA update. http://www.game-debate.com/news/22985/amd-ryzen-agesa-1-0-0-6-update-adds-support-for-memory-clocks-up-to-ddr-4000mhz

>>60789908
They're gaymers, they think that CPUs are being made only for games.

>>60792390
MUH 640x480 gaymen

>>60792199
after 3nm it is

>>60792428
Engineers don't stop engineering, anon.
The lab created tests of today were the theory of yesteryear.
That means that when a practical proof of something comes to be there are mathematicians, physicists, and engineers theorizing on something which well surpasses it.
The study of quantum junctions is already a thing.

Proof of a 1nm device from October last year:
http://spectrum.ieee.org/nanoclast/semiconductors/devices/onenanometer-gate-dimensions-for-transistors-have-been-achieved

>>60792353
LMAO you know that AMD will get to Fab on this process right? Intel is pretty fucked though.