Reminder that a Zen core is currently more advanced than a Skylake

>>58839488
where that 4 core nomenclature comes from ?
i thought they had only 6-8 cores for now

How much did AMD inc. pay you for this ad?

File: zen-die.png (387KB, 750x412px)

387KB, 750x412px

>>58839503
Zen currently only comes in complexes of 4 cores, there are two of these 4 core complexes on the Summit Ridge CPU.

>>58839488
>equal performance
Slow down there. All the rest are true, but from the looks of things Integer performance will be around the same level as Broadwell, and Floating point performance around the same level as Ivy bridge, which isn't a huge difference, but still not "equal".

>>58839523
ok thanks

>>58839504
http://www.eetimes.com/document.asp?doc_id=1331317&page_number=2

Intel engineer admits Zen core is competitive.
The numbers in the chart just confirm it.
If your company admits it, it's time to face the music.

I can't wait for actual benchmarks to come out so I can enjoy another year's worth of AMD tears.

>>58839595
>I can't wait for actual Netburst benchmarks to come out so I can enjoy another year's worth of AMD tears.
>I can't wait for actual IA64 benchmarks to come out so I can enjoy another year's worth of AMD tears.
>I can't wait for actual iAPX benchmarks to come out so I can enjoy another year's worth of AMD tears.
>I can't wait for actual FDIV fix stepping to come out so I can enjoy another year's worth of AMD tears.

Intel with 3 spectacular architecture fuckups, and a mass recall due to errata error, AMD had a single CPU fuckup in 45 years.

Clearly someone has a better history here, but too bad you're too underage to know any of this.

>>58839488
benefits of a full on 5+ year R&D effort

the thing is, intel could also build a new uarch from scratch, and their existing technology would carry them there, unlike bulldozer which almost put AMD under

>>58839488
>missed the boat at $1
>missed the boat at $7
not gonna miss the boat at $13.
selling all my dogecoin from 2014 into this shit.

>>58839563
Sir sir sir my friend my friend please delete this.

>>58839665
Intel is building a new arch, but it's not coming out before 2021.
It started looking for engineers a month ago.

>>58839665
Zen is just the starting point for AMD, they're going to continue to evolve it. At the Financial Analyst Day in May 2015 Su stated that work on Zen+ was already well underway. How much of an uplift do you think AMD's teams can pull out in an additional 3 years on top of Zen which is already trading shots with Broadwell?

Intel's family of arch here is almost dead ended. They have no big performance uplifts coming. Cannonlake, Ice Lake, Tiger Lake, all 10nm parts with tertiary features being their selling points, not increased IPC. They won't have a truly new arch until after all of this in 2021-2022.

>>58839712
>not hiring the shit wrecker for 6 million a year.
>settle for pajeet instead

>>58839544
If it can clock higher than Broadwell(what can't lol) while using less power it's clearly faster.
If we're looking at pure IPC then yeah, it's Broadwell level.
Now the thing is how do you measure FP? Because 512bit FP ops aren't really gonna be useful to anyone on /g/, to some 15% of the server market? Yeah.
So you could say its maximum throughput is lower, but it won't affect most workloads.

>>58839736
When I read Keller's interview he seemed like a guy to go for challenge, there's not much challenge at Intel since they wouldn't allow him total control like AMD did.
So he's hopping from company to company making killer cores that WILL let him have control.

>>58839661
>AMD plays it safe
m-muh fuckups
meanwhile Intel is driving improvements over technology not like poodozer

>>58839755
even video encoding doesn't use FP.
FP should be offload to gpu.

>>58839768
>now at tesla making killer cars

>>58839771
More like Intel is incompetent and is still using a bolted P6 core, that's the definition of playing safe.

Every large architectural change AMD did has succeeded besides Bulldozer, every Intel attempt at matching that success failed.
Too bad OEM bucks didn't fail, you just need good lawyers for that, not engineering talent.

>>58839785
2nd pass uses FP I believe, it's mixed if anything.

Anyhow yo can't offload FP yet because branchy code exists and GPUs can't deal with that, they're good for doing dumb and simple..

>>58839488
Weird, AMD's L2 is 8 way associative, Intel's is 4 way, meaning AMD uses more die area for faster performance, but it's also smaller per KB, awfully strange considering the SRAM numbers in at the bottom.
Everything else Intel wins in pure lithography, by a good 15-20%

Good process, but it's also a good thing a process doesn't make or break a product, you can fill out the lacking stuff with a good architecture.
Seems AMD did exactly that.

>>58839822
x265 has avx, so it has FP.
x264 last i heard was mostly integer.

>>58839712
>yfw intel next architecture features long pipes and high frequency.

File: single-socket-holocaust.jpg (28KB, 550x550px) Image search: [Google]

28KB, 550x550px

>>58839989
mage longer bibeline XDDDD

>>58839989
Would be good only if they're using some exotic material that can reach over 7GHz and not silicon that will burn down their fabs and summon Gehenna

File: 1426467217270.jpg (40KB, 435x512px) Image search: [Google]

40KB, 435x512px

>>58840010
>Jayhawk was going to be 50+ stages

>>58840020
I doubt intel will be using an entirely different substrate sub 10nm. Everything will still have a silicon starting wafer for a long time to come.

>>58840075
LONGER

File: 1422561205843.png (7KB, 645x773px) Image search: [Google]

7KB, 645x773px

>>58840075
that would have been the first sustainable fusion reactor
>tfw not building a tokamak around your Intel cpu

>>58839488
>Equal performance

But that's wrong you mongoloid.

5% lower IPC while being their budget 4c CPUs are a whole GHz behind stock Kaby. Ryzen will compete against broadwell-e but in terms of budget gaming performance it's doa. I still can't wrap my head around their ultra high end 8c/16t actually having higher clocks than budget r5 and r3

>>58840075
Clearly these heathens had no qualms about summoning Satan to the Mortal Realm.

>>58840127
>5% lower IPC while being their budget 4c CPUs are a whole GHz behind stock Kaby. Ryzen will compete against broadwell-e but in terms of budget gaming performance it's doa. I still can't wrap my head around their ultra high end 8c/16t actually having higher clocks than budget r5 and r3

That wccftech rumor models going from 1900X to 1100? That's just made up anon, no sources were cited.

>>58839806
>Every large architectural change AMD did has succeeded besides Bulldozer

https://arstechnica.com/business/2013/04/the-rise-and-fall-of-amd-how-an-underdog-stuck-it-to-intel/2/

>To stay competitive, AMD created its first original x86 design, a Pentium-equivalent called the K5. This CPU was roughly comparable to the Pentium's performance, but it came to market in 1996, a year after the Pentium Pro.

>A tiny Silicon Valley company named NexGen, meanwhile, was doing what AMD couldn't: taking on Intel with a relatively miniscule team of just 60 engineers and a little funding from IBM. The company was headed by Atiq Raza, a brilliant Pakistani engineer educated in the United Kingdom and the United States, who had arrived in the US in 1979 with only $3,000 in cash to his name. (By 2009, Raza put his Palo Alto estate up for sale at $24 million; he ended up not selling it.) Raza wanted to take on Intel, too, but he knew that he couldn't do it at NexGen's current size.

tl;dr AMD fucked up so hard they ended up buying another company to save their ass

>>58840127
That's easy. Every Ryzen is a 8c/16t but some of them failed tdp/voltage/frequency/smt/core targets and were recycled.
The 4c low-clocked ones are the biggest failures but still good for low end.

>>58839488
will get this or whatever comes after it when i upgrade from intel i5 2500k. i dont want that encrypted binary shit on my devices

File: IMG_20170206_074008.jpg (38KB, 720x400px) Image search: [Google]

38KB, 720x400px

>yfw all unrest Intel shills are spreading the less than 3.4ghz myth

File: 1267467696350.jpg (63KB, 500x388px) Image search: [Google]

63KB, 500x388px

>>58840425

they're [RATTLED]

I don't play games. Why is 3.4ghz bad for an eight core processor?

>>58841029
Its not.

>>58841029
It's not and it was already confirmed to have SKUs with much higher clocks.

>>58841029
It's not. Intel's 4 cores can clock higher so they've reverted to the "4GHZ is higher than 3.4GHZ!" mentality. Basically, Pentium 4 all over again.

>>58839488
Zeppelin die confirmed 200mm2 or slightly below, a smaller die than Polaris 10.
AMD is going to be just shitting these things out, high yields. What an exciting time.

>>58841129
If AMD at launch has a 3.6ghz/4ghz chip, its bodes very well. The original FX 8150 launched with a 3.6ghz base clock, and that was on a process specifically geared towards high clocks. The later FX 8350 got bumped up to 4ghz base, and I think Ryzen could likely do the same without increasing TDP. Higher base clocks at a given power level means dramatically improved clock scaling as power decreases. If they managed 3.8ghz at 10w~ then they could hit maybe 3ghz at 5w.

>>58841347
Hey how in the fuck is the die around 200mm2 when the core+caches are 88mm2?

How can the uncore be 112mm2 with a dual channel IMC? Do those GMI/Infinity fabric links use that much space?

File: Zen_Summit_Ridge_First.jpg (73KB, 369x676px) Image search: [Google]

73KB, 369x676px

>>58842173
See for yourself.
The CCXs are not placed directly next to each other, a die is usually not laid out to be as dense as possible. Thermal considerations tend to be of pretty high importance.

>>58842253
That seems like a lot of empty silicon but it could be interconnects on some lower metal layer.

>>58839488
It's means nothing when you don't know the number of cycles each operation takes, the access time to memory, and how well they prevent cache misses.

>>58842276
Theres never any wasted area, everything is laid out with a purpose.

>>58842500
Table of instruction throughput and latency is already out there floating around.

>>58839523
Hol up... So is you sayin da 8 core Ryzen finna have 16mb ah dat L3 cache n sheet? I think you be talking some bull sheet mang.

>>58842718
It also has 512kb L2@core.

>>58839488
BUT WHERE THE FUCK IS IT MY FX 8 CORE NEEDS TO BE REPLACED

>>58840170
Didn't raza leave?

Also hasn't every AMD CPU since the phenoms been a flop?

>>58842882
>Also hasn't every AMD CPU since the phenoms been a flop?
So you mean the Piledriver family?

Trinity was shit.
Everything after that brought notable improvements and fixes to the arch.

The arch was flawed to begin with.

so yes, the 1 product line of CPUs that AMD has had for 5 years, was bad.

>>58840010
Fucking loved my Prescott Pentium 4. Saved me $$$ on heating!

>>58842799

Can I have it?

>>58839661
>netburst
>intel viiv
Oh those were some wild years man.

File: 1486382593903.jpg (417KB, 2500x2500px)

417KB, 2500x2500px

>>58840010
>filename

10/10

>>58839799
the guy who died was retarded.

do Zens come with an iGPU?

>>58845597
The Ryzen line is CPU only.
The APUs come later in the year, the platform codename is called Raven Ridge.

>>58845597
Those come later in the year and are all 4 cores

>>58839488
It'd better be, because otherwise this'll be AMD's last stand.
They won't be able to recover if Zen ends up being a flop.

>>58839488
Larger cache in a relatively smaller space (factoring process size) means it's more dense, which means it is slower.