shit thread
tl;dr

zen's unlikely to be ground breaking but it lays a solid foundation for zen+ to be a shit kicker

as per the usual i5 2500k 4 lyfe

>(((tom's hardware)))

File: 1472088665709.png (1024KB, 1300x975px) Image search: [Google]

1024KB, 1300x975px

>>56260168
Get a load of this pajeet.

>>56259828
AMD announced at their 2015 FAD that work on the Zen+ core was already well under way.
How you can manged to try and spin this into a negative is beyond me.

>>56259828
AYYYYYMD FINISHED AND BANKRUPT

>>56260069
Oh the denial and rationalism has already begun I see

>>56260195
I don't understand what the thought process is- this is good news. This is AMD expecting Zen to do well enough to support a Zen+ a few years later.
Intel is working on shit like 4 generations ahead

>>56259828
As hilarious as the massive shitstorm of Bulldozer's release was, really don't want a repeat of that. AMD needs this to sell, Bulldozer flopping almost caused them to back out of the high end CPU market entirely.

can't wait for
>j-just wait for zen+!!!

>>56260333
and for intel, "4 generations ahead" means "15% improvement" at cost parity.

>>56260366
nobody wants/expects zen to beat kaby lake

this is going to turn into the same fucking situation as the rx 480 where you shills are going to superhype the fucking architecture with retarded outlandish claims and then act like it's a disaster when it doesn't live up to it

>>56260544
>shills are going to superhype the fucking architecture with retarded outlandish claims and then act like it's a disaster when it doesn't live up to it
Fanboy retards always do this. They're so completely adverse to reason its almost like they're paid to spread disinfo to make the product seem worse at launch by the competition.

>>56260333
Zen is AMD's P6. What their K6 arch could have been if they had stuck with it after K10/10.5.
They told reporters that they had an internal roadmap showing their plans for the Zen core arch several generations down the line. Its here to stay for a number of years, and they're aiming for high performance with it.

>>56260322
>Oh the AMD Hype has already begun again

Fix'd

Drones will never stop overhyping

>>56260652
Not to say Zen is guaranteed to fail, but AMD having roadmaps for followups is no guarantee of success either. AMD had plans for several generations of the Bulldozer line as well, and we all know how that went.

>>56260753
Zen however is guaranteed to be better than anything in the BD family.

>>56260821
True, but that isn't exactly an accomplishment.

>>56260866
It is for AMD who haven't been able to do anything meaningful on the CPU front for performance since the Phenom II.
Excavator barely manages to be neck and neck with the Phenom II line clock for clock, and Zen is promising a 40% average IPC increase over it.

For a new starting point with brand new arch thats pretty good.

>>56259828
KABYLAKE CONFIRMED DOA

CANNONLAKE IN THE WORKS

File: 40mm cadillac.png (58KB, 784x309px) Image search: [Google]

58KB, 784x309px

>>56260892
For those of us still rocking Phenom IIs it will finally be the chance to upgrade. I am cautiously optimistic.

>>56261080
http://www.anandtech.com/bench/product/362?vs=700
http://www.anandtech.com/bench/product/362?vs=1684
http://www.anandtech.com/show/10436/amd-carrizo-tested-generational-deep-dive-athlon-x4-845/7

A 40% uplift over Excavator is a decent uplift over the highest clock Phenom II SKUs. Going by the shear size of the FPU in Zen its safe to say it'll be better on all fronts, no trade offs in int vs fp performance.

If the core really does follow Excavator's power per clock scaling, then 7w per core equates to 2800mhz. An 8 core 65w part thats a direct upgrade from the 125w Phenom II X6 1100T would be hard to ignore. 3200mhz should be just 10w per core. A quad core desktop chip with moderate clocks could come in under 65w and be a worthy upgrade to any Phenom II X4.

FinFET devices scale completely differently from planar ones though, so the Zen core can't have the same voltage/clock curve that Excavator has. It likely only meets on a few key points. It should scale down in clocks much, much better.
Not that this is of any interest for the average user.

File: 5-1080.2494205094.png (64KB, 1920x1074px) Image search: [Google]

64KB, 1920x1074px

>>56261309
relevant

>>56260069
>Zen is complete shit but Zen+ will surely be good! Just wait!
The cycle of AMD continues

File: avfs scaling.png (152KB, 725x381px) Image search: [Google]

152KB, 725x381px

>>56261442
Power per Core Pair

XVR with AVFS
3700mhz: 22.5w~
3500mhz: 16w~
3150mhz: 10w~
2800mhz: 7w~
2450mhz: >5w~
2100mhz: 3w~
1750mhz: >2.5w~

STR
3700mhz: 25w~
3500mhz: 20w~
3150mhz: 14w~
2800mhz: 10w~
2450mhz: 7.5w~
2100mhz: 6w~
1750mhz: 4.5w~


So thats what AMD is comparing Zen to power wise. Its safe to say peak perf/watt frequency isn't on the higher side since the Summit Ridge dies are primarily fabbed for Opterons.

File: 1470074811298.gif (391KB, 900x823px) Image search: [Google]

391KB, 900x823px

>Intel+Nvidia consistently release great products
>AMD keeps making huge promises and falling short of them
>j-just wait!
Why do people willingly cuck themselves into buying AMD? You fags couldve bought an i7 and a GTX 1080 now and enjoyed great performance instead of waiting for a unicorn to appear.

>>56259828
Really hope my FX6350 lasts until Zen+

/g/ knows absolutely nothing about how cpus work which is why there is only shitposting when it comes to cpu news.

>>56261783
In general the average post here doesn't know a single thing beyond what smartphones are coming out, and what meme programming languages are trending.

File: 17-1080.573265681.png (251KB, 1920x1079px) Image search: [Google]

251KB, 1920x1079px

Not too long ago there was some autismal fuck constantly posting FUD about Summit Ridge only have single channel memory per CCX, and doubly claiming that the L3 was isolated per CCX as well.
He'd show up in every single thread repeating the same thing.

Funny how the little shill autist disappeared.

>>56261949

I wouldn't be surprised if its The Stilt posting on /g/.

>>56262006
Ohh that fags on ocn all the time shitposting but making it look legit lol

>>56262006
Dunno, could have also been Seronx. He's been pretty fuckin derpy when it comes to Zen shit as well.

>>56259828
>people here have no single fucking clue about how semiconductor businesses work
By the time the first design of the architecture hits silicon they've already designed the next two updates based on that architecture.
AMD made some shitty bet on more cores and thus lost to intel for a several years no matter how competent they are. By the time they realised it they couldn't do anything but ship whatever they had at the time.

>>56262006
That would be hilarious, and entirely believable.

>>56262112
Hes a legit schizophrenic, and a tremendous AMD fanboy. The Stilt however has some vendetta where he tries to contort everything to somehow be a negative for AMD.

>>56261607
But my Fury X and FX 8 core works great for me

>>56262142
Huh, learn something new everyday.

The Stilt though I agree, he's always starting shit on OCN. Fortunately block lists are a thing.

>>56260544
>nobody wants ... zen to beat kaby lake
I want Zen to beat Kaby Lake. I want Zen to be the second best processor ever made, beaten only by Zen+ as the most stunning achievement of mankind with the unlimited power fanboys said the Cell had. I also want my girlfriend to be permanently attached to my penis and to never age. I don't expect either of those things though.

>>56261607
>Nvidia
>Consistantly great

Shill Shill go away
Earn your pay another day

>>56262197
Fanboyism knows no bounds. Doesn't matter how educated(allegedly) someone is.

>>56262237
let me rephrase that. I want a platform that's not made by Intel. it doesn't matter to me whether or not it's faster. only that it's more than fast enough, which Zen is looking to be.

I am most assuredly not the only one.

>>56262197

OCN is so fucking lenient on the threadshitters in the news section. There really are about 10 posters that routinely go into an AMD focused thread and shit it up with off topic comparisons to Nvidia or Intel.

At least Chev Chelios got banned.

>AMD product comes to reality
>It sucks
>Drones start talking about how great the future is going to be

Every fucking time, this is getting sad.

>>56262259
80% marketshare doesn't lie.
Having the best GPUs that wont set your house on fire doesn't lie.

File: IMG_5884.jpg (1MB, 2272x1704px) Image search: [Google]

1MB, 2272x1704px

>AMD executives repeated the "Zen is not a destination; it's a starting point" mantra during both of the Zen presentations in an obvious ploy to tease the forthcoming Zen+.

JUST KEEP WAITING
GOOD JOB PAJEET

>>56262583
It is actually 70 percent now. AMD is up at 30 percent as of most recent numbers.

AMD is the Chicago Cubs of the tech industry.

>>56262583

Its not 80% anymore. In fact AMD has been gaining marketshare for over a year.

http://jonpeddie.com/publications/add-in-board-report/

why are people so obsessed with what other people spend their money on? is it autism or are people just that pathetic?

>>56262669
Baser human tribal instincts, and dire autism.
Fanboyism is basically just an extension of my tribe vs your tribe after all.

I actually hate that Zen isn't looking like a dud because now I can't wait for it to be released
REEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

>>56262774
Yeah bro i really want to make a build but still no zen till probs feb 2017

/requesting/ the AMD fanboy starter pack image

AYYMD IS FINISHED & BANKRUPT

How can it be a failure if it's never going to be released?

Where's looking at the end of 2017 at the earliest for desktop Zen

>>56261607

>download intel driver app
>never finds the right drivers or claims an update even though you have the same one
>ends up having malware in it

yeah intel is awesome

>falling for the AMD waiting meme

>>56259828
what a shitshow

>>56262491
>I am most assuredly not the only one.
I agree with you there. I wouldn't mind if Zen gave mind boggling performance but for now, it's of little interest to me. I'm buying Bristol Ridge on launch because it nicely fits into the "good enough" bracket and because I enjoyed overclocking every part of my Kaveri chip. Going by how AMD have previously done things, I should be able to just slot in a Raven Ridge chip when that time comes.

>>56262774
Yeah cant wait for that Haswell performance in 2 thousand fucking 17.

>>56262862
>the end of 2017 at the earliest for desktop Zen
Erm, what? If you had said mobile Zen I'd be on board but AMD have been fairly consistent about saying late 2016 with full availability in 2017. I'm guessing that means a CES launch.

File: 1471392041531.jpg (100KB, 640x480px) Image search: [Google]

100KB, 640x480px

>>56259828
>AYYMDkeks have been waiting half a decade for this

>>56262944
You say that as if Intel has long since left Haswell in the dirt, markedly improving performance since that time.

>>56262950
A few low end mobile Zen chips MIGHT be available in 2016 just so they can tell their shareholders they launched Zen in 2016

Actual good Zen chips MIGHT launch in 2017 if they don't get delayed again, and by that time Intel will have launched several new lines of processors putting them even further behind

>>56262972
still doesnt change the fact AMD is years behind the competition

>>56262950
Hes shitposting, just spammed the same thing across a bunch of threads.

Raven Ridge APUs are due out mid 2017. Summit Ridge beginning of the year with limited availability end of this year.

>>56262979
Stop talking about things you're clueless about.

>>56262979
>low end mobile Zen chips MIGHT be available in 2016
Wow, as the other anon said, you really are clueless. You're a mucous eating oaf. Just for clarification, Zen won't be hitting APUs until Raven Ridge, which is not this time around. The FX series won't have an IGP and you'll never get laptop manufacturers to put two chips in a low end laptop because of costs. I seriously doubt you'll get manufacturers to put any chip without an IGP in a laptop any time soon.

>>56260821
>Zen however is guaranteed to be better than anything in the BD family.

That's what they said about Bulldozer against the Phenom too.

File: bulldozer-module.jpg (139KB, 620x362px) Image search: [Google]

139KB, 620x362px

>>56263096
No, its actually not.
One marketing guy on Overclock.net was spouting bullshit like this totally without anyone's approval.

What AMD themselves said about Bulldozer was this:
>Throughput advantages for multi-threaded workloads without significant loss on serial single-threaded workload components
>without significant loss
>admitting very clearly that there is some loss

AMD showed what the Bulldozer arch was, only internet fanboys refused to listen. Exactly as they always do. AMD was showing Zambezi competing against some intel chips in very specific integer bound workloads, not making direct clock to clock comparisons vs their own prior arch.

Stop twisting the facts around to fit your bullshit narrative.
The situation with Zen is nothing like Bulldozer.

>>56260069
>It's shit but the next one will be better
Always Massive Disappointment does it again.

>>56263096
>>56263201
I wonder how an improved, 14nm K10 would fare against Zen.

>>56263254
Look at the CPU component of Llano.
Per clock it was up to 5%-10% faster than a comparable Athlon II X4 without L3. They conceivably could have kept evolving the core arch, but it would have taken considerable work.

Per clock Llano can still hold its own against Kaveri.
http://www.anandtech.com/bench/product/399?vs=1270

Llano has no turbo core, its fixed 2.9ghz all the time. The Kaveri chip turbos up to 3.8ghz. Multicore scaling in Steamroller is still worse, and given how far behind Trinity was in FPU performance it wouldn't have changed much in Kaveri.

>>56263254
That's so long ago at this point that you might as well be speculating about how things would have been if the bizarre code morphing thing Transmeta had going on worked well enough to be competitive.

>>56263254
Not that good.

In fact Zen is more like K10 than Bulldozer+ since it basically adds hyper-threading, improves cache performance, new instruction sets, etc. compared to K10.

The only real thing wrong with Bulldozer+ was the cache performance. But Intel was so successful with SMT (hyper-threading), that AMD basically just wanted to go that route instead of trying another CMT design.

>>56263344
Not him but Kaveri has some fairly serious issues with turbo. There's been tons of speculation about it but no one seems to have figured out why. Regardless, I can guarantee that Kaveri isn't actually hitting, let alone maintaining, those turbo clocks. Putting that aside, one can't normalise for clock rate in a situation like that as that's an intrinsic part of the architecture. There's also the argument to be made that Kaveri was optimised for lower TDPs than Llano ever was which some have speculated is the cause of the terrible turbo problem. Essentially what I'm saying is that it's SERIOUSLY hard to speculate on this because there's just so much going on.

>>56263201
>its good, you just dont understand!
AMD

File: power-line.png (21KB, 577x401px) Image search: [Google]

21KB, 577x401px

>>56263536
There is no issue at all, its a design feature.
The APUs throttle CPU clocks when the IGP is used, otherwise they would exceed TDP. Llano and Trinity/Richland all drew up to 130/140w. Kaveri cut it down to 95w through heavy throttling.

If you're only using the CPU component of the chip then the clocks are stable.

So is there an AMD CPU which can replace my current i5 3570K @4.5GHZ in gaming?

>>56263521
Cache performance was horrible indeed, but Bulldozer also has low single thread performance by design (few execution units).

>>56263585
>If you're only using the CPU component of the chip then the clocks are stable.
Unfortunately not, no. There seems to be some weird voodoo shit going on with the way Kaveri turbos and throttles, that's speaking from experience. I tried mine with a gt 610 and the IGP disabled, it still wouldn't turbo properly. Even when overclocking, it has problems maintaining clocks but only in some workloads, not in others. It just seems that AMD didn't really figure out of turbo until Carrizo.

>>56263550
But it *is* good, just not for what the average end user would want to do. AMD misjudged the market by around half a decade, shame really.

I really want Zen to be great, but I doubt it will change much. Odds are Haswell will outperform it, which gives most people little reason to care.

Plus it's hard to forget the hype Bulldozer had just to fail hard.

File: FX-9590-57.jpg (74KB, 537x568px) Image search: [Google]

74KB, 537x568px

>>56263695

This is /g/, if it doesn't play call of duty its shit because anything else is beyond this board's understanding.

Pic related - ignoring the highlighted 9590 (as the picture is from a review of it) consider the 8350's age and where it sits in this benchmark.

>>56263667
> but Bulldozer also has low single thread performance by design (few execution units).

Core width is not indicative of performance. Two ALUs and AGUs per int core in the BD derived chips would have considerably higher IPC if they could actually get fed instructions in a timely and reliable manner. Excavator could conceivably match Sandy Bridge without being any wider. Its entirely a matter of feeding the ALUs.
Keep in mind that Sandy Bridge's 3rd ALU(all prior SMT Core i arch as well) is so under utilized in a normal workload that its available to process the additional logical thread. That additional logical thread provides a 30%~ uplift in multicore performance.
The multicore scaling of CMT is around 80%~


>>56263695
I definitely wouldn't call it good. Its passable for certain things, but most certainly not good.
There are far too frequent pipeline stalls in the arch for it be deserving of any praise.

>>56263743
Its absurdly dishonest to look at 2nd pass results without 1st pass.

File: 1469948394822.jpg (322KB, 1200x1600px) Image search: [Google]

322KB, 1200x1600px

>>56261466
>>56263235
>>56260322

Honestly though, why is this STILL a thing? This board sounds like a fucking broken record when it comes to AMD shit and it's just fucking unbelievable. People here go out their way to literally market anything AMD related, even going as far as that fucking autist that made a 17 minutes long vocaroo recording licking amd's ass regarding async compute, as well as spreading FUD over every single thing intel and nvidia related. Then they hype the ever living fuck out of fucking EVERYTHING amd makes and it never - E V E R - delivers. It has never fucking delivered since the goddamn phenom and people are STILL doing this crap here.

This is just fucking unbelievable, seriously, I just have no words. Part of me wishes this is just the same people wanting to have some kind of fun but I know it's not true and it just makes me cringe as hard as humanly possible when I just glimpse the sight of a poor sad fat fuck hunched over his keyboard defending whatever amd POS he bought as if his life depended on it. It's not even funny anymore, this is just fucking sad.

File: image.jpg (26KB, 325x325px) Image search: [Google]

26KB, 325x325px

>>56263695
>But it *is* good
Just like the Pentium 4

>>56263828

Your ideas intrigue me and I wish to subscribe to your newsletter.

>>56262574

Don't forget how they start backpedalling on the hype they themselves created.

I have lost count of many times the newest AMD turd came out and their /g/ pajeets started going
>n-nobody s-said i-it was g-gonna be a-any b-better anyway! I a-actually l-like it this w-way!
Like I've seen these miserables fucks do after the so hyped polaris turned out to be yet another overhyped piece of shit.

>>56263695
>it's good! It's just a piece of crap for anything real world but in a theoretical world it's good!

Indeed, AMD.

>>56263743
And then you remember that rendering is literally the one thing this shitty CPU line up is good at whereas for everything else it's IPC that matters. Big fucking whoop.

>>56263888
Calm down pajeet. Your pay isn't high enough to afford trips.

>>56259828
>ZEN+ IN THE WORKS
No shit. And Kaby Lake has been in development since before Sky Lake was released.

>>56259828
>toms goyware

>>56263903
It's fine for anything but manchild tier gaming

It's also 90% the same performance as an i7 without a 300% markup. Pretty sweet deal actually.

>>56261466
Of course; they built it into their Zen logo (circle of smeared poo)

>>56264072

>It's fine for anything but manchild tier gaming

And so is any i3 that is cheaper, better, doesn't need an aftermarket cooler just to not sound like a jet taking off while throttling the ever living fuck out of itself and uses way less electricity while producing less heat.

>It's also 90% the same performance as an i7 without a 300% markup

For encoding and useless synthetic, yeah.

>>56264095

Fun fact: number 3 of the top500 is Opteron based.

>>56263922
>it's IPC that matters
I've seen this repeated so many times. No, it's not just IPC, that's just one aspect that makes up performance. IPC doesn't count for shit if the chip doesn't reach high enough clocks. IPC is not absolute performance and you should stop saying shit that makes it out to be absolute performance.

File: IMG0051653_1.jpg (141KB, 960x440px) Image search: [Google]

141KB, 960x440px

>>56262583
>>56262628
>>56262657
http://www.hardware.fr/news/14755/amd-continue-gagner-parts-marche-gpu.html

>>56264142

Mercury research normalises their numbers - you should read their methodolgy.

>>56264132
>IPC is not absolute performance
Uhh Yes it is in like 99% of cases you

don't speak if you don't know what your blathering about peasant.

>>56264132

>IPC doesn't count for shit if the chip doesn't reach high enough clocks

Are you somehow implying intel chips don't reach high enough clocks? You should stop talking about things you understand nothing about because it's AMD who always goes LOOK AT HOW MANY GIGUHURTZ WE HAVE!!! while having half the IPC of any intel chip, so it's more of a case of the other way around.

In any case you're brightfully retarded if you don't think that IPC trumps multi-threaded performance in the vast majority of case scenarios because guess what shithead, the vast majority of software isn't properly multithreaded optimized, unlike video rendering.

File: G-Force-soundtrack.jpg (42KB, 300x300px) Image search: [Google]

42KB, 300x300px

>>56263828
The irony here is that this board is filled to the brim with Nvidia shills that fit the exact description you posted.

>>56264142

>Mercury Research is
reporting the shares as a four-quarter volume-weighted average to smooth
the noise of seasonal inventory cycles and reveal ongoing share trends.


http://mercuryresearch.com/graphics-pr-2016-q3.pdf

>>56264183

Uh huh sure, it's nvidia shills that keep making those cringeworthy vocaroo recordings and spreading fud about competition while any mistake they make is the opposition's fault amirite, pajeet? It's also nvidia fanboys that have become famous for overhyping and shilling the ever living fuck out of anything their shitty company produces only to have said promises fall flat on its face, in which case they start churning their favorite "w-wait f-for n-next y-year!!!" meme huh?

Where is that guy that made a 10 thousand (5 full posts) characters long AMD shilling/nvidia trashing copypasta by the way? I dare you to show me anything even remotely close to that coming from intel and nvidia fanboys.

>>56264218

>anything even remotely close to that coming from intel and nvidia fanboys.

That is because Intel and Nvidia fanboys are too stupid to form a coherent argument backed up with evidence and citations.

>>56264241
>backed up with evidence and citations.

Yeah, just like your solid """evidences""" that always turn out to be false when the latest amd POS comes out and gets trashed by competition, huh?

Nice sidestepping by the way.

>>56264167
>Uhh Yes it is in like 99% of cases you
No, IPC times clock rate is. IPC is not absolute performance no matter how many times you repeat this.
>>56264181
>because it's AMD who always goes LOOK AT HOW MANY GIGUHURTZ WE HAVE!!! while having half the IPC of any intel chip, so it's more of a case of the other way around.
I never implied that Intel doesn't you mouth breathing knuckle dragger. I'm saying you need both to have good performance. Bulldozer performed like shit because it didn't have good enough IPC and Zen could very well be a lemon if it doesn't clock high enough.
>you're brightfully retarded if you don't think that IPC trumps multi-threaded performance
What the ungodly fuck does that have to do with anything? That's completely unrelated to what we're talking about here. I do believe that you're the one "brightfully" retarded here.

Not surprising that both the anons who confuse IPC and absolute performance are idiots incapable of typing coherent sentences.

>>56264218
Its Nvidia shills like typing up these cringeworthy blogposts and sucking Nvidia's dick on this board 24/7

File: 1469743235983.png (1MB, 1200x10000px) Image search: [Google]

1MB, 1200x10000px

>>56264263

Do you not believe man never walked on the moon because you hate neil armstrong?

>Yeah, just like your solid """evidences""" that always turn out to be false when the latest amd POS comes out and gets trashed by competition, huh?

Now again, you have provided nothing beyond "lol it r not true because I sez so!". The picture in question is backing up the claims it makes and the typical response just lilkes yours now is lacking any counterpoint, just a statement based upon your belief.

For those who don't know, this is half of the picture this other anon is discrediting.

Reminder that the opinions of amd shills are financially fueled and thus it is actual shilling. Nvidia and Intel "shills" on the other hand are just hardware enthusiasts with no motives other than to discuss hardware.

File: 1469743310719.png (2MB, 2000x8571px) Image search: [Google]

2MB, 2000x8571px

>>56264317

Part 2.

>>56264218
>Part of me wishes this is just the same people wanting to have some kind of fun but I know it's not true and it just makes me cringe as hard as humanly possible when I just glimpse the sight of a poor sad fat fuck hunched over his keyboard defending whatever amd POS he bought as if his life depended on it. It's not even funny anymore, this is just fucking sad.

This is what you're doing right now, anon. Look how fucking angry you are over a computer, making these autism filled blogposts, yet you project it onto AMD fanboys.

Real men have fabs.

File: 1470152323676.jpg (106KB, 1280x720px) Image search: [Google]

106KB, 1280x720px

>>56264317
Youre so fucking bootyblasted right now I might as well post a smug anime girl and be done with it

Well, this WAS a civil thread.

Then the fucking shills and trolls from both sides showed up.

>>56264380
Zen is shit and so is Polaris, just buy Intel+Nvidia and be done with it.
case closed.

>>56264365

We were having a (relatively) rational discussion about veracity of claims but I suppose that is over now.

>>56264406
case in point.

>>56264380
Sorry, I get very annoyed when people use the wrong terms. It's part of what's fuelling the unreasonable Zen hype. People see "40% higher IPC" and think it must mean 40% better performance because hey, IPC = performance, right?

I'm not saying Zen will be bad, just that caution and not getting technical terms mixed up are two very good ideas.

>>56260069
jim keller is no longer working for amd.

this will be another fail where amd makes slight improvements for the same architecture for the next 10 years like they did with athlon64

>>56264471
I am aware and understand that, at least you as an individual is trying to keep it civil.

Personally for me when it comes to Zen, I expect Haswell-level performance +/- a few percent in both directions depending on the workload. If it proves to actually have close to Broadwell-level performance, great. If not, then at least AMD has managed to close a fairly significant chunk of the IPC gulf separating Excavator and the post-Haswell chips, which in turn puts them in a far better position than what they are at right now with the construction cores.

>one guy makes a totally factual statement about total performance being a function of IPC x frequency
>a bunch of spastics lose their minds without even reading and comprehending what was said

Top tier posting.

A Sandy Bridge core at 3ghz out performs a Sandy Bridge core at 1mhz. Both have the same IPC.
Performance is a function of both IPC and frequency.

An extremely high IPC design, of which there are many(google some whitepapers you plebs) won't have sufficient real world performance if it can't clock high enough. There are tons of RISC implementations with absurd IPC that won't clock beyond 400mhz because logic density is far too high, and they often double for double, or in one case even quad clocked logic. A 400mhz core clock can equate to FPU logic inside the core running at 1.6ghz. One of Intel's designs used such an arrangement and had a double clocked FPU, so in a sense they nearly did get to 10ghz computing by having an 8ghz FPU.
Regardless, the statement holds true. A core needs both sufficient frequency and IPC to have adequate throughput.

>>56264471
Don't sweat it, nigga.
Most posters here flat out don't understand whats being discussed.

The Zen cores in Summit Ridge at least aren't going to clock high. They're fabbed on a dense variant of the LPP process that trades off frequency for density, particularly because low leakage is ideal in an enterprise chip and they target low clocks to begin with.
I expect max OC of about 4ghz for Summit Ridge. Raven Ridge may clock a bit higher, but not by much.

AMD stated Zen has the same energy per cycle as Excavator, and that works out perfectly for an 8 core part at 3.2ghz having a 95 TDP. If it scales in even remotely the same way then 3.5ghz would be about 15w per core, making the 8 core chip pull 120w.
Of course FinFET scaling sill be somewhat flatter than what we see in Excavator, but it should still be ballpark.

>>56264604
>One of Intel's designs used such an arrangement and had a double clocked FPU, so in a sense they nearly did get to 10ghz computing by having an 8ghz FPU.
That would be Netburst with the double-pumped hardware. If anything it made the heating issues worse.
Nvidia had the same issue with hot-clocked shaders in their Fermi and earlier GPUs.

>>56259828
nice b8, OP.

I don't care if it's even 10-20% slower clock-for-clock (and it likely won't be), since the closest Intel chip to Summit Ridge is a $1100 Broadwell-E.

Zen = full performance >4 core CPUs at non-Jew-tastic prices

>>56264604
>I expect max OC of about 4ghz for Summit Ridge. Raven Ridge may clock a bit higher, but not by much.
That's around what I'm expecting too. I'm not sure if I expect Raven Ridge to clock higher than Summit Ridge though, as none of the APUs so far have been able to keep up with their IGP-free counterparts in terms of clock rate. Of course, it's very possible that it has nothing to do with the IGP and I'm just speculating blindly.

Either way, Zen should be a nice step in the right direction. I look forward to seeing what it brings.

>>56264724
do we even know anything about zen prices?

>>56264741
>do we even know anything about zen prices?
Not him but absolutely nothing aside from the fact that AMD have previously stated that they don't want to be the budget option any more. Clearly, old habits die hard as the rx 480 was priced competitively and the rx 460 is dirt cheap.

>>56259828
OMG OMG OMG
a company announced a successor for a product

>>56264737
If Raven Ridge is fabbed to favor frequency more than Summit then it should be able to clock the CPU portion somewhat higher.

>>56264741
Nothing official yet.
Everyone has been expecting Summit Ridge to top out at $500, half the price of Haswell-E's top SKU.
I think either Anandtech or Computerbase wrote that they needed to be priced around $300-$400 to maximize value based on their performance estimations.

All AMD has officially stated thus far is that ASP will be higher than prior FX parts. The 8350 launched at $195 if I recall, but it had horrible profit margins. I think less than $300 for the 8 core top binned part would be ridiculous.

>>56264784
>>56264827
hmm. good to know.

in any case, even at price/perf parity with intel, there's added value in having an unlocked CPU on a socket that lasts longer than two years.

>>56264827
>If Raven Ridge is fabbed to favor frequency more than Summit then it should be able to clock the CPU portion somewhat higher.
Very true, which is why I said I was speculating blindly. I'm still not sure we should expect Raven Ridge to clock higher though, as AMD are probably going to be using the same die for both laptops and desktops. Surely they'd focus on power consumption first?

>with the same TDP than the FX

D R O P E D

>>56264784
>>56264827
AMD doesn't need for bargain basement prices, they just need to sell decent 8-core chips for $400 or under, which would still make them a decent profit.

In a way they're in a little better spot than Intel, who can't sell desktop chips with a bunch of CPU cores and minimal/no GPU cores without cannibalizing their workstation Xeon market.

>>56264949
Same TDP as Excavator

>>56264945
It all comes down to the curve and where peak frequency/watt lands. I hate to sound like a broken record, but if we hold to Excavator's clock scaling we see that 2.45ghz is roughly 5.5w, 3.2ghz is about 10w.
The ES Summit Ridge had a 2.8ghz base clock and 3.2ghz turbo inside of a 95w TDP. Final silicon will probably have a 3.2ghz base clock and slightly higher/varied turbo states.
The 32c Opterons, 4 dies on one package, if all clocked at 2.4-2.5ghz would fit perfectly inside of the rumored 180w TDP.

Those are key points on how the clock scaling of this core works on that specific process. The sweet spot is below 3ghz.
A less dense variant of the process could have it right at 3ghz. Say we're now hitting 3.2ghz at 8w per core instead of 10w, and we can hit 3.5ghz for 12w instead of 15w. That'd still scale to usable frequencies for mobile SKUs, but offer slightly more frequency for desktop systems as well.

>>56264604
FUCKING RETARD

you act like cpus won't be 3ghz and up

kek KEK

>WHAT IS CLOCK VS CLOCK PERFORMANCE

>>56265010
not really better desu

>>56264956

considering you can get a 5820k for under $400 Zen is likely going to have a tough time.

File: AMD_Steamroller.jpg (102KB, 949x579px) Image search: [Google]

102KB, 949x579px

>>56265010
energy per cycle*

>>56265296
It really is.
Piledriver in the average FX 8350 is pulling a bit over 31w for its stock 4ghz base clock.
Steamroller module needs 25w for 3.7ghz while being considerably larger comparatively, higher transistor count.
Excavator takes 16w for 3.5ghz, and scales lower in frequency better than anything else. 3.15ghz costs just about 10w.

Thats 80% of the frequency of Piledriver for 1/3 the power draw. In a considerably higher IPC core without a half FPU.

>>56265522
I mean it's still too high

>>56265864
You're right, it is.
As it stands the stated clock scaling is really only ideal for enterprise, and just acceptable for HEDT if you're not looking at OC potential. Its not going to cut it for mainstream parts when the competition is Skylake and Kaby lake hitting well over 4ghz at reasonable power levels per core. Intel still has a huge lead here, but they also have z dozen variants of their own process that they pay absurd amounts of money to develop.

AMD needs to hit about 4ghz at 15w per core. Zen+ will undoubtedly push frequency higher and energy per cycle lower, likely on IBM's 14nm FinFET process instead of Samsung's LPP.

File: 15-1080.2713563378.png (94KB, 1920x1076px) Image search: [Google]

94KB, 1920x1076px

Known factors:
Relative IPC
target frequency range
general energy per clock

Unknown factors:
multicore scaling
SMT throughput uplift per core

I'm pretty curious to see what they'll pull off in successive iterations of the core arch.

File: stallman chip.jpg (237KB, 1280x1080px) Image search: [Google]

237KB, 1280x1080px

>Piledriver will save AMD!
>Steamroller will save AMD!
>Hawaii will save AMD!
>Fiji will save AMD!
>Polaris will save AMD!
>Zen will save AMD!
Now:
Zen+ will save AMD!
Vega will save AMD!

Fool me once, shame on you
Fool me 8 times, then holy shit why isnt this piece of shit company bankrupt.

>>56266879
The 4th gen GCN Polaris cards are selling well.
AMD already has major enterprise wins for Zen based parts including two super computer supply deals.
One firm is even paying for a Zen based semi-custom HPC chip to build a cluster with.

Notice how they managed to not go bankrupt? Yeah. They are "saved" already.

File: 1456197374470.jpg (9KB, 184x184px) Image search: [Google]

9KB, 184x184px

>>56266914
>AMD already has major enterprise wins for Zen based parts including two super computer supply deals.
One firm is even paying for a Zen based semi-custom HPC chip to build a cluster with.
But they still cant make a single desktop product worth buying.

Sure, you AMDtards can keep typing up your little essays about how the RX480 is magically better than anything Nvidia will make because of buttcoin mining performance, or just outright make up bullshit like Intel is run by Jews or Nvidia gimps cards, but facts are facts, and real results always show AMD is behind the competition.

The worst part is you all go out to defend them despite this bullshit, like your lives are at stake.

>>56261949
The L3 *is* isolated per CCX. Unclear about number of memory channels, he might be right about that too, no one knows yet.

>>56266981
>The L3 *is* isolated per CCX.
NOPE

There is a ring buffer and any core can access any slice of L3 with the same average latency.
A 4 die MCM can have any core snoop any slice of L3.

>>56266981
It is worth noting that a single CCX has 8 MB of cache, and as a result the 8-core Zen being displayed by AMD at the current events involves two CPU Complexes. This affords a total of 16 MB of L3 cache, albeit in two distinct parts. This means that the true LLC for the entire chip is actually DRAM, although AMD states that the two CCXes can communicate with each other through the custom fabric which connects both the complexes, the memory controller, the IO, the PCIe lanes etc.

Every core can access the L3 in any CCX, and memory channels are not per CCX. That isn't how a memory controller works.
Why would you go on the internet and just lie?

>>56264142
The GTX 970 is the biggest success story. wew lad

>>56262979
Summit ridge will be the first Zen. Summit ridge is either an enthusiast desktop chip, or a workstation chip. No one is really sure which.

>>56267022
Correct. And there are four cores per CCX. But if there are two CCXes, then as I understand it they are isolated from each others' L3.

>>56267498
They are not isolated at all. Neither are the memory channels.
Hardware.fr, Computerbase, Anandtech, everyone covered this at length. Its even discussed at length on the RWT forum.

Mindlessly regurgitating FUD as if its factual is shitposting.

>>56264310
>No, IPC times clock rate is.
Damn, those 2.8GHz test chips sure were impressive.

>>56265396
I'm sure fused-off Summit Ridge variants will be priced favorably against 6-core Haswell-Es.

>>56267532
where did any article (or the presentation itself) talk about inter-cluster behavior?

as far as I can tell, it was never discussed whether a whole or half Summit Ridge die would be the base NUMA node.

it's frankly probably smarter to only stripe over a single CCX anyway, just to keep latency minimal.

>>56261949
> only have single channel memory per CCX

it does you idiot, otherwise AMD would be die harvesting by disabling the redundant controllers on the 24 and 32 core MCMs.

>and doubly claiming that the L3 was isolated per CCX as well.

it is. you need to run your access to another ccx's L3 through fabric, on intel's uarch the l3 is usually directly accessible through every core. the latency for l3 access on zen is going to be really bad compared to intel's, which is why they're beefing up the size and speed of l1 and l2 cache to compensate.

>Funny how the little shill autist disappeared.

everyone who disagrees with you is a 'little shill autist'? what exactly are your qualifications kiddo?

>>56263828
Copying this

>>56268595

it didn't. he is blindly claiming that AMD's marketing material backs his assumptions when it does no such thing.

>>56264167
>Uhh Yes it is in like 99% of cases
IPC literally means "Instructions Per Cycle", if that doesn't clue you in about how at least one other metric (the famous clock cycle frequency!) is essential to getting at least an estimate of a CPU's performance then I have some bad news for you.

Honestly though, why is this STILL a thing? This board sounds like a fucking broken record when it comes to AMD shit and it's just fucking unbelievable. People here go out their way to literally market anything AMD related, even going as far as that fucking autist that made a 17 minutes long vocaroo recording licking amd's ass regarding async compute, as well as spreading FUD over every single thing intel and nvidia related. Then they hype the ever living fuck out of fucking EVERYTHING amd makes and it never - E V E R - delivers. It has never fucking delivered since the goddamn phenom and people are STILL doing this crap here.

This is just fucking unbelievable, seriously, I just have no words. Part of me wishes this is just the same pe wanting opleto have some kind of fun but I know it's not true and it just makes me cringe as hard as humanly possible when I just glimpse the sight of a poor sad fat fuck hunched over his keyboard defending whatever amd POS he bought as if his life depended on it. It's not even funny anymore,this is just fucking dumb.

>>56263828
AMDfags have serious serious mental problems if they are still waiting to upgrade their shitty CPUs. Imagine being an AMDfag since Phenom II, you saw Sandy Bridge get released which was amazing, you hyped up Bulldozer as the Sandy Bridge killer, waited a year after SB release for it, it totally bombed and ended up being even worse than the Phenom IIs and then you waited for AMD to release an actual upgrade from your current shit, here we are 5 years later and still nothing on the market from AMD matches Sandy Bridge, yet alone the stuff that Intel has released since then. AMDfags have literally waited 5 years to save, what, $50 off the price of an i5? Talk about penny pinching, that's the cost of a night out, and AMDfags call Intel jews. 5 Years of wait with still nothing to show for it, I wonder if AMDfags would have made this choice after Bulldozer was released if they knew what the future held.

>>56260069
>zen's unlikely to be ground breaking
I seriously doubt that. I mean compared to what Intel offers it might not be ground breaking but compared to what AMD has offered it's looking like a rather large step up. AMD just has to find the right price for Zen chips and they'll probably sell.