Who's waiting for AMD ZEN here?

File: intel.jpg (1MB, 1845x1923px) Image search: [Google]

1MB, 1845x1923px

>>51376996
I'll buy one if it matches current Intel performance and has moar corez. Having an Intel CPU makes me feel dirty.

>inb4 JIDF

>>51376996
whats the x86 about

>>51376996
Everyone is.
Summit Ridge and its server countpart Opteron line are AMD's battle horn. The official sounding that they are back in the fight, they are targeting enterprise, they are targeting high performance, and performance per watt.
I'm holding out for Raven Ridge in Q1 2017 to see how they diversify it from their binned Summit Ridge chips, and entry level GPUs. That'll give real insight into how Su sees the direction of the company in regards to the consumer market.

That being said, Zen is not the arch to be excited about. The Zen+ core arch is the one that will really change things for AMD.

>>51376996
I am.

I need to replace my n54l and my current desktop so Ill wait.

File: AMD-Zen-Quad-Core-Unit-Block-Diagram.jpg (97KB, 961x540px) Image search: [Google]

97KB, 961x540px

enjoy your gimped cores that are called 4 core 'units' now AMDpoorfags

>that shoddy writing

wccftech detected

>>51377222
>still reposting fake slides from a german forum after they were proven to be shitty photoshops numerous times

Nice shitposting, child.

>>51377088
Its an instruction set, its basically amd saying it wont be ARM

File: 1446807674683.png (3KB, 364x138px) Image search: [Google]

3KB, 364x138px

>>51377248

you'll be denying it until the very end, eh bud?

>>51377279
http://juanrga.com/en/the-fake-zen-slides.html

100% proven to be fake.

More so some recent patch notes detailed the core width of the Zen arch. Those notes prove that Zen does not use FMACs.

ZNVer1 is a 10 wide core, 4 ALUs, 2 AGUs(load/store), and a 4 port non fused FPU.


Shitposting is against global rules, btw.

>SMT for high throughput
LESS COARS

>>51377387
Equal amount of cores, wider cores.

>>51377400
1 Bulldozer module = 1 zen core

I don't think they will do a 8 zen cores, 16 threads model for a regular consumer cpu aside from servers. Maybe in some sort of super expensive extreme edition

>>51376996
Me. I want a mean apu with ddr4 in a nice miniature package.

>>51377341
>shitposting
>links to a post on juanrga

really

he's a well known shitposter on the semiaccurate & realworldtech forums

I know I am. Looking to upgrade from my a10-5800k now that I have an actual GPU. Gonna have to change boards to upgrade anyway, might as well wait for zen.

>>51377458
1 module is not equal to 1 Zen core, resources are not utilized the same way.
The Bulldozer arch has two integer cores in the hopes that each seat of two ALUs per core will sustain higher throughput in a well threaded workloads than a single core with 3 or 4 ALUs. It was about trying to maximize compute performance per mm2, a paradigm that while valid, can be achieved in numerous other ways.

Summit Ridge is capped off by an 8 core part with a 95w TDP. This is certain. Its one of the first things Sweclockers ever posted, and they've been right on the money about everything else, including the Summit Ridge codename before it was ever used elsewhere.

>>51377552
All he did was make a blog post showing the fake slides run through a site that checks if images have been edited.
You're still just shitposting like an underage /b/tard, and its still against the rules.

>>51376996
Are you sure AMD isn't jewish as fuck to?

>>51377573
>completely ignoring the fact I pointed out that he's constantly posting shit on various cpu arch forums

he's got REAL industry engineers calling out his shit and yet he still persists

the slides are fake - but don't take anything else seriously from this guy

>>51377573
>1 module is not equal to 1 Zen core, resources are not utilized the same way.
In terms of FP count, it is.

>Summit Ridge is capped off by an 8 core part
It's a server CPU.

>>51377341
Please elaborate on this. You sound like you know what your talking about. Also what are your impressions on that chip design. (4 + 2 +4)

Will it be in laptops?

Could be more power for money since AMD.

>>51377655
"2 AGUs + 4 ALUs would be rather disappointing and also at a severe disadvantage for HPC to Intel."

>>51376996

I'm interested but more for potential power efficiency and a successor to the Athlon AM1 series. I want to build an always-on home server that scrimps on power but doesn't scrimp on features, like Intel's offerings in the space do. The AM1 processors have the features but not the efficiency and I'm hoping Zen cores will change that.

In the desktop space, I have aged out of needing desktop upgrades because I'm still playing older games.

>>51376996
Ye boi.

Gonna pick up the cheapie quad-core model Zen chip and an R9-490 with 8GB of RAM.

Can't wait to play XV in 4K.

>Zen + GPU + HBM on a single package

fugg

It's not going to be a match for Kaby Lake

AMD is ded bruh

>>51377600
>shitposting child with the IQ of a bar of soap

Nothing else he says matters, you are not addressing the facts of the matter, you are deflecting like a shiposting 12 year old faggot child. You posted fake, photoshopped slides. I stated they were fake, and provided a valid source.

You attacking the source for an unrelated issue has no bearing on this whatsoever.

>>51377653
They are not even remotely similar in their FPUs.
The Bulldozer family uses 2 128bit FMACs, and one or two MMX units. Later revisions like Steamroller and Excavator only have one MMX.

Zen has a 4 port non fused FPU, it does not use FMACs, and it has wider data paths. They are not even remotely similar. Not to mention the entire front end aspects of the FlexFPU.

Summit Ridge is as much a "server CPU" is as intel's i7E line. This statement is utterly meaningless.

>>51377655
Elaborate how? Thats a seriously open ended request.
Each Zen core is at least physically as wide as an entire Bulldozer family module, and they are not reducing core counts in their consumer chips. More execution units over all tends to be a good thing.

>>51377702
David wrote this entirely out of context, and didn't explain himself for some time.
Core per core with equal clocks no one expects AMD's Zen to compete against Haswell and Skylake directly. Intel does extremely well in how they handle and tag memory ops within their core. They've spent and awful long time working on their implementation of SMT.
AMD however have touted utterly disruptive memory bandwidth and latency, so things won't be so clear cut.

>>51377787
i'm not even the guy who posted the slides senpai

>>51377779
what the fuck is Kaby Lake even supposed to do besides help stall for Canon Lake after the 10nm schedule slip?

>>51376996
meh, amd has tried to revolutionize the cpu time and again and mostly failed. I'm beginning to think x64 was a fluke

>>51377796
Thats exactly what its supposed to do.
They had to delay their 10nm node, they undoubtedly would have had even worse issues than they did with their 14nm node. Kaby Lake is a stop gap with very minor arch changes.

>>51377817
What?

History repeats itself.

First intel, then amd, then intel, now...

I fully expect zen to push ahead of the (at the time) current intel line. Its a completely new arch, so its just logical to assume that it will be superior. Being that AMD also had great price points for its cpus, it stands to believe that such CPUs will also be very, very competitively priced.

Now, intels response will be interesting...

>>51377852
>Its a completely new arch, so its just logical to assume that it will be superior.

>what is Netburst
>what is Bulldozer

2 new cleansheet archs for you right there

Has there been any word on the SKUs of Zen?

Are they gonna do a 4, 6 and 8 core lineup? Or maybe not 8 cores, since the cores are bigger this time.

>>51377872
>Has there been any word on the SKUs of Zen?
Not yet. Basically just "wait until later."

>Are they gonna do a 4, 6 and 8 core lineup?
They're not going to throw away usable dies, they don't have the luxury. They'll have one 8 core base die, and they'll sell various binnings of it. Lower clocks, less cores, etc. Just as they did with all of their previous Bulldozer family chips.

>>51377869

Again, you just prove my point of history being cyclical.

Good, mistake, good, from both company's.

My comment is meant to be taken in context of the history. (new arch after bad arch). My final point was a bit nebulous, sorry.

>>51377787
I didn't say their fp architecture was similar. A bulldozer module and a zen core has same fp scheduler count and it is.

>Summit Ridge is as much a "server CPU"
Su is clear on their position, no more being underdog. If they are making an 8 zen core cpu it will be a server one, at least for quite a while into Zen's life.

>>51377817
>>51377852
the thing is that superscalar/OoO x86 microarchitectures have been very good for a long time for general purpose computation, and everything Intel and AMD has tried to do to "revolutionize" shit just ends up being optimization for rather specific computation cases.

> long ass hyberbibelines are great for 5 niggerhertz low-branching code!
> SMT is great for hiding memory stalls! there will never be more cache contention!
> a split front end will be great for everyone who always runs identical threads everywhere! nobody will ever have i-cache contention or bottlenecked fetching/decoding!

most coders can't architect their software in ways conducing to low-level optimizations on modern mem-latency limited CPUs, so assuming anything but the bare minimum about binary quality will see performance tank for 99% of software with architecture "optimizations".

big cache/tlb/btb and short pipelines seem like unnecessary extravagances until you realize that existing software needs every crutch it can manage.

>>51378026
>conducive

fuck i don't even know

>>51378004
>. A bulldozer module and a zen core has same fp scheduler count and it is.

The Bulldozer family uses the FlexFPU. The scheduler itself only functions the way it does because of the arrangement of the integer cores. It utilizes an implementation of SMT to allow it to partition instructions to either of its FMACs.

The Zen core does not have this arrangement, and it does not utilize the FlexFPU or its scheduler.

>>51378062
I also didn't say they have the same structure for scheduler. I said count.

lol @ all these armchair EE's

you guys sound so clueless

>>51378087
>scheduler count

Time to stop posting, kid.
You clearly are just saying words and have no idea what a scheduler does in relation to a fetch, dispatch, and decoder.

>>51378116
You are trying too hard. Don't feign ignorance what is what.

Please AMD, give me a reason to upgrade my 2500k.

>>51378107
>you guys sound so clueless
and with that the shit flinging begins

>>51377279
You do know AMD is using smt instead of cmt for zen though, right?

Fucking dumbass.

>>51377032
fuck. ive been searching that pic for ages, thanks senpai

>>51378227

this. i need something with higher than 2500k ipc and 8 cores.

>>51377124
>Zen is not the arch to be excited about
amd shills backpedalling already

I am Inte i7 950 over here

When will threading/scheduling models be modified to accommodate SMT? Right now, at least on Windows, you basically have to hope that your thread affinity hints make the scheduler stagger your threads across all the physical cores before using multiple logical cores

>>51376996
Yes.
why not?
still waiting for leaks on AM4 motherboards
why do manufacturers not allow consumer input?
Then i can compere to skylake to zen.
skylake has ugly motherboards so its going to be interesting

>tfw need a new computer now
I can't wait for "october 2016", and making a cheap right in the meantime feels like a waste of money

>>51377124
the Raven APUs are hopefully in a position to completely negate low tier GPUs entirely. This wouldn't surprise me because it would seem to coincide with AMD's release of only top level/high level dedicated chips (Raja is stated saying Arctic will only have two GPU chips which may turn out to be only two full chips and two binnings)

Hell, the only thing stopping current APUs from dominating in graphics/compute performance is sharing the DDR3 bus with the CPU. Their on-chip 7750-plus-hawaii-improvements (the a10-7870k) would be about 20% slower than the real card, and only because of TDP limits, if just GDDR5 could be integrated.

I imagine that a 1024 shader, 2GB HBM(2?) Arctic Islands based 16nm unit could perform as well as or better than a 7870 with 65 watts of power.
Paired with an overclockable CPU as powerful as the current i3's, this could undercut every single GPU in the $90-120 range as of expected 2017 prices (7870, 750Ti/950 tier)

File: power_maximum.gif (74KB, 400x1227px) Image search: [Google]

74KB, 400x1227px

>>51384636
If the 8 CU in Kaveri had adequate bandwidth it would perform near identically to the reference Radeon 7750. The reference card has a core clock of 800mhz, flagship Kaveri's IGP is clocked at 720mhz and its newer arch. TDP limitation is not a factor in the equation here. The reference card with 1GB of GDDR5 has a peak draw of just 43w~
Newer power tune alone would bring that way down, which was implemented among other things in Carrizo.

Raven Ridge could easily fit 20CU on die inside of a 95w TDP with no units binned off. At 14nm they could accomplish this along with a quad core CPU, and die size would still be right around 100mm2, possibly less.

ARM simply, i use a few shitty servers are they are just doing fine. Same with dekstop, jewchrome, android.
Most distro and soft easly compile to arm. And ofcore battery life is amazing

>>51384711
>570
>310w maximum

>7850 (570 performance competitor)
>140w maximum

Why does AMD always seem to capitalize on die shrinks so very well?

>>51376996
>2011
>WHO HYPE FOR BULLDOZER! AMD IS ABOUT TO SHOW INTEL WHO'S BOSS!
>it's shit

>2015
>WHO HYPE FOR ZEN? AMD IS ABOUT TO SHOW INTEL WHO'S BOSS!

like cockwork. Like fucking pottery. It's gonna be the same shit again.

>>51377341
>juanrga
lmao

>>51384815
Bulldozer was a design that AMD had been sitting on for years. It never got the funding or time necessary to turn it into what was envisioned. Many units were hacked together and recycled to save time in bringing it to market. AMD's management was incredibly unstable before it launched, and they had no idea what to do post shakeup when all of their design talent was walking out the door.
No performance characteristics were shown before it was released, and it had made major regressions in certain areas.

Zen is a brand new core arch, uses an entirely new cache, new memory controller, new interconnect fabric. We know IPC on average is 40% than Excavator per core. That is tangible. We already know it will at least provide a certain level of performance.
The company is internally stable, was financially restructured, and Zen is part of a long term cohesive plan instead of something rushed to market in a panic.

The two are not comparable.

>>51384815
New buisness direction, we will ditch the MOAR COAR philosophy and follow a new vision, a unified vision. our new motto:

>ONE VISION, ONE PURPOSE, ONE CORE TO RULE THEM ALL.

Ladies and gentlemen I present to you neuvo Zen, this product is destined to break all bounds at a blazing fast 32Ghz clock speed by channeling the power of 8 cores it brings the CPU as a whole into a state of equilibrium.

You will not need to worry about your power bill, the chip at its core is a fusion reactor supplying its own power, it also functions as a heat and light source that shines as bright as a star in the comfort of your basement.

Perfection

>>51378107
>EE's
You mean CE's you fucking nigger.

In the mean time
>implying this won't be delayed for too long in the same way that the R9 300 series was
>implying this won't be as underwhelming as the R9 300 series when it actually does come out

>"But AMD has made a division between Radeon GPUs and it's CPU division!"
That won't matter. 40% better IPC than Bulldozer puts it at par with Haslel, not exactly game changing. AMD will only make a victory if they release the CPUs cheaply, which they'll overprice, given the pricing on things like the 8370e, and the R9 300 series being literally 200 series cards with a bigger price tag and some factory tweaking.
That being said though, making a cheap product isn't what AMD, or the market, needs. The market needs AMD to actually make worthwhile products for the premium/enthusiast end of things.

>>51378227
>>51380099
another 2500kfag here

since intel are incapable of producing any worthwhile upgrade my hopes rest on amd

>2015
>still upgrading CPUs

>>51385037
back in my day there was a valid reason to

>>51385052
>you will never live GHz wars again
Nothing is giving excitement of ATi 9800 and Athlon 64 anymore

I doubt that the Zen CPUs can compete against Intel.

Intel pretty much already have the possibility to trash AMD, but the market doesn't demand for it, so they hold back.

>>51387236
If they can ast least get similar performance for a few bucks less maybe they can gert their foot in the door again or at least make jewtel drop their prices a bit. Even Intel's lowend products are pretty pricey for budget builds.

I for one am looking forward to what lowend ecc options AMD will provide with the new lineup for my future fileserver. I'm well aware that that is a massive niche market so I don't have a lot of hope.

>>51376996

im waiting to see how it performs before i upgrade.

but i think intel cpus and especially chipsets are so far ahead right now, i dont even think amd can catch up, even if the performance is equal, the intel chipset has everything i need.

But i am hoping amd will bring out the big fucking guns and actually make me want their shit.

i do love AMD gpus, so im hoping if its a good eco system, that i can get amd zen + arctic islands.

have the entire system be team RED

>>51385052
>tfw upgrading from P3 to Athlon XP in 2005
>tfw upgrading from some single core crap to Core2DUO E8500
>tfw compiling Linux kernel on a 4+ core system after doing it for years on a dualcore

File: 1447420360569.jpg (109KB, 601x592px) Image search: [Google]

109KB, 601x592px

>>51384815
>bulldozer
>shit
You fucking what m8

File: 1447663214831.jpg (92KB, 523x440px) Image search: [Google]

92KB, 523x440px

>>51387623
These processors are way too underrated.

>>51387623
>>51387660
Who cares about gaming benchmarks

>>51387886

everyone.

but those charts are kind of stupid because sure the amd cpus perform well, but they use twice the wattage of the intel cpus.

>>51377245
(You)

>>51377561
(You)

>>51377787
(You)

>>51387623
(You)

>>51388564
Anon, that's inefficient. Observe:

>>51377245
>>51377561
>>51377787
>>51387623

(You)


See?

>>51376996
I'm waiting for anything that will let me upgrade from my 2500K @ 4.7ghz. Pretty soon I'll have had this thing for 5 years.

>>51388881
(You)

Thanks.

>>51376996
>x86

????????

>>51388943
Again, not as efficient as it could be. A more correct response is "Thank (You)".

>>51388958
Intel cut off AMD's license to x64 so they have to rely on 32 bit x86 again.

>>51388940
Do you really need to upgrade though?

>>51389391
but amd introduced x64 first

But, isn't this just more APUs?
Wasn't AMD dropping the high-end cpu market after the bulldozer failure?

>>51377796
Kaby Lake brings ~10% IPC improvement over Skylake, adds 4 more PCIE3 lanes to the 200-series chipset, adds native 5k resolution support, adds support for that new Intel/Micron 3D XPoint SSD-DIMM, and hardware decode support for newer audio and video formats.

>>51389447
AMD is making regular CPUs as well, the highest Zen chips are supposed to be enthusiast level.

>>51376996
>x86
Does this mean I can't use a 64 bit OS?

>>51387623
>>51387660
What's missing from these pictures is a recent i3 BTFOing anything AMD has to offer.

>>51389598
Yes AMD is going backwards in time and they're going to release 16bit CPUs only from now on.

>>51389504
I can believe all those points except for the 10% IPC bump.
Please provide sauce on this.

My expectation is that it's best case factoring in some special purpose acceleration instructions while going downhill with a tailwind.
Either that, or Skylake was the real punt, and Intel wasn't really stalling for the 10nm shift after all.

>>51389622
>Please provide sauce on this
Historically, Intel was aiming for a 5-10% IPC increase per generation, with wildly varying results.
But Kaby Lake is more focused on power efficiency over pure IPC improvement, so it might be more like a 5% IPC increase at a lower TDP than Skylake. The unlocked SKU will probably have the same TDP, thus a much higher IPC improvement over the rest of the Kaby Lakes.

>>51389598
It means it uses a x86 architecture. Should've called it x86_64.

>>51389654
so basically you pulled this out of your ass?

>unlocked SKU will probably have the same TDP, thus a much higher IPC improvement over the rest of the Kaby Lakes.

I can't even begin to understand what you're thinking if you thing watt budgets can help you make accurate inferences about internal architecture/IPC.

>>51389675
They don't have to, AMD hasn't made an x86-32 CPU in a decade. Why would someone assume they would be making a 32bit processor now?

>>51389744
No, that's literally what Intel stated for their design goals with Kaby Lake.
Skylake was the same thing over Broadwell (higher IPC at lower TDP) and Broadwell was the same over Haswell (higher IPC at lower TDP).
It's been like this for the past three years. Compare a Skylake i3-6300 with an i3-4160 and you'll see a lower TDP with better IPC/single-threaded performance/multi-threaded performance/etc. Half that difference and you'll et Kaby Lake's expected improvements, i.e. it's still moving at a snail's pace.

Fucking AMDretards

>>51389041
Thank (You).

File: Generational CPU IPC_575px.png (38KB, 678x451px) Image search: [Google]

38KB, 678x451px

>>51389809
nope, just a haswell user, m80.

skylake is dog shit, and you're shill-level retarded if you think a last-second asspull half-generation is going to be great somehow.

>>51389979
I never said Kaby Lake was going to be great. I'm waiting for Cannon Lake to upgrade. Kaby Lake is just going to primarily add more features and some improvements over Skylake, but not enough for me to upgrade either.
Nowadays a 10% IPC improvement really doesn't yield all that much. We should be demanding at least 15% improvements, but that's supposedly what the 10nm shrink is going to do.

>but our die shrink is totally going to make your CPU outdated!
But where have we heard that before, Intel?
Oh right, the last four generations.

>>51390017
Moving to a smaller process node does not affect IPC in any way.

>>51390400
Wow, aren't you Captain Obvious

>>51387583
It's actually quite a shitty situation for AMD.

Intel will let AMD survive, but not enough so that they can ever again gain back major market shares.

AMD thus look elsewhere to find some profitable business e.g. gaming consoles and APUs/SoCs in Apple products and wherever.

They certainly must do that to survive and to be somehow profitable again.

And then there is Nvidia, who is binding all major game publishers on their side with shit like Gameworks etc.

The AMD Fury series has failed thus far, because the GTX980Ti is better in every way.
HBM made the Furies unecessary expensive too.

>>51390017
Skylake is basically Broadwell with:
> infinitesimal bump in power efficiency
> infinitesimal bump in IPC
> more iGPU
> same 4c/8t bullshit as last 6 or 7 years

Kaby Lake is going to be Skylake with chipset-type feature bumps.
Cannon Lake is just going to be a die shrink of Kaby Lake with probably a couple more new gfx/chipset features.

The biggest difference between a 2013 Haswell and a 2018 Cannon Lake will be modest power efficiency gains unless you're a poorfag with no discrete gpu.

You'll be paying ~$300 for a 4 core CPU running in the ~3 to ~4 GHz range, just like everyone has since the i7-860 (nehalem) in 2009.
Hell, even Core 2 Quads at ~2.5 GHz sold for only a little more in 2007.

Intel has been accomplishing very fucking little since AMD face-planted with Bulldozer.

>>51390699
If they released a Fiji card without HBM and priced it the same as the 390/390X, it would have been a godsend.
But no, Fiji's refinements can never be passed down to Hawaii in any shape or form because of HBM.

Why did AMD have to go after the big fish?

>>51390838
>more iGPU
It's the other way around. Skylake has a weaker iGPU than the unlocked Broadwells, since it only has the HD 5000s whereas Broadwell had Iris Pro 6000s.
One of Kaby Lake's SKUs will be like Broadlake with a mobile iGPU for the HTPC market.

>>51390853
HBM is the future. Thats why.
Smaller memory PHY, less die area wasted, higher bandwidth, lower power.

>>51390699
>>51390853

I think the hbm thing was more of a 'look what we can do' and will only really be relevant in future generations.

I do realise this makes me sound like the 'just w8 for next gen m8' guys but maybe they're playing the long game.

Also hbm could be interesting for their APUs and whatnot.

For what it's worth I'm happy with my r9 380.

>>51392557
it's probably more about spreading development costs across a broader span of time while they had the chance.

going with the 14/16nm node change and an entirely new mem architecture on the same model change would be a big and unnecessary risk.

maybe Nvidia can pull it off together with Pascal, but they still have every opportunity to fuck it up like they did with Fermi and the original dustbuster.

>>51392557
No consumer APU will use HBM. It's already been confirmed on their leaked Zen roadmap and reinforced by the fact that APUs will remain an entry/budget product.

>>51393809
>No consumer APU will use HBM. It's already been confirmed on their leaked Zen roadmap and reinforced by the fact that APUs will remain an entry/budget product.

This hasn't been confirmed anywhere. Whether or not Raven Ridge will have HBM on package is entirely up in the air, and APUs are not going to remain budget chips forever. Kaveri already showed the beginning of this. Llano and Trinity/Richland launched with an MSRP round $130~. The Kaveri A10 7850k launched with an MSRP of $173.
They want an IGP to outperform Iris Pro, and HBM is the only way to accomplish that.

They'll always have cheaper binned parts, but it is abundantly clear that APUs are targeting "mainstream" and not entry level or "essential" market segments.

>expected in October 2016
Looks like I'm sticking with my Phenom 2 for another year then.

If it can get close to the hype: I'll do a fresh build (5.5 years out of this one isn't bad).

>>51384982
>AMD should focus on enthusiast market
AMD needs money and a lot of it. They need to move large amounts of product and take back market share. High end sector is not the way to go about that.

>>51393855
>This hasn't been confirmed anywhere
My bad, I didn't realize that the Bristol Ridge APUs are still on Excavator/28nm.
But APUs will remain a budget option for two reasons
>iGPU will always be behind the vast majority of dedicated GPU cards
>there is not enough die space to make room for an 8-core processor with a full-fledged R7 iGPU
>even if there were, thermal management between a full-fledged processor and R7 GCN will be a major problem
>the only market that this sort of product makes sense to are the low-end budget market who likely will not purchase a mainstream or high-end card or the HTPC market
>HBM will be a marginal benefit over conventional DDR4 compared to its massive individual cost and production costs due to die changes
Outside of servers and mobile, I highly doubt APUs will move into the high-end desktop/gamer market because of its costs. HBM needs to be a fraction of its current price (I think SK Hynix said their HBM2 costs about 5 to 6 times as much as GDDR5, which is still more expensive than DDR4).

The only thing I can think of that might change this is Intel's RAM-SSD XPoint hybrid coming to market. That might push AMD to implement HBM for their high-end CPU for some semblance of performance parity.

Will this use AM3 socket?

I'm assblasted about having to buy a new mobo every 2-3 years thanks to Intel.

>>51394129
No, both pre-Zen and Zen CPU AND APUs will use the same AM4 socket

>>51394144
Well, unless I see some crazy deal on Black Friday I'll hold on my i3 2120 until next year.

>>51392557
AMD are playing the long game. With DX12/Vulkan they were able to force a sea change in the gaming segment to make their processors more valid. The fact that their current generation CPU's are going to be useful for gaming says a great deal, but if Zen can manage up to 16 threads at a time it makes those even better.

With the console wins AMD has basically forced most game devs to develops for GCN. While it hasn't payed off yet it's about to. When DX12/Vulkan come online fully it's AMD that will most benefit. Not just because the console ports, but because by getting major engine studios on their side the engines that most games are going to be made in will be geared to support AMD hardware.

Lastly, the server space is looking pretty bad for AMD, but only on the surface. With new data centers being built, and older ones needing an upgrade, 2016 is going to be a good year for AMD in the server market. This is especially true if Opterons are 16+ core beasts that can handle 32+ threads. Even better is the Server APU's. An APU with 8+ cores and a substantial number of GCN CU would be a beast in an HPC enviornment.

The physical distance of RAM and CPU is becoming an issue, but one which can be remedied by increased CPU cache memory.

Why dont they make chips with like 10 gigabyte L1 cache which is like 1 nanosecond away from the CPU itself?

>>51376996
With the way AMD have reported that Zen has met all expectations in lab testing, I am starting to get very moist about it.

Even though a little voice at the back of my mind keeps telling me it's going to be Bulldozer all over again.

File: ps4 chipworks.jpg (109KB, 640x618px) Image search: [Google]

109KB, 640x618px

>>51394045
Look at the PS4's APU.
A 348mm2 die, operates inside of a 100w power envelope.
8 Jaguar cores separated into two quad core clusters with a large interconnect between them.
20CU on die, just two of them binned off, operating at a 800mhz core clock.
256bit GDDR5 PHY.
Its a cheap 28nm HPP part.

The only issue this 28nm part has is the CPU cores not being able to clock high enough to deliver strong serial performance. Its not a particularly hot running chip, and there is no observable thermal throttling.
Directly scaled down to 14nm the die would only be 105mm2, and power consumption would be cut in half.

Without the large memory PHY, needless CPU cluster interconnect, and other misc units removed a quad core Raven Ridge APU could manage the same 20CU IGP, have a die under 100mm2, and stay well within a 95w TDP. Its newer arch, newer PowerTune, and a massively more efficient process node.
AMD needs their IGPs to outpace intel's Iris Pro graphics, its the only real performance lead they can maintain. Dual channel DDR4 doesn't provide enough bandwidth even for 8C, HBM is the only solution, and its likely that their top binned desktop and mobile APUs will have at least one module on package.

Get ready for Raven Ridge to be capped off by a $250 chip. Even by AMD's own marketing the top Kaveri SKUs are mainstream chips.

>>51394264
>The fact that their current generation CPU's are going to be useful for gaming says a great deal
Possibly not because DX12 and Vulkan relies entirely on how the developers implement the available hardware. 4 core/threaded processors will be the sweet spot because of the market size of computers running four core/threaded CPUs.
It also depends on how strong the cores are individually, since eight weak cores mean nothing if almost half the cores are left underutilized.

>>51377032
amazing effort put into that pic

>>51394368
Except the consoles use AMD processors too, thus developers need to use 8 cores to get the most out of the hardware.

>>51389830
ayyy

>>51394366
>mainstream chips
And I hope you realize that Pentiums and i3s are considered mainstream chips to Intel. The i5 is an enthusiast chip and the i7s are prosumer chips.

If AMD's own expectations turn out to be true (Excavator's IPC being 5% better and Zen being 40% than that), then a four core non-hyperthreaded Zen APU will be incredibly close to Skylake i3-6300's multi-threaded performance, but still fall ~20% behind in single-threaded performance.
It might take an eight-core non-hyperthreaded Zen CPU to match a Skylake i5-6600 (non-k) in multi-threaded performance, but a hyperthreaded Zen might still fall short of a four-core i7-6700 (non-k). The only unknown factor is how strong hyperthreading will be under Zen's architecture compared to Intel. If they can get within 60% of Intel, then that's a pretty amazing accomplishment given that it's AMD's first major foray with hyperthreading in years.

>>51394486
i3s and locked i5s are mainstream chips. Thats what you're likely to find in most OEM systems.

How Zen will compare to intel's Haswell and Skylake in serial performance will come down to clock speeds. A dual core i3 should be easily outperformed in well threaded workloads by a quad core Zen based chip provided clocks aren't disparagingly low. The benefit of an additional logical thread can't compare to an additional physical thread. Even the performance gap between Kaveri and Haswell i3 narrows to a spitting contest at roughly equivalent clocks when the workload is well threaded.

Moderately high CPU performance paired with an IGP that easily outperforms a Radeon 7850 is absolutely mainstream performance. Raven Ridge could provide that without any thermal or power issues whatsoever. Binning off CU and lower clocks to make mobile parts is trivial.

>>51394342
I will most likely get called a retard but I think I remember reading that bigger CPU caches increase latency so there is a tradeoff there somewhere.

>>51394359
>Zen has met all expectations in lab testing,

I'm sure that's what they said about the fury and whatnot. For their sake and everyone's else's I hope it's a success.

>>51394800
Larger blocks of cache do function slower.

>>51394702
>Even the performance gap between Kaveri and Haswell i3 narrows to a spitting contest at roughly equivalent clocks
You sure about that?
https://www.cpubenchmark.net/compare.php?cmp[]=2133&cmp[]=2522
The overall score includes the iGPU which makes the A10's win a no-brainer, but check out that disparity in single-threaded performance.

http://www.cpu-world.com/Compare/449/AMD_A10-Series_A10-7850K_vs_Intel_Core_i3_i3-4170.html
Now scroll down to the multi-threaded benchmarks and you'll see that even with four physical cores, the A10 lags behind or barely meets the i3 due to its much weaker physical cores.
You'll see much of the same on these benchmarks as well
http://cpu.userbenchmark.com/Compare/Intel-Core-i3-4170-vs-AMD-A10-7850K-APU/m28214vs2937

If we swap the i3 for a slower i5-4430S at 2.7GHz, you'll still see a difference in performance.
https://www.cpubenchmark.net/compare.php?cmp[]=2133&cmp[]=2111
Even when clocked at 70% of the frequency A10-7850k, the i5 still manages to outperform the A10 by over 15% per thread.

Zen's expected improvements in IPC and core strength will bring it up to parity with Haswell, or maybe even surpass it slightly. Given that Skylake really hasn't brought more than 5% improvements over Haswell, it might be enough to sway the market, but Zen+ needs to be a serious step up from Zen to catch up to the expected improvements from Cannon Lake.

>>51394928
>he actually thinks Passmark is a real bench

Oh boy, time to educate yourself
http://www.anandtech.com/bench/product/1200?vs=1197

The A10 7850k has a 3.7ghz base clock, and a buggy 4ghz turbo for lightly threaded workloads.
The i3 4360 has a fixed 3.7ghz clock.
This is the closest you can get to a direct comparison without trying to normalize performance to clock speeds or manually setting the clocks and voltage for Kaveri.

Handbrake LQ the i3 wins by less than 20%
Handbrake 2x4K the i3 wins by less than 10%
Hybrid x265 Kaveri pulls ahead
Multithreaded Cinebench R15 the i3 is less than 20% ahead
Multithreaded Cinebench R11.5 the i3 is less than 15% ahead
Multithreaded Cinebench R10 the i3 is less than 15% ahead
1st pass x264 the i3 is less than 5% ahead
2nd pass x264 Kaveri pulls ahead
7Zip bench Kaveri pulls ahead
POV Ray Kaveri pulls ahead
C Ray Kaveri pulls ahead

Cinebench is notorious for being ICC compiled and being ill suited to represent a real rendering scenario, POV Ray exists as an alternative because of this. Even still the performance gap between the two chips closes, or completely disappears when it comes to well threaded tasks.

>>51395094
>ignores the fact that single-threaded programs pretty much aligns with the Passmark single-threaded results and multithreaded results
And I already have. Passmark is even used by AMD themselves for getting a generic guideline on a processor's performance. There's a reason why it's called a benchmark, it's so that you don't have to run dozens of programs to know what sort of power the CPU can put out.

My point was that per core, the Kaveri A10 lags behind even an underclocked i5, which actually has less single-core performance than an i3 at similar clocks. AMD claims that Excavator will bump the IPC by 5% at best, and Zen 40% after that. Going by basic maths, we can conclude that a 4-core 4-thread APU will be close to a Skylake i3, which is an amazing achievement in itself and will definitely reflect well upon sales given that it has four physical cores rather than two. This might be the push Intel needs to stop dicking around with dual cores and go quad-core or more for future chips.
The unknown factor is how Zen's hyperthreading will compare to Skylake's, or even Sandy Bridge. If it is on par with Skylake, then holy shit a 4-core hyperthreaded APU will even give the low-tier Skylake i7 S a run for its money.
The other unknown is the cost. AMD might pull a stupid move and price Zen on par with Intel's Core series based on core count and hyperthreading. This is what Lisa Su hinted at when she said that Zen could not afford to be considered a "budget" option. If a quad-core HT APU costs the same as an i7, then that would immediately kill the hype for Zen and drive down sales to all but its entry SKUs.
Hopefully, an 8-core HT Zen will undercut the 6-core Skylake-E i7 in price so that the 8-core non HT Zen competes with the lower i7s and the 4-core HT competes with the i5s. That could be a slaughter on AMD's part if HT turns out to be on par with Intel's HT.

>>51395266
And given that the 14nm process has been giving both TSMC and Intel grief, we could see very low initial yields for full-fledged Zen 8-cores CPUs or 4-core APUs. That might drive up the starting price even further, dampening Zen's appeal to the mass market. Hopefully, one year's worth of ironing out production and yield problems will be enough to prevent what caused desktop Broadwells to get delayed.
Intel's been ironing out their 14nm process for well over a year and only now have they gotten production up to speed to release non-enthusiast i3s and Pentiums.

File: EZSL7Hw.jpg (85KB, 1200x675px) Image search: [Google]

85KB, 1200x675px

>>51395266
Passmark is not a valid bench, any similarities you see presented there and elsewhere are coincidental. It is not representative of any workload. The page I linked uses real benches, numerous ones, and it directly flies in the face of your tech illiterate anecdotes. It does not matter if AMD's marketing division uses it, they also use PCMark8 which is completely worthless.

I made an incredibly simple statement:
>The benefit of an additional logical thread can't compare to an additional physical thread. Even the performance gap between Kaveri and Haswell i3 narrows to a spitting contest at roughly equivalent clocks when the workload is well threaded.
This statement is 100% factual, and the plethora of evidence I posted verifies that. The Kaveri chip can roughly compete with, or flat out outperform the i3 when both have more or less equal clocks. The i3's hyperthreading doesn't provide the same multicore scaling benefit as a physical core, despite the fact that intel's core arch has significantly higher serial performance.


As for Excavator, the 5% IPC uplift is an average. It is not *up to* 5% higher IPC as you're incorrectly asserting. In Cinebench specifically Excavator shows a 9-13% IPC uplift over Steamroller.

A 40% average IPC uplift over Excavator would outperform a Haswell chip with equal clocks in tons of metrics. Your assertion that it would be 20% behind Skylake, like many things you've stated, is entirely baseless.

>>51395358
Oh jesus fucking christ. Child, stop. Stop posting on this board all together and go back to reddit.

TSMC does not have a 14nm process node.
Intel's 14nm process is nothing like Samsung/GloFo's.
Having the same marketing name does not mean they are even remotely similar whatsoever.
Samsung's yields for their 14nm LPE 7420 were reportedly 20-30% when they first entered volume late last year. They've since had no trouble providing adequate numbers of chips for their own devices, or for Apple's. They've delivered over 60% of all A9 chips for the new iPhone 6s line. Those are 78.23mm2 and 96mm2 dies respectively.

A 300mm2 28nm die would only be 90mm2~ scaled down to Samsung/GloFos 14nm node.
An FX 8350 would be smaller than 90mm2 in 14nm.

There are no issues to be found here.

all these mad intel fags because the only way they could get ahold of Master Keller work was buying DEC

>>51395382
I'm not arguing with you about physical cores being stronger than virtual cores, I'm arguing about how four of Kaveri's physical cores are still deficient to two of Intel's physical cores plus virtualized cores. I'm just showing how much of a disparity in core performance Zen needs to overcome if AMD is to regain market share from the gamer market, which is the only desktop market that's bucking the death-spiral that is desktop sales. I have my doubts that Zen will be able to match Skylake's per-core performance based on AMD's own claims and expectations, but that's still a good thing for AMD if the pricing is right.

Benchmarks may be worthless for representing real world performance, but they still give a grounds for comparison between dissimilar processors. Benchmarks only shows the processor's performance in an overly-generic and simplified manner, but it still reflects accurately with real world performance as you yourself have shown with actual program comparisons. Single-thread dependent programs do almost twice as well on the i3 than it does with Kaveri. Multi-threaded programs are a split between Kaveri and the i3 based on whether higher-speed memory is utilized, whether the program utilizes L3 cache (which Kaveri lacks), and so on.

>>51395495
>TSMC does not have a 14nm process node.
http://www.tsmc.com/english/dedicatedFoundry/technology/16nm.htm
I was mistaking Global Foundaries for TSMC. But that still doesn't change the fact that the 16nm process has caused TSMC problems as much as Global Foundaries with 14nm, shitlord. Die changes usually causes production and yield issues.

And no, I'm going to continue because I have a financial stake in both SK Hynix and AMD. How well AMD's Zen, K12, and GPUs does is in my best interests.

File: 4170vs6300.png (69KB, 600x450px) Image search: [Google]

69KB, 600x450px

>>51389608
Like the way the 4170 is BTFOing a stock fx 6300 here?

i think it was mighty nice of AMD to leave all their fx processors unlocked, letting fx6300 get quite close to i5 performance, for a much lower price.

>>51395643
>, I'm arguing about how four of Kaveri's physical cores are still deficient to two of Intel's physical cores plus virtualized cores.

And you're wrong. As has already been amply demonstrated.
No amount of talking around in circles is going to change this. Relying on Passmark for anything makes you sound like a tech illiterate retard enjoying his very first day here on /g/.

>>51395643
Global Foundries has no issues with their 14nm process. Their internally developed 14XM node was dropped years ago, they are offering Samsung's 14nm LPP at Fab #8 in New York. Test wafers are continually run across all common facilities to ensure uniformity.
If Samsung can produce an 80mm2 die with 30% good candidates per wafer at their own facilities, then GloFo would do the same, and we know that Samsung is having no issue doing exactly that.

You're just talking out of your ass continually.

>>51395495
>>51395382
>Passmark is not a valid benchmark
Then why does it reflect so well against your own proof that Kaveri lags in single-threaded programs and matches or slightly leads in multi-threads?

AMD themselves said that they expected a 5% IPC improvement. That's why I'm going on their number, not other speculations that Excavator is 10-15%. I could be over 5%, but we know that it's at least 5%. It also depends on SKU-to-SKU comparisons: are we taking about Kaveri A8 to Excavator A8? Or Godaveri A10 to the flagship A10?
For all we know, it could be that Cinebench was comparing the original Kaveri A10-7850k to Excavator's A10. Or even an A10-7800k. We don't even know what Excavator's clocks are like compared to Kaveri or Godaveri.

The 20% number is based on the absolute worst case scenario, where 5% is 5% between the regular Kaveri A10-7800k and the flagship Excavator A10 APU and 40% is just 40% between the flagship Excavator A10 APU and the regular Zen A10.
We just don't know because the final numbers aren't out yet. The next A10 isn't due until the Spring of next year.

>>51395778
First of all, Carrizo is an already released product. Carrizo based laptops have been on the market for months, and they've been benched to hell and back. You not knowing this is just laughable.
5% is an average IPC uplift, not the maximum. What I posted is not speculation, its straight from AMD's slide deck presented at a trade show. IPC is IPC, clocks are completely independent from it. IPC is the same regardless of what SKU you'd be looking at.

>>51395752
>Global Foundries has no issues with their 14nm process
Nigger, are you kidding!? They only just managed to hit their production milestone this month after months of delays. AMD was even rumored to look elsewhere for their source.
Global Foundaries inability to produce results with 14nm was one of the main reasons why most financial analysts were predicting AMD shares to plummet to $1.50 by Q3 2015.
>but GPU market shares
AMD's GPU revenue don't amount to that much compared to their bread and butter CPU and enterprise sales.

>>51395882
>I have literally no idea what I'm talking about

GloFo fully qualified their 14nm LPP lines at the end of Q2 this year, they were LPE qualified at the beginning of the year.
What you read in WCCFtech rumors doesn't matter. AMD is bound to GloFo by their wafer supply agreement, and porting a design to an entirely different process doesn't just happen over night.
The odds of AMD producing the bulk of their CPU logic outside of GloFo is literally zero.

Flat out you're completely wrong about every single thing you just posted.

>>51395942
>What you read in WCCFtech rumors doesn't matter
Nobody takes currytech seriously. Everyone takes no news, no results, and visible delays very seriously. It doesn't take Patel and Sanjeet shitting side-by-side to realize that GF didn't report a successful pre-production run months after they scheduled.

>>51395964
>further proving you're just a dumbass in over his head

GloFo had no scheduled customers. No one expected them to start shipping 14nm chips this year. You cannot have a "visible delay" when you have nothing entering production. Their lines were qualified within two months of Samsung's lines being qualified.

I'm set for probably another 4 years with my skylake tb.h. I've also got me 16gb of ddr4.

>>51376996
Still rocking my Phenom II x6.
And secretly hoping they make the socket AM3++

>>51396002
>GloFo had no scheduled customers
Everyone knew where AMD were going to get their fabs from. They're not going to shit out chips from nowhere and GF's relationship with AMD pointed to a no-brainer answer. But when GF's own deadline passed with no news on the status of their new fabrication, people knew that there was a fucking issue. Yes, Intel uses a significantly different method for producing their own 14nm dies, but they have a much better cushion to rest upon if things went wrong due to their market share. Broadwell was supposed to come out months before Skylake, but problems with 14nm yields cause them to push it back until roughly around the time Skylake came out, and even that was a paper launch.
GF and TSMC has had issues with die shrinks in the past, which could be why GCN was unambitious refreshed rather than shrunk to a 20nm die as was rumored. We knew that AMD was planning on a die shrink after the 290X's launch before stepping into 16nm, but that never panned out. People were worried that Zen could be the same story, but with much bigger repercussions.

With Zen, that 14nm die shrink isn't as important when it comes to IPC gains, but it is important since rescaling the architecture at the last minute would have added even more delays to Zen's launch.

>No one expected them to start shipping 14nm chips this year
>You cannot have a "visible delay" when you have nothing entering production
GF planned full production for this earlier year though. AMD even thought about pushing back their launch to 2017, back in 2014
http://www.eteknix.com/leaked-amd-slides-show-8c16t-zen-cpu/

Why the hell are we even arguing? We're both hoping for Zen to live up to the hype, I'm just being more cautious about my own expectations. I don't think Zen will be 100% on par with Skylake and I fear that AMD will price their chips wrong.

>>51396244
>more things pulled from your ass

>GF planned full production for this earlier year though.
No, no they didn't. They qualified their 14nm lines in the same time frame as Samsung. They could not have produced anything unless it had taped out before, and targeted that window.
There was nothing. Literally nothing.

The only argument happening is due to you being a dumbshit who keeps talking out of his.

gf was literally created by AMD
what relationship are you referring to?

>>51396345
Then go fuck yourself, dickhole.
Zen was supposed to be released early in 2016 with K12 following in 2017 according to Rory Read in a Bloomberg interview done on 09-04-2014.
Then AMD went quiet about the release date this year after Su took over.
Only know do we know that it's being pushed to Q4 2016.
There was supposed to be a 20nm die stopgap released this year.
http://www.kitguru.net/components/cpu/anton-shilov/amd-new-chip-designs-are-more-than-18-months-away/
That never happened when AMD announced Skybridge was dropped. It was rumored that that same 20nm was going to be used in AMD's GPU by 2015. Back in 2013.
https://www.techpowerup.com/188551/new-amd-gpu-family-codenames-volcanic-islands-and-pirate-islands.html

Pulled out of my ass, huh? Go back to reddit, your lead-addled memory does you no good here.

>>51396375
That GF was spun off from AMD a few years ago.

File: dem nodes.jpg (138KB, 1348x1012px) Image search: [Google]

138KB, 1348x1012px

>>51396567
>Zen was supposed to be released early in 2016 with K12 following in 2017 according to Rory Read in a Bloomberg interview done on 09-04-2014.

No it wasn't. No exact time frame was given, the only actual figures we've ever had came from a slide deck from March this year.

>There was supposed to be a 20nm die stopgap released this year.
The only 20nm parts that were planned were a Cat core APU, and an ARM based APU.
All 20nm parts were scrapped because cost per transistor wasn't worth it.

AMD themselves made it explicitly clear that 20nm was terrible in the industry. There was literally never going to be any 20nm GPUs. Never.
Unsubstantiated rumormill bullshit is not what an educated person would rely on to make an argument.


You are just pulling things from your ass. Over and over. Nonstop.

>>51396638
>You are just pulling things from your ass. Over and over. Nonstop.
I even referenced the EXACT DATE of the interview Rory gave to Bloomberg about Zen's planned release in Q1 of 2015, cumsucker.

>>51396778
*Q1 of 2016
You're still a cumsipping sissy bitch

>>51396778
http://www.bloomberg.com/news/articles/2014-09-04/amd-ceo-says-company-faces-bumps-in-road-before-new-chips-help

Oh man, I wonder why there isn't any reference whatsoever to an expected Q1 2016 release date in there.
Looks like you just can't stop pulling things from your ass.

The only reference to 2016 is this line:
"AMD engineers are now proving they can deliver new designs on time, something that didn’t happen in the past, Read said. New chips will help AMD gain in new areas and improve its performance in PCs and servers from next year increasing into 2016, he said."

That had nothing to do with the release of Summit Ridge.

>>51396784
Cry more, redditor child.

>>51396843
>links a fucking blurb of a 20-minute long televised interview
>wah, where's your proof?
You seriously cannot be this stupid

Going to buy it just for the socket. Not buying into another Intel socket that is declared obsolete a year later.

>>51390838
>unless you're a poorfag with no discrete gpu.
So you're leaving out laptops and tablets?
Jesus you would get raped in a freshmen marketing class.

>>51394359
>Zen has met all expectations in lab testing
Translation
>It has met all of AMDs abysmal internal standards which we would never publicly reveal because it is literal shit,

Ive already pruchased $3000 in amd stock, i suggest for you all to do the same

>>51384990
Yet another 2500k here.
No reason to upgrade yet, but I'll be watching amd very closely... A year or two down the track I'll need an upgrade and Intel feels kind of stale now.

>>51381289
Intels releases have very small performance gains. If Zen has the gains projected it will match current intel cpus, and if this happens again, amd will come out on top..

>>51376996
>Zen
Well, I'm hoping it doesn't suck, because if it does, Intel will have us by the balls.

>>51399647
Or intel will just pay off OEMs again like they did in the Pentium 4/Athlon days.

>>51388958
x86 is the instruction set. Yes there will be x86_64 processors in the Zen lineup.

>>51389747
Because most people think that x86=32 bit and x64=64 bit. Anon misread

>>51399696
>Yes there will be x86_64 processors in the Zen lineup.
All of them will be x86_64 processors.

>>51377341

it's pretty clear that the 'unit' part is true considering all the information there is correct and has since been confirmed by AMD

i'm guessing we'll only see zen CPUs with groups of cores in multiples of 4, with the entry level cpus being low clocked 4 cores with igpus.

>>51399870
> considering all the information there is correct
It isn't correct. Zen does not use FMACs.

>>51399919

[citation needed]

I'm more of an Intel guy but definitely interesting, will try to stay up to speed one this.

>>51399983
This isn't wikipedia, and this isn't reddit.
You're not going to be spoonfed for behaving like an insufferable dumb faggot.

Various patch notes and technical details lifted from technician's linkdin accounts already spelled it out. Non fused.

>>51400052

>i can't back up my assertions so i'll fall back on ad hominem

>>51400160
>literally has no idea what an ad hominem is
You are legitimately retarded, not a good combo when you're a shitposting childish faggot to begin with.


+;; Decoders unit has 4 decoders and all of them can decode fast path
+;; and vector type instructions.

+;; Integer unit 4 ALU pipes.

+;; 2 AGU pipes.

+;; Floating point unit 4 FP pipes.

+ 32, /* size of l1 cache. */
+ 512, /* size of l2 cache. */


Excerpt:

4 wide decoders
4 integer ALUs
2 AGUs (for 2R 1W L1 cache according to a LinkedIn profile)
4 FP pipelines

That makes z ten pipelines with a general four wide design.

There is a lot more information, which I will collect over the next days. Some stuff is copy pasted from Excavator (bdver4) or Jaguar (btver2) and modified then. But careful comparing did show some clear differences, while at other places it's not clear, if there is new information or not (e.g. div latencies). But as btver2 has 2048 kB L2 and the rest of the block is more similar to bdver4 or btver2 than btver1 (Bobcat), which has 512 kb L2, it looks like no btver1 files were used as a source. So I assume, that this is a new entry of an L2 cache size, indicating fast L2 caches per core. The L1 data cache still has the same size as that of Jaguar or Excavator. Some patents mention an 8-way 32kb L1 D$.

Interestingly, as there are two 128b FP mul and two 128b FP add units (with only 3 cycles latency for these ops), the FMA instructions will be executed by combining one FP MUL and one FP ADD unit, resulting in 2 issues and 5 cycles latency (as that of the Bulldozer family)

>>51397184
You could use the phenom ii in the am3+ sockets, right?

Also why does AMD need so many sockets? AM1, FM2, AM3...

>>51400350
>You could use the phenom ii in the am3+ sockets, right?
Yes

>>51400350
They kind of went somewhat tard with the sockets
AM3+ was kept for the FX's, the FM sockets were needed for the APU's, but they did introduce newer versions kind of fast, I'm not sure why they needed the AM1 socket

>>51400350
AM3/AM3+ are old as shit.
They only carried on to AM3+ as a stop gap, they could sell new boards that still supported old chips. They knew Bulldozer was turning into a shitsandwich and needed a platform to soften the blow of it.

FM1 came about for Llano APUs, this was originally going to last a couple generations but never did.
FM2 for Trinity/Richland, this too was supposed to last several generations but never did. AMD started treating binned chips with IGPs fused off and refreshes as new product lines.
FM2+ as a stop gap for Kaveri/Godavari because the design was delayed significantly, has the same backwards compatibility.

AM1 is for their reused Cat core chips. Single channel memory, super inexpensive. They sell in good numbers in poorer parts of the world since you can get a whole computer for like $100.

>>51400439
I can get a j1900 for $55. Or an athlon 5350 and mobo for $130 last I checked.

>>51400500
You can get similar single board systems with an AMD chip on them as well. See: http://www.newegg.com/Product/Product.aspx?Item=N82E16813157595

The point behind AM1 is that you have a socketed system. The upcoming Stoney Ridge APUs use the same socket as their predecessors, they'll undoubtedly end up on AM1 as well.

File: file.png (648KB, 874x818px) Image search: [Google]

648KB, 874x818px

>>51376996
>Who's waiting for AMD ZEN here?

Not me, I'll be looking for this bad boy :^)

>>51400790
Don't Xeon Phi processors use extremely old cores?

>>51400790
Knights Landing it's a coprocessor, it's serial performance it's pretty shitty, it's made for massive parallel workloads
Which AMD's and Nvidia's GPGPU's are better at
>>51400804
They use a P5 derivative
But well, pretty much anything Intel does is a P5 derivative

>>51400804
> The Knights Landing chip can deliver over 3 teraflops of peak performance, which is roughly in the range of some high-performance graphics chips used in the world’s fastest supercomputers.”

My conditions for buying zen

>$500 or less
>10 cores
>same ipc as skylake
>4.5ghz or better average oc

File: smug afro american.jpg (23KB, 205x256px) Image search: [Google]

23KB, 205x256px

>>51400817
good luck on that shitty tdp :^)

>>51400813
>>51400812
So they're using real cores now, sweet. I just double checked on why I was thinking that, the original MIC arch used modified P54C designs

>>51400790
Hope you've got $4000+ laying around and adequate cooling for that 300w beast.

>>51400804
They're part of the Atom family, though they're apparently going to become radically different from any current Atom family core arch.

>>51400812
Knights Landing can function as a coprocessor, or a standalone.
Its being sold as a socketed chip as well as an add in card.
Decreases wirelength, and really streamlines multisocket racks.

>>51400817
The only one of those going to happen is the price.

>>51400350
>>51400402
Are you retards serious? Can someone actually compare the frequency of AMD sockets to Intel sockets. Cause I'm sure you'll find that AMD have significantly less.

>>51400350
>>51400402
>>51402227

Flagship Consumer x86 sockets in the past 6 years

AMD
>AM3 2009
>AM3+ 2011

Intel
>LGA 1156 2009
>LGA 1155 2011
>LGA 2011 2011
>LGA 1150 2013
>LGA 1151 2015

>>51402375
115x is the consumer line.

2011 is the enthusiast/server line.

For 2011 there are also 2011-w for wide and -L for slim.

You also forgot fm1,fm2, fm2+ and several other sockets for their whole shitty lineup of apus.

I'm sure they can reuse the sockets but the pinouts are different from chipset changes so thwy change a pin or two to prevent retards fucking their shit up and general confusion.

>>51402375
>>51402420

And 2011v3

>>51400817
>same ipc as skylake
Well most people are rumoring it to be Haswell performance so it'll be even better than Scamlake

>>51402946
Knowing AMD cpus it will be pretty good at OC.