Intel and TSMC on suicide watch, IBM Nazi science rules once again

Intel and TSMC on suicide watch, IBM Nazi science rules once again.

semiwiki.com/forum/content/6879-exclusive-globalfoundries-discloses-7nm-process-detail.html

Not mentioned is that TSMC and Glofo are the only ones that can hit high frequencies out of those, Intel's 10nm+ is coming latter.

Other urls found in this thread:

electronicsweekly.com/news/business/ibm-samsung-globalfoundries-develop-5nm-gaafet-ic-2017-06/
twitter.com/AnonBabble

DELET DELID THIS RIGHT NOW THERE'S NO WAY A HUNGRY AND MORE AGILE COMPANY COULD BEAT INTEL WHO HAS RULED SEMICONDUCTORS FOR OVER 20 YEARS IT'S NOT POSSIBLE

Daily reminder.

>Intel's 10nm has higher mask count than 7nm LP
>Intel's is not even TALKING about using EUV
Oh fuck.
This is bad.
Really, really bad.
Like, new levels of bad.

While Intel will get 10nm++(which is about as good as a half node) in 2020, Glofo/Samsung will already be on 5nm using GAAFETs

electronicsweekly.com/news/business/ibm-samsung-globalfoundries-develop-5nm-gaafet-ic-2017-06/

Intel is getting FUCKED

>Not mentioned is that TSMC and Glofo are the only ones that can hit high frequencies out of those, Intel's 10nm+ is coming latter.
Looks to me like all the nodes besides GF/IBMs one is focused on cost/density, no high frequency operation at all.

Hoo boy IBM really wants the node crown.

Yes, only 7nm LP is high performance from the get-go. Thus AMD gets the node lead first time in semicon history.

Whatever is after Zen3(7nm+) is likely gonna be using that 5nm node.

It's gonna be a slaughter

THANK YOU BASED NAZI SCIENCE

Don't forget MCM bibecards, friend.

I have little interest in GPUs

Can't wait for 2nm Zen6+++

I don't think AMD will iterate Zen for that long if they have the R&D money.
AMD loves R&D.

AMD said Zen will last 4 years, as long as the AM4 platform.
So there's very likely a new arch after Zen in 2021

Reminder than AMD will have a LOT of money very soon for all the R&D stuff they always wanted to do but never had money for.

Can't wait for the new intel marketing strategy:
10>7 therefore our chips are better

Yeah and Vega is going to BTFO nvidya

inb4 AMD makes a better arch when they're poor than when they're loaded.
Then again they did make K8 and K7 when they were loaded so who knows, at least they have good management this time and for all intents and purposes their exclusive fabs without the cost of maintaining them

Vega? No, its simply competetive. Navi? Maybe, but it's gonna be hard.
Ye, no Ruinz helps a lot.

Gayming toys are useless, I hope AMD drops out of gaymer GPU market soon and focuses completely on CPUs
They can continue making small GPUs for APUs and other stuff, but big shit like Vega? No need.

AMD still want's at least some of the meme learning pie.
And bega is bretty good for meme learning since meme cores are irrelevant when ASICs exist.

Drop all huge monolithic die R&D down the drain, focus on making sub 300mm^2 chips and try to get them somehow to work together in a MCM.

That should be AMD's goal.
Navi might or might not be this, but this is obviously where they should focus

Navi is obviously that.
They're several years ahead of everyone in a race to MCM GPUs anyway.

>They're several years ahead of everyone in a race to MCM GPUs anyway.
lmao do you actually believe that?

>Navi is obviously that.
That's RUMORED.

But I certainly hope that's the case.
There's no beating the sheer amount of silicon MCMs allow.

They have working interconnect in real retail silicon.
>That's RUMORED.
It's pretty obvious since Vega has some kind of IF links in it. Be it GMI or simply translating IF over PCI-E.

MCM has been around for decades. Why do you think AMD's shit is anything special?

Have a wild guess.

Hi Brian

Why don't you tell me?

Because you and your Oregon dudes will have to reverse engineer it, Brian.

Say it with me

>SLOWER THAN 14NM

TSMC looks identical to GF, why should they be on suicide watch?

Because TSMC is finally getting a competitor that's not Intel and TSMC's 7nm is apparently low power as well.

It's lower power. And no EUV any time soon.

well GF is LP as well? Thats why nvidia is doing refresh at 12nm, 7nm for gpu's will be available at like 2019~

No, LP means Leading Performance (aka IBM speak). It's intended for 5ghz operation from the get-go.
PREPARE FOR SHOAH BOYZ

This is exactly what's wrong with fanboys. You actually trust what marketers tell you.

>interconnect
>some
>marketing interconnects
The fuck?

thats so retarded naming it like that i just assumed it meant low pwer

...

That's IBM-speak for you. Anyway yeaaah high power 7nm sounds like the worst nightmare for Intel.
Like, period.
They can lose even HPC markets.

What makes AMD's interconnects special? That's what I want to know. They've existed for decades.

well considering IPC is literally the same and only differnce is basically higher ghz, if this 7nm is really that and reaches 5ghz, it will definetally fuck over intel.
Zen2 is the 7nm LP right? when is it scheduled?

I don't want to sound like 3rd work marketer, but SCALABILITY. And simplicity since IF it totally agnostic to physical transport layer, EPYC uses PCI-E lanes for socket-to-socket links.

>Zen2 is the 7nm LP right?
Yes.
>when is it scheduled?
Tapeouts for Zen2 (and supposedly Navi) are Q4 2017 so it's Q4 2018 at worst for Zen2.

it's rumored a 10% ipc increase too, so it will have both ipc and clockspeeds parity with intel's offerrings

The difference between fab processes isn't performance versus power, it's efficiency versus price/yields/time to market. Notice the sizeable performance difference between GloFo's 14LPE (low power early) and 14LPP (low power performance) processes. In this sense, low power = high efficiency = high performance in a given power envelope. The difference is between moderate efficiency with better yields and a faster turnaround, and higher efficiency with lower yields and a longer development time.

The 7LP nomenclature probably still means low power to the engineers, it's just the marketing team rebranding it as "Leading Performance" for consumers derping about how much they want leaky housefire chips instead of good products.

Also, note that Intel still has the lead in fab technology. Intel's 10nm process is actually 10nm, while Common Platform's 7nm is actually 14nm, and 14nm is pretty much just their 20nm process with FinFETs added. Intel might lose their lead if EUV pans out, but so far EUV has proven to be something that sounds like the future but is actually unmanufacturable bullshit like graphene. Intel funneled billions into EUV development back in the P4 days, and their failure to get anywhere with it was the last nail in Tejas's coffin.

>while Common Platform's 7nm is actually 14nm
Sup Charlie!
No, 7nm LP is both high power and a bit denser than Intel's 10nm.

Samsung and GF 7nm is different, as is TSMCs

Their densities are all close to Intel's 10nm, while promising a increase in density with EUV the year after, but only GF's 7nm is designed for high frequency(5GHz) since it's a IBM process for POWER

All the numbers are in the OP, there's barely any difference between the processes on the density front

>Intel
>rish production Q12017

Ahahaha, yeah, and we're getting first MOBILE SHIT in Q1 2018?

What the fuck is this company doing? There's a very good chance AMD will have 7nm products in late 2018 on inferno clocks while Intel will still be fucking around with 15W toys

>What the fuck is this company doing?
Sucking dicks.
>There's a very good chance AMD will have 7nm products in late 2018 on inferno clocks
Not a chance, AMD is doing tapeouts of 7nm products Q4 2017.
Having your own fabs is both a blessing and a curse.

One year from tape out to retail doesn't seem bad.
Vega taped like.. September or somewhere late summer, not even a year from its launch and its's a huge 500mm2 die.

In comparisons having a 200mm2 die sounds much easier to do in a year.

Vega10 is 484mm^2 and it's so late due to being new uarch and RTG being unable to write working drivers in time. Hell, it took nVidia delays to launch Maxwell2 and nVidia is one hell of a powerhouse of software development.
So Zen2 will be swift and dealdy for Intel.

Am I the only one hoping AMD bloats the silicon and puts out a 330mm^2 zen2 die? Imagine 4 of those in EPYC2

There's no point to. Zen2 will probably be moar goarz and nothing else. AM4 is not designed for moar memory channels or moar lanes.

>I really wanted to sound like I know what I'm talking about
>I just copied words I saw on Sup Forums
Thats cute, kid. I suppose imitation is the sincerest form of flattery however. Global Foundries is offering a 7nm node, that is denser than intel's upcoming 10nm+. Smaller HD SRAM cells, lower total area per xtor, all along with higher fmax and lower vmin to boot. Simply put it is more advanced on the FEOL as well as the BEOL.
Intel has fallen behind the curve. The best part? Global Foundries will be offering a refined variant of the same 7nm node they've developed, but with a phased in EUV light source. The increased resolution will see yet another area reduction, as well as lower vmin from reduced variability.

When Global Foundries has their 7nm EUV node online and in production the construction on intel's 7nm facility might not even be finished yet. They've got a long way to go before they ever handle any 7nm wafers. They're no longer the foundry leader, and you can thank management for that.

If it isn't desinged for more channels then it can't have more cores you twat, you're contradicting yourself.

Oh well it sounds so-so-so bad for Intel.
Fab advantage was the only thing going for them.
Zen2 will demolish both desktop and mobile markets.

"Low power" when its even applicable, is a relative term. Low power does not mean low frequency. Global Foundries is targeting a nominal fmax of 5ghz with their 7nm DUV line.
14nm LPP aimed at >3ghz.

Trying to write something off based on terminology you don't even understand is just silly, user.

I like the term "SoC/Mobile process"

Zen2 worst case
5% IPC 10% higher clocks

Zen2 best case
10-20% IPC, 25% higher clocks

In both cases Coffeelake would have a hard time competing.

>10% higher clocks
Maybe base clocks and they'll advertise it as an overclocking champion.

I'm talking about Fmax, base clocks are generally useless since Turbo/Boost showed up

If Zen2 had a base clock of 3.7ghz it'd be a pretty substantial uplift. They could easily work out SKUs with all core turbo at 4ghz instead of just having single core there, any IPC uplift would be icing on the cake.
3.9ghz to 4ghz base clock for an 8 core part would be pretty game changing for AMD at their current levels of performance.

Assuming that 3.4ghz is their nominal base clock for comparison.
It would be significant industry headlines if AMD had a Zen2 chip with 10% higher IPC clocked at 4.5ghz on all 8 cores inside of 95w.

Spec-wise (OP), it looks like Intel's 10nm is about par with the rest of the 7nm line, but they have worse efficiency and lower clocks? what's the reasoning behind this? Doesn't everyone buy the same machines from ASML?

It's just tuning a process to target different frequencies, ASML's equipment only allows them to make the stuff, they're in charge of deciding what to do with the process.

That why when planar was still cutting edge,, foundries usually had several processes targeting different frequencies or densities on the same node

So is it just because Intel was trying to rush 10nm/7nm? If they were on target back in 2015/2016 they would have beaten the competition by a full 3-5 years, even if they had worse eff/clocks. And then they could use that time to improve the process to fix the issues.

The tooling isn't the issue, its the IP they're trying to build, and how they try to build it.
Intel's major issues started with their 14nm node back in 2014 when they first tried to launch Broadwell CoreM. The 10nm saga is just a continuation of their 14nm woes.

If Intel was on time they'd have 10nm ready in late 2015 or early 2016.

They clearly were not on time, either management took a shit, or it was far more difficult.
I'm inclined to believe it was far more difficult since they haven't lowered fab R&D, the mask count in the OP is quite telling.