AMD Has Plans To Create The ‘Starship’ Processor With 48 Cores / 96 Threads On The 7nm

wccftech.com/starship-amd-processor-48-cores-7nm/

>inb4 moar cores

Other urls found in this thread:

www-03.ibm.com/press/us/en/pressrelease/47301.wss
pcworld.com/article/3020172/ces/amds-lisa-su-explains-how-amd-will-survive-and-yes-even-thrive-in-2016.html
twitter.com/SFWRedditImages

>wccftech
designated shitting forums

The data center and HPC markets aren't going to want less threaded hardware any time soon. It only makes sense to continue to add cores when its economically viable, and process allows inside of reasonable power envelopes without sacrificing too much frequency.

We have 32c/64t Opterons on 14nm LPP. Seeing 48c/96t chips on a process technically two nodes smaller than current isn't exactly earth shattering news.

AMD doing stupid shit yet again, what a surprise

>7nm

doubt this rumor is credible, 7nm won't be here until 2019 at the earliest, i doubt amd even knows the specifics of the process or how to design for it yet.

Cant wait to pay 5 grand for it

www-03.ibm.com/press/us/en/pressrelease/47301.wss

Global Foundries absorbed IBM's foundry business.
They'll be offering this 7nm SOI based process to AMD, the wafer supply agreement as just amended for it. As a matter of fact IBM started detailing technical aspects of their then in development 10nm and 7nm nodes back in 2012. The process IP is complete, and more than a handful of test wafers with complete functioning chips were produced on it.

14nm SOI FinFET is online and running right now, its the process used for their new POWER9 based chips.

lol no. AMD has already stated that they will be on 14nm for 3-5 years.

pcworld.com/article/3020172/ces/amds-lisa-su-explains-how-amd-will-survive-and-yes-even-thrive-in-2016.html

>AMD is establishing a number of products on 14-nm FinFET in 2016, Su said. “But it will be a long node. It will last three, four, five years. But within that node we can do a lot in optimization, and within that node, we can do a lot on power... once you’re in the node, it’s all about architecture.”

also, IBM's fab business failed horribly, I wouldn't expect their 7nm to be anything cutting edge relative to their competition's process tech. We could be seeing yet another rendition of GlobalFoundries being stuck with subpar/stolen tech.

>look at me shitpost
Making use of a process node does not mean they'll leverage it for high performance parts. They'll continue to produce mid range and entry level hardware on 14nm FinFET for years because its priced to become the next commodity node like 28nm bulk was.
Sounds like you're the exact same tech illiterate retard for yesterday who tried to claim IBM had never done anything innovative when it comes to process tech. As a matter of fact, I know you're that exact same shitposting autist.

IBM developed a ground breaking SOI based FinFET process. Their 7nm node is their 3rd generation FinFET, and the first ever to utilize a fully SiGe channel. AMD will be leveraging this process. They amended their wafer supply agreement with GloFo specifically for IBM's 14nm and 7nm SOI FinFET nodes.

Why are you so dedicated to doing nothing but spreading disinfo and shitposting? Are you really that much of a mentally defective fanboy?

>Hurr AMD is bad
>durrr everything around AMD is bad
>uuuuuuuuurf if AMD is using IBM's process then it must be bad too
>huffffrrrrr by extension IBM is bad

Seriously, stay in Sup Forums where you belong. All you're doing is behaving like a child day in and day out.

Wow you're optimistic. I would be surprised if this cost less than 10k.

It wouldn't be priced out of range of intel's then competing parts.
Current xbox hueg Xeon is the E7 8890v4 with 24 cores and 48 threads. MSRP for that monster is $7,174.
They'll likely bump their top end up to 30 cores over the next two generations as they've been doing consistently for years now. I don't think price on the top binned SKUs will increase much beyond that though.

>48 cores
kill this fucking meme right fucking now

Intel has no competing parts right now mate.
The 24 core Xeon is 7k as you said, and that's the best intel has and it's already extremely expensive to fab. AMD is promising to deliver a chip that has like twice the transistors if not more, it's not gonna be cheap to fab. Unless they have like tons of smaller dies.

Which they might I guess.

The real question is if 7nm is even stable.

If 11nm is shit how is 7nm supposed to work?

The keyword being "then" as in a couple years from now.
30 cores vs 48 cores likely would still be competitive in a number of metrics, intel's Core i arch is substantially larger than AMD's Zen core arch. Intel produces monstrously large cores with ungodly large FPUs and they have years of pedigree in doing so. AMD doesn't so I don't see them trying to tackle intel there.

AMD will probably do pretty well in the enterprise market in coming years. That being said I don't see intel being left in the dust in any way.

What exactly do you mean by stable?

The issue with CMOS scaling is always the short channel effect. A shorter channel lends to gates needing less current to switch, but its easier for current to leak out of them. The interim solution to that was to move away from planar gates and start using vertical Fins. They facilitate decreasing channel lengths while effective gate surface area actually increases, and leakage plummets.
Though we get to a point where you can't realistically create a fin tall enough to keep an electron from shooting out of the channel, silicon no matter how you dope it just won't keep the fuckers in place. So SiGe is used. Completely different set of insulating and charge carrying properties depending on how its employed. Physically smaller channel, lower drive current, lower leakage, better ultimate channel control, and the current stays where it should.
In that sense its as stable as anything else. Its not like its going to break down after a certain time limit(within reason)

SiGe has actually been widely used for years in mobile chips, just not as channel material.

No. 30 cores would not be competitive with 48 cores in no metric other than power consumption unless the 30 core processor had a huge clock advantage. Which is unlikely.

AMD Zen cores are going to be huge compared to Bulldozer cores. It's no longer gonna be a tiny Bulldozer core vs a huge Intel core. It's a huge AMD core vs a huge Intel core.

No matter how you slice it, a fucking 36 thread deficit cannot be overcome. That is a HUGE number. Most processors don't even have that many threads today.

>AMD is promising to deliver a chip 10 years from now that will beat what Intel offers today

>I dont know what IPC is

>That is a HUGE number. Most processors don't even have that many threads today.
My workstation has more today and it is almost 3 years old

You'll have to see how well AMD's 32 core Opterons compare to intel's 24 core Xeons.
A weak FPU per core is easy to beat out in throughput. If the workload is crunching 256bit Vectors then AMD is at a massive disadvantage they're not soon to overcome. Zen is still lacking 256bit data paths in the FPU, they'll handle 256bit ops but at half the rate of 128bit ops. Intel's cores don't have this handicap.

Yea well maybe if AMD cocks it up as usual, but that's a huge fucking deficit. I'm pretty sure AMD integer performance will be decent, so even if they did cock up the fpu that'd still be tough competition.

Not that I think Intel will just try to compete with 48 core processors by building 30 core processors. They can build more cores just the same.

I happen to know you're the exact same autist who accuses everyone else of being the exact same autist who did something you didn't like in every AMD thread.

Except I'm always right.
The exact same shitposts in every single thread he shows up in.

>hurr IBM can't do anything innovate
>hurr they'll just license stolen tech
>hurrrrrrrrrr

This childish shitposting ruins the board. There are literally only 1-2 posters responsible for this. The same little faggots who spam the X IS FINISHED HOUSEFIRES AMDPOORFAGS bullshit.

>only 1 or 2 people could possibly be responsible for posting insipid memes on Sup Forums

sounds like you've forgotten which site you're on.

Its not people off handedly spouting meme nonsense thats the issue. Its this couple die hard autists that show up in every thread to derail it. The exact same tone, the exact same phrases used, the exact same words choices, even the same common typos. Its the same couple of people behind the overwhelming majority of it.

I don't spread fanboy bullshit when I'm in a thread discussing intel's latest Core i arch, or the latest ARM SoC. These children however come to any thread tangibly related to AMD and post the same ill informed crap without fail.
Brand fanboyism belongs in Sup Forums.

>i dont like shitposting
>>>/reddit/

>not only will it be more economically viable (in terms of cost effectiveness and $ per transistor) than the 14nm implementation of Zen but it will also be much more power efficient.
This is damage control for the upcoming failure of Zen isn't it?

Its guesswork on part of the author for the most part
SOI does decrease costs to a degree by requiring fewer critical masks than bulk FinFET processes, but the starting cost is higher from the more expensive wafer, and ultimately if the yields aren't there then all of it means nothing. The trend had been cost per transistor leveling out post 28nm, as in it hadn't been decreasing which was troublesome for the industry as a whole. Speculating on it actually increasing in future nodes is the reason why everyone is holding on to their 14nm/16nm nodes for so long.

It goes without saying that it'll be lower power, but cost per transistor is still an unknown factor to everyone outside of IBM/GloFo right now.

MOAR CORES

MOAR NIGGAHURTS

MOAR HOUSEFIRES

Are you people retarded? This is a server/workstation/supercomputer CPU. In these use cases MOAR CORES is king.

We know but AMD use moar core for the lack of performance. It's a meme

mfw after 5years those will cost under $50
just like those 20core cpus