ITS OVER INTEL IS FINISHED

>Hot on the heels of the 5nm announcement in partnership with IBM Research and Samsung, Globalfoundries is making two announcements around its 7nm manufacturing process. While the 5nm process is still in the test phases, Globalfoundries 7nm is much further along. In fact, today, Globalfoundries is announcing that their 7LP (Leading-Performance) process and FX-7 ASIC platform are available today for partners to start planning to utilize with design kits.

>The 7LP process is a continuation of Globalfoundries low power work in previous nodes like 14nm with a focus on cost and power. However, with the introduction of 7LP, Globalfoundries is touting a 40% device performance improvement over 14nm while also talking about reducing power by up to 60%.

>Globalfoundries 7LP process uses optical lithography, but is EUV “compatible” if a customer’s designs require it for one reason or another. Globalfoundries also claims an over 30% reduction in die cost with an over 45% reduction in target segments. 7LP is still a 7nm FinFET technology, with an expectation that the company will commercialize it in 2018, which isn’t far off, likely thanks to the strong early customer engagement they’ve seen on it.

>Products expected to launch in 1H 2018, ramping volume in 2H18.

>7nm FX-7 for ASICS
In addition to 7LP for high-performance logic and caches, Globalfoundries also announced initial customer availability of FX-7 for ASICs. FX-7 is a follow-on for FX-14 which has a leading 56G SerDes. This process would be used to build chips for hyper-scale datacenter acceleration, automotive, 5G networking and machine learning and deep learning applications. ASICs are short for “Application Specific Integrated Circuit”, which you may be familiar with relating to audio and video decoding and even machine learning like Google’s TPU. Design kits are available now with volume production expected in 2019.

>>Products expected to launch in 1H 2018, ramping volume in 2H18.

Other urls found in this thread:

forbes.com/sites/patrick.../#6b4b6e444730
globalfoundries.com/sites/default/files/product-briefs/7lp-product-brief.pdf
twitter.com/NSFWRedditVideo

Cont:

>In addition to the 7nm FX-7 and 7LP news, Globalfoundries in partnership with IBM and Samsung research teams together at the VLSI Technology and Circuits conference in Kyoto, Japan, announced the feasibility of 5nm with a functioning 5nm chip. This is a big deal for Globalfoundries and their partners because there were major doubts that anything under 7nm would be possible or coming any time soon. However, the research and testing conducted using Extreme Ultraviolet (EUV) lithography technology. Using EUV along with stacked nanosheet technologies IBM in partnership with Samsung and Globalfoundries built a functioning 5nm test chip.

forbes.com/sites/patrick.../#6b4b6e444730


INTEL JUST GOT DEMOLISHED

RIP Intel

BASED IBM WIZARDS

oy vey it's like anudda shoa

the fourth? holohoax

Brakes on the rape train? Nah, add boosters.

The ride never ends

SHUT IT DOWN GOYIM.

You're not funny.

Oy vey. Delid!

You can't, GloFo is owned by terrorists :)
Murikka and jew lawyers can't do shit

You're not funny.

>anudda shoah

FFFFFFFFUUUUUUUUUUUUU-

globalfoundries.com/sites/default/files/product-briefs/7lp-product-brief.pdf

Product brief.

You're not funny.

...

...

>5ghz
Let the shoah begin.

Oh my God

>Globalfoundries also claims an over 30% reduction in die cost
Single most important bit of information is right here. Radically reducing cost on a process that is radically more complex is not something which happens by chance. This process would be employing new alignment techniques developed by IBM.

The future of Global Foundries is completely disruptive. The semiconductor industry is going to be making massive strides in the next 3 years. This is a bigger jump than moving on from strained silicon. Implications are tremendous.

Meanwhile at GF and AMD

>guys we got no high clocking process to compete with Intel at the very high end, what do we do?

-Make their own process?
>Nah, too costly and time consuming

-Tweak Samsung's process?
>Nah, could backfire and it's not guaranteed to hit high clocks

-.. Call IBM?
>GENIUS

I was planning on getting Threadripper this year but I'll hold on to my 3570k for a while, Zen2 should be both higher IPC and higher clocks, no way I'm not waiting for this gem.

Actually it was IBM that called GloFo and PAYED them to take their fabs.

You're not funny.

That's what Intel thinks of its situation too.

>up to 17 metal layers

That's IBM's work allright

>5ghz
Gotta go fast.

Meanwhile 10nm Coffee Lake is delayed to 2018 and the 8th generation core i series is 14nm yet again, a refresh of a refresh.

I didn't believe it when I heard that Intel didn't think AMD could actually compete and relaxed their development. But the writing's on the wall now.

Time to start wildly fucking speculating about Zen2, guys.

6 core CCX/ 12 core die
24 thread mainstream parts
48 thread workstation Threadripper replacement
96 thread enterprise chips
4.5ghz+ clocks at 95w for 12 core mainstream
3.5ghz+ for 65w parts with 12 cores
6 core/12 thread APUs
Higher IPC
higher frequency DRAM support beyond 4000mhz
Data fabric limitations dramatically reduced, higher speed Infinity Fabric all around
lower latency IMC
still on socket AM4

>makes hamfisted attempt at comparing calls of Intel's demise to Trump
>too lazy to even edit the image

0/10 see me after class.

>ryzen running at 5 GHz and with 6 core ccx
[Sweating intensifies]

If they can manage a 10% uplift in IPC in common workloads they'll absolutely slay.
>making substantial gains in IPC and clocks in one gen
>without increasing power
>while adding cores

Someone stop Su. This is absolute savagery.

I don't think you're gonna serve a 12 core with two memory channels.

4die MCM will still have 8 channels.

We have 8 core chips with two memory channels running 2400- 3000mhz DDR4 typically.
Thats 48GB/s memory bandwidth with 3000mhz memory. You could break it down to 6GB/s per core, though this obviously isn't a very realistic metric since all cores will never be arbitrarily just performing endless memory ops. Still its napkin math, and everyone loves napkin math.

4000mhz DDR4 is 64GB/s. That'd give 5.33GB/s per core for a 12 core dual channel chip. Not a huge reduction from what we're at now, and looking at current trends we'll probably be seeing significantly faster DDR4 a year from now.

Besides AMD said they'd stick with socket AM4 until DDR5 was entering the market. They aren't adding more memory channels, but we have at least some indication that they are adding cores to the CCX.

Yes..? But 1x die(AM4) platform is unlikely to have 12 cores and two channels, especially if the cores are notably faster from higher frequencies and IPC, I think they're gonna have two different dies, one for server, one for desktop.

Or they could actually do 12 cores on 2 channels, but the memory controller would have to be significantly more powerful than it is now.

hopefuly AM4+/AM5 wil be compatibe with gen1 ryzens

This is what I think.

AM4 CPUs will be stuck to 8 cores, which is fine.
AM4+ will go up to 12 cores, because they obviously require different motherboards.

I don't think AMD is going to make the financial investment to start fabbing multiple dies when they're just now starting to turn around. They'll be sticking with the most cost effective measures just as they are now.

Reminder that shoah starts in 7 days.

what doe the layers do?

You're overestimating the cost of a new photomask, it would be a short term hit on AMD's cash but certainly a good long term investment, it lets them get the most out of the server chips.
These won't be big chips anway, smaller than Polaris, the mask should cost some 200 million USD

Route power through the chip, more layers = usually more power hungry chip.

ex: Intel's Kabylake is 12 layers, Ryzen is 11, but it also has twice the cores and a bunch more uncore.

does this mean ibm-based ryzens will be more powerhungry?

What are they gonna to do with shitty binned server dies?

It says "up to", 7nm Ryzens probably won't go over 12 layers.
Those 17 ones are for IBM's POWER chips.

Threadripper 2

Volume is too small.

Intel is adding more cores to all their chips in every segment, and AMD isn't likely to fall behind.
Nor are they going to tarnish their newly won respect by segmenting their platforms again.

Metal layers are signaling, power routing to the logic, and the logic/sram/edram/etc itself. More layers means you can build a denser chip depending on how your cells are oriented. IBM's POWER series of processors have more metal layers primarily because of their immense and well connected EDRAM.

>5ghz


how will they keep it from going full chernobyll?

Not use TIM

>7nm HIGH POWER
You got your answer.

By not using toothpaste for the IHS.

Intel needs new management
>vague legal threats at MS/APPLE because they want off x86 as if that can stop them
>AMD kicks their ass with ryzen/threadripper
>the threadripper panic mode fiasco
>trying to get into new markets like iot and mobile and failing every time
>no dedicated GPU line despite making GPUs that can already play games
>Nvidia partnering with IBM to get general purpose CPUs for their ML setups, literally don't need intel anymore
>"x86 or death!"

Intel management basically lost contact with reality.

>.75v drive voltage

That's low? How does it compare with current 14nm nodes?

Ryzen's vmin for hitting 3600mhz is around 1.1v or a bit lower depending on luck of the draw. With Ryzen it seems that .7v is the min voltage required to get reliable switching. If .75v is the nominal drive current for 7nm then the low end is going to be drastically lower as well.
60% lower power for a design at a fixed clock would be quite outstanding if it materialized in an end product for AMD. They'd have a full 1800X under 45w.

>They'd have a full 1800X under 45w.
*hysterical laughter*

>I can't handle people making jokes on the internet because my skin is far too thin

>denser
Does this mean they'll try and stack chips?

You're not funny.

No, it means higher margins for Dr. Su. Or bigger dies with moar everything.

And jews are not white.

I think the future lies within chip stacking

Yes, but software is so far behind we have to wait. For now we need for kikes to finally adopt fucking HBM.

Yes, the process characteristics are that significant.
25w~ 1700 would be a possibility.
It'd be no challenge at all to put that in a reasonably thin laptop.

This 7nm process from IBM isn't bulk silicon based, its IBM's advanced FD-SOI. Its actually their 3rd generation of SOI based FinFETs they've designed internally. There is industry changing IP at work here.

No, stacked logic is still a long way off.
When I said density I mean in the structures on the front end of line. DRAM cells can be oriented so they're partially overlapping by having parts of the structures on alternating layers. The net effect is you get massively more transistors per mm2.

Imagine an EDRAM cell looking like a 3D letter "L" and you're stacking them so the legs are overlapping one another.

>SOI-based FinFET
I though they went for SiGe fins.

IMPOSSIBLE, HOW DID THEY GENERATE SUCH INNOVATION WITHOUT INVESTING IN FEMINISM?!?! I HATE MEN!

Wasn't that for 5nm?

The channel material likely is SiGe, or something similar, but the process starts with an SOI wafer, and there are no quasi neutral zones below the gates. They're all fully depleted of carriers.

>CMOS bulk FinFET
What did they mean by that?
Nah, 5nm uses silicon nanosheets.

*melancholy of no stacking*

...

Wow, not only was Intel's only good point its fabs, it lost that advantage now too, now they have both a worse deadend arch and worse or at _best_ equal fabs.

The next 4 years will be a rough ride for Intel

Exciting times to be alive, user.

>FD-SOI 7nm with quad patterning
>FD-SOI 7nm with EUV
>450mm wafers
>5nm GAAs faster to market than anyone expected even 2 years ago

AMD already stated that Zen is gonna last some 4 years, so 2021, right when Intel releases its new arch.
Clearly AMD is not sitting on its ass, what are these newgen architectures gonna be like? Still Von Neumann? RISC-V-like on x86? Just regular x86 but chalk full of accelerators for desktop tasks?

>Clearly AMD is not sitting on its ass, what are these newgen architectures gonna be like? Still Von Neumann? RISC-V-like on x86? Just regular x86 but chalk full of accelerators for desktop tasks?
We don't know. At least AMD is part of RISC-V foundation.

>Tfw am dumb and don't understand why this is a big deal

That's why I included the translation in the subject field for your type, you know that blue text above the post?

You're not funny.

>5nm GAAFETs
How in the FUCK?

You are funny.

At this point looking forward even 3 years its damn near impossible to tell whats going to happen. There have been major boons to the industry that are just pouring money in, and there is fiercer competition between foundries today than ever before.
The automotive industry is driving production of RADAR, and LIDAR sensors, along with ARM SoCs for infotainment systems, a even full GPUs for self driving systems. More and more LED lightbulbs contain a tiny IC for controlling flicker rate, or even for adjusting the brightness, or hue of the bulb. Ever take a look at Broadcom's stock? The radio market has been steadily rising without fail for years, and annual shipments of radio devices have never been anywhere near this high.
Samsung being the giant they are have been putting endless billions into their foundry business to compete with TSMC.
TSMC is trying to stay competitive enough that intel doesn't have a chance to really take off as 3rd party foundry business.
GloFo's acquisition of IBM's foundry business and subsequent IP/engineers/tooling took a shaky business and have propelled them to the top of the industry.

Theres more money flowing, and more R&D going on than ever before.
Speculating on processor architecture a few years out is total guesswork. The things that these foundries will be able to facility will be absurd. 10b transistor GPUs will be entry level. A CPU like Ryzen will be in a sub 15w ultra book.

Shit is going to be fucking wild.

The finest Nazi science that a DARPA funded research company has to offer.

Now that I think about it, silicon runs the entire world, why don't governments and countries literally throw money at fabs and chipmakers, compared to other companies, their profits are small time, yet they provide hardware, possibly the most important cog in the system after electricity itself.

They do, just not directly.
These companies get massive tax incentives and all kinds of special treatment. They just don't get blank checks for R&D, but the end result is near the same.

In the case of Samsung, they're basically like 1/3 of Korea's GDP, and the head family that runs everything is pretty much married to the government. For all intents and purposes Korea is the nation of Samsung. They even used to produce military weaponry but decided to sell off the division because it wasn't making as much money as smartphones, but they created the automated turret system that guards the DMZ.

lol. well, intel shill, maybe your company would not be falling apart right now if you spent your money on actual products instead of paying taxes to feminazis who fucking hate everything about enlightenment civilization?

as a technology company, your ability to employ the principles of the enlightenment is fundamental to your ability to compete in the marketplace. if you shit on that, you are fucked. you did shit on it, and now you are fucked.

do not blame me, blame yourself.

I want radar in my phone

>AMD releases 5GHz stock Zen+ with over 10% IPC improvements
>destroys intel in almost* everything
* - if only they fixed their fucking IMC latency

Until graphene takes over, but that wont happen until the far future

any chance for a GAYDAR sensor?

>muh gaymergate!

What? Intel literally donated $300million to fucking Feminist Frequency instead of spending the money on fucking R&D.

>muh gaymergate!

So why is all this innovation happening all at once?
The CPU market has been shit for what 5years now?

What the fuck IS gaymergate?

AMD is back and they want money and jew blood (preferably jew blood). Also RISC-V is in constant development and Common Platform Alliance is shitting nodes every fucking year.

tell me more about these... nodes

The stars just happened to align.
AMD really lucked out on the foundry side of things to enable their new arch to be super efficient.
The rest of the foundry industry has more cash than they know what to do with so its all getting pushed right into R&D.
Expanding semicon industry means more money into all the peripheral industries around it as well. Having money to blow means being able to hire more engineers, and it just goes on and on.