AMD showed a pic of the Summit Ridge die during their shareholder meeting.
Its a long rectangular die with two clusters of cores.
>>54691592
This faggot better cooked up a miracle here. Intel dominating the market is getting irritating.
>Getting excited for sandy bridge performance in 2016
>>54691389
So how do the individual clusters communicate and execute tasks?
>>54691736
Uncore, the same way everything on die communicates. Everything is always connected by same sort of fabric or another, nothing is an isolated structure.
>>54691389
I love die photos.
>>54691389
The cache looks neat.
I believe each core has its own private L2, and the L3 is shared.
The L3 is easy to identify, but its arrangement is very unique. It appears to have been fabbed into several 1MB blocks, and there is logic sitting in the center of all of it. The square in the center is what I'm referring to. I believe this to be similar to how the Jaguar quad core clusters share their L2.
>>54691389
Pls be good.
I'm tired of Intel monopoly.
>>54692039
Here is the Jaguar cluster for reference.
It looks like Zen is using a far more advanced version of this.
Also worth noting is the memory pads are distributed along the exterior on the die, they are not centered to one location. There also appears to be some interposer IO long one edge.
Two things AMD has consistently bragged about with Zen is the high memory bandwidth for enterprise chips, and the high bandwidth low latency cache.
There is good evidence in the die show alone that these are pretty beefy.
Sweclockers was the first group to reveal any information about AMD's Zen based parts, and we now know most everything they stated to be confirmed. They were first to reveal Zen being an SMT architecture, they were first to reveal it being an 8 core die, and they were first to use the codename Summit Ridge. All of that is confirmed. They also stated that Summit Ridge would be a 95w chip.
If they can hit a base clock moderately north of 3ghz inside of that 95w TDP then they should have something that can compete against intel from Xeon Ds to e5 and e7 Xeons.
General info on the Zen core found in linux kernel patches, with a bit of speculation on part of the author.
http://dresdenboy.blogspot.de/2016/02/new-amd-zen-core-details-emerged.html
Chinese mom is whipping AMD back into shape.
She'll tolerate nothing less than A+MD.
Relevant
Hans de Vries of Chip-Architect has been examining the image.
Its pretty interesting.
>>54691722
It'll be worth it if it's actually fucking affordable. Intel jewing the market to death helps no-one.
>only launching with 6 and 8 core SKUs
fuck
I'm not gonna be able to afford zen until 2018
>>54693694
Mainstream Raven Ridge APUs will be available sometime in 2017.
Summit Ridge is competing against intel's Extreme line of processors.
>>54693714
>APU
>implying I want to waste money and die space on a worthless iGPU
>>54693761
Thats where the market is. Its the reason why all of intel's mainstream chips have a significant amount of die dedicated to graphics. This has been a growing trend for the better part of a decade now.
>>54693781
PC builders who mainly use their desktops for games don't want iGPUs.
>>54693789
5% of the market
Zen is in my opinion the most exciting thing coming out this year. Finally, after what felt like an eternity, AMD is actually going to update their CPU lineup.
>>54693803
>AMD launching Zen as HEDT first
They clearly care very much about that 5% of the market.
>>54693789
If you want to render videos then you pay for a prosumer grade chip like an i7E or Summit Ridge. If you don't need the threads then you get media accelerators and tertiary features that fill a check list that looks good from a marketing standpoint.
You're not getting a beast of a CPU for a budget price.
>>54693789
businesses who build pcs in bulk for enterprise use don't give two fucks about dropping in a gtx1080. less stuff for their it department to worry about
>>54693824
No, the consumer dies are just derived from the enterprise ones.
The highest binnings get saved for enterprise SKUs, AMD is launching Summit Ridge just to get cash coming in while they build inventory.
>>54693825
All I want was 4 Zen cores with no iGPU for a cheap price. Guess it ain't happening.
>>54693847
>All I want is a chip that has no appreciable profit margin for the company making it
Do you not see the issue here?
>>54693042
>>54693870
What else is AMD good for? They exist to undercut Intel. They can't innovate worth a shit because they have no R&D budget. I was hoping they'd at least make themselves useful and release a viable budget alternative to Intel's overpriced yiddish silicon.
>>54693841
Actually they do. A lot of google's machines are powered by graphics cards, because graphics cards are incredibly good at computing.
>>54693789
If there enterprise market apus have trickle down to the enthusiast market, it'll be something all the tiny machine builders will want. Zen is supposed to have support for on board hbm2. But that aside, we'll likely see 4-8 core zen apus with p10 gpus that can be paired with another dedicated p10 gpu. Seriously, if you can get a mid tier Polaris 10 (380(x?) performance) integrated with 8 cores. You got yourself something nice to play with.
>>54693931
Why would I want an overpriced APU featuring a 380-tier GPU when I'm going to be buying a separate card that's at least as fast as a 390 anyway?
Zen is shaping up to be a huge disappointment.
>>54693887
>AMD can't innovate
This is just pure ignorance that borders on violent.
Dockport
HSA
Coherent memory buses
AMD64 extension
HBM
Hypertransport
DVFS
per part adaptive voltage
AMD has more innovative IP on the market than any of their competitors, they have only failed in establishing a solid brand presence through marketing.
>>54693969
>its a disappointment because I'm a whiny poorfaggot
Fixed for accuracy.
>>54693887
Are you some illiterate kid from /v/
What are the odds of Apple putting Zen APUs in MacBooks?
>>54693988
>not wanting to spend $500 - $1,000 on a HEDT chip = poorfaggot
Ok.
>>54694032
Who knows? I wouldn't count on it, but it is possible. For a long time everyone suspected that Apple was the number one driving factor between the Samsung/GloFo deal.
>>54693895
They actually "graduated" from graphics cards a while back. They have custom hardware. Some sort of highly parallel ASIC setup they call "Tensor".
>>54693969
It's just a shot in the dark. In all honestly we could see apus with Vega chips on them. They've already stated plans for massive hppc apus with 32 zen cores, a gpu die the size of 64 zen cores, and as much as 32gb of hbm2 all crammed onto an interposer. Noone really knows just how well the zen architecture will perform, nor gcn4. It's all just speculation at this point.
>>54694103
To add to this, the image op posted is likely their 32 core cpus. All the information for zen states it scales with 4 cores per module and up to 4 modules per die. The 32core cpus are going to be 2 dies on an interposer.
>>54694103
The one enterprise chip that AMD has coming sometime next year is purported to have 4+ TFLOPS computer power. Thats DP, so its roughly equivalent to Fury X in an APU.
Big HPC chip they have planned for later on has a goal of 10TFLOPS DP.
If they don't have a consumer variant then I'll be buying an enterprise board to get my hands on one of these.
>>54694157
The image I posted is 8 core Summit Ridge.
You are incredibly confused.
>>54694163
Nice, keep in mind that they're going back to a unified socket design, so we might see something like lga1366 where you can put enterprise chips in enthusiast boards. You'd just likely lose certain enterprise features like ecc memory. And likely a much lower memory cap. But you'd get the trade off of overclock ability.
>>54694174
I seen you said it was summit Ridge, but it didn't say anything about it in the image, nor did you cite any source. So I just kinda jumped the gun.
>>54694163
AMD APU
>4+ TFLOPS
Who are you kidding buddy? Current AMD APU's have 0.8 TFLOPS and you expect a jump of over 400%?
>>54694230
There are clearly two clusters of cores per die.
Each cluster is only four cores. Very easy to see.
>>54694230
>we might see something like lga1366 where you can put enterprise chips in enthusiast boards.
That definitely isn't happening. Socket AM4 only supports dual channel memory.
>>54694287
Current APUs are 95w and lower TDP parts fabbed on cheap 28nm bulk, made for consumer desktops, and OEM machines like all in ones.
They are not enterprise hardware. The big APUs in question would be 300w~ parts like a Xeon Phi card.
Pull your head out of your ass or stay on a board like /v/ where you rightly belong.
>>54693988
>HBM
To be fair this wasn't AMD, they just bought it from another company who developed. And AMD themselves aren't exactly geniuses for thinking "you know what'd be cool? Faster memory".
HSA unfortunately never took off.
>>54691722
Intel doesn't offer affordable 8 cores with hyperthreading
>>>54693042
She should be /g/'s hero, not that fat fraud "Dr." Stallman.
>>54694348
This is 100% incorrect, stop pulling things out of your ass.
AMD partnered with a company called Amkor to develop HBM, They did not purchase IP from anyone else to do it. Sk Hynix was chosen to handle manufacturing and packaging of he die stacks with an agreement that AMD would pay for all known good die stacks coming off the line.
The entire concept and early versions were entirely done by AMD.
>>54693789
But couldn't they make it so that when you're not gaming or doing graphically intensive things, the iGPU is used like on some laptops? That would be sort of nice because it would allow your system to run quieter.
Anyways, they are surprisingly capable. I've been using the HD530 because I built my PC recently, but am waiting until Polaris comes out to decide between getting a 1070 or whatever AMD puts out. I don't have any super demanding games in my library yet, but it can play Mass Effect 1 on high 1080p 60fps and CSGO on high 1080p 60fps with some AA on even. I haven't tried anything else yet though, but I wouldn't be surprised if it can handle modern games on medium/low settings.
>>54694418
FYI this picture is from 2011. AMD had prototypes as far back as 2010, and the working concept is even older than that.
>mfw retards in /g/ just say shit and expect not to be called out
>>54693789
I use the iGPU's for dual monitor & QuickSync for improved encoding over NVEnc for things like Steam In-Home Streaming. They're pretty rad.
>>54694394
How about stop being poor?
>>54694508
It's like a thousand dollars for current 16 thread intel chips
>>54694520
>ou don't have to buy the latest chips. There are surely really affordable xeon 8 cores for a low as 200 bucks, which will still trash any comparable AMD CPU, probably including zen, once it's out.
>>54694560
Sorry fucked up the greentext. Get a xeon, something like a E5-2670.
>>54694310
>am4 will support dual channel only
Can you post a link to this? Everything I've ever found is that zen will support up to 8 channel ddr4, but that specifically is only on articles for opterons.
>>54694508
If zen pushes the haswell level IPC everyone is speculating it to have. Why would you spend twice as much for the same performance on a larger form factor?
>>54694608
Only a few enterprise Opterons support 8 channel configurations. A 4 die MCM with 32 cores would have 8 channels.
Look at the memory pads on die for christ sake.
8 core/16 thread = dual channel
16 core/32 thread = quad channel
32 core/64 tread = octa channel
>>54694593
That only runs at 2.6ghz base, and lacks things like 10 gigabit and USB 3.1
I am now of the opinion that these "leaked" slides aren't fake, but they are in fact clever marketing put out by AMD themselves. They contained far too much correct information to be guesswork, they're just peppered with a few inaccuracies. The perfect piece of guerilla marketing to start people talking and keep them talking.
Most things on this slide in particular are legitimate, though Zen does not have 256bit pathways in its FPU, and the FPU is 4 wide instead of 3 as per kernel patch notes.
>>54694593
What's a cheap motherboard look like for that? I could use something cheaper than a 5820k for running VIRL & some other VM's
>>54694838
Some of them are very well put together. They may have been legit slides that were taken months in advance by someone at AMD and the details were changed as the FAD got closer
>>54694838
Zen isn't going to have that shared FPU bullshit like the Piledriver is it?
>>54694593
xeon is a meme since they can't be overclocked
>>54694916
Zen cores are independent like everything prior to bulldozer
>>54694916
The FlexFPU from the Bulldozer family is done with, each Zen core has its own independent FPU.
Zen marketing has really been kicking into overdrive on /g/ over the past month. A lot of astroturfing going on.
>>54695029
Its rare to hear good news from AMD so when they release something people swarm all over it.
>Expect /g/ to be shitting their pants about Zen
>They're pessimistic like everyone else
Why is no one hype as fuck for Zen? It's not even a fanboy thing, Zen kicking ass is literally good for everyone. Based on literally everything that has been seen, there should be nothing but cautious optimism for this CPU yet I still see so many people dismissive of the idea that AMD could ever make anything good.
>>54693988
The truth is NVIDIA is the capitalist who brings innovation because people know that if they innovate there, they will get big bucks.
Socialist AMD comes and ruins NVIDIA's inovation money grinding by trying to make them free for all(open source)
>>54693042
Kinda cute desu.
>>54693789
I'd appreciate having a minimal radeon igpu along with the hardware accelerated codecs so I can still use the pc even when the discrete gpu dies.
Future vulkan/dx12 games can even use the extra igpu cores to offload compute tasks.
>>54695274
Its a damn good thing that fanboyism isn't rampant. Hyping Zen to hell would be a terrible thing to do, just look back to the Bulldozer threads we had here.
Zen isn't going to perform on par with Skylake, and anyone claiming otherwise is only hurting AMD by inflating expectations beyond whats reasonable and factual. A 40% IPC uplift from excavator isn't hitting any home runs, and Zen won't tackle intel's enormous FPU prowess.
Zen will be around Ivy Bridge - Haswell in most integer workloads, it will probably outperform them in some metrics, and its going to fall behind in heavy FPU and vector stuff. It all comes down to how well the parts clock.
All of this
>HURR BASED KELLER AMD GONNA SMAESH INTEL XD
bullshit needs to stop. We should at least pretend to be sane adults here.
>>54695274
I'm just waiting to see how it really performs and the price point. I already know I'll need a new motherboard and memory and to upgrade to Windows 10 to fully take advantage of it. I don't need hype I want numbers.
>>54694508
Jesus christ you know laurance rockefeller invested in intel and apple when they were just startups?
>>54695459
See like this asshole right here, exactly what I mean.
First of all, no one is saying Zen will compete with Skylake in performance and no one ever said that, all this "HURR BASED KELLER AMD GONNA SMAESH INTEL XD" is your own dumbass putting words in people's mouths. But what it represents is AMD catching up and actually providing a compelling alternative to Intel's near monopoly. Bulldozer was a different beast entirely, AMD seems to have actually have their shit together in terms of how this thing works.
Ivy-Bridge to Haswell performance is fucking amazing, and it's exactly the performance we were expecting. Skylake was barely an upgrade from those chips anyway. If we're looking at a chips with near Haswell performance, Skylake's feature set, eight cores, and a lower price of both the motherboard and processor, then I say we do actually have a potentially compelling product on our hands. That's worth talking a little more optimistically about.
Also if you're going to tell people to be sane adults quit with the HURR DURR comments. This isn't /b/.
>>54693042
Man, she cute
>>54695508
That's what I'm waiting on too, but based on the numbers that have leaked it's hard to not get excited. Of course leaks don't mean shit if it doesn't actually perform that way on release.
>>54695650
>First of all, no one is saying Zen will compete with Skylake in performance
Yes, yes they have. All over the web retards are saying this, and someone in every single Zen thread parrots it. A bunch of children like yourself on reddit took some Passmark results and multiplied them to try and simulate Zen, and took that figure and said Summit Ridge would compete against Skylake.
>all this "HURR BASED KELLER AMD GONNA SMAESH INTEL XD" is your own dumbass putting words in people's mouths.
No, you shit eating child, that is literally the meme that everyone here spouts. Find a single Zen thread where someone doesn't talk about the "certified shitwrecker" and how hes a miracle worker. Every single thread is filled with children, like you, who do nothing but harm AMD with their astroturfing bullshit.
Your post did nothing but highlight how fucking stupid you are.
>>54695696
Well that was a mature response, I've been reading Zen leaks/articles/threads since the beginning and I can't say I've seen any good proportion of people claiming it'll beat Skylake. Everything estimating performance put it below Skylake, and more people have been comparing it to Sandy Bridge. Keller is also a talented man, and he likely did have a substantial influence on Zen's development, but that's hardly a "meme" but no one is singing his graces that much at all especially since he left AMD.
Your post did nothing but prove what an immature man-child you are.
>>54695760
>even more bullshit
You immediately launched a personal attack against me for telling people to have reasonable expectations for Zen's performance. You then proceeded to dismiss and outright lie about perveiling rumors on the web. Acting like you're unaware of popular memes just makes you sound like this is your first day on 4chan as well. You're a complete shit eating child, hired shills have more tact than you do.
Pic related is the issue. This is exactly what everyone believes, its been going this way for month. Children just like you regurgitate this type of nonsense constantly. I argue against this bullshit, and you have a problem with me? Do us all a favor and go get murdered.
Shills like you do more harm to AMD than their hapless marketing department ever could. Retards like you created the Bulldozer media fiasco by constantly hyping everyone up, constantly blowing smoke up eachother's assholes. A lackluster CPU isn't the end of the world, it only got attention for so long because shit eating children spent a year shitting all over the internet about how it was going to be the best thing ever.
We'll I've been paying passing attention to the market for years, and I can assure you AMD will miss all their deadlines (twice), release a bugged chip, and then have gimped the whole thing with a cost cutting compromise.
>>54695886
Bulldozer wasn't that bad, though. For multithreaded applications, it was amazing. The problem is that nobody developed multithreaded workloads, and if you were paying attention during the Bulldozer hype, people were talking about that. After the launch, discussion about multithreading basically died while everyone and their mom shit themselves over Intel having better single core performance instead of asking why we weren't getting proper threading for our quad cores. Fast forward the better part of a decade, and it's AMD that's finally dragged the industry kicking and screaming into doing what they should have done years ago. Meanwhile, Intel continues to eek out incremental single core gains while not doing much else and somehow everyone believes they're doing a good job.
AMD wasn't hurt by its own hype, AMD was hurt by idiots, like you, misunderstanding market interactions. It's exactly the same as the 90s when Microsoft's blue screen errors were rampant, not because W95 and 98 were bad, but because hardware manufacturers refused to implement fixes for hardware conflicts so that they could build walled gardens. The difference is that AMD couldn't throw money at the problem like Microsoft did.
Speak from ignorance but call people children. Lame. You showed up saying that it's good that there's not much hype for Zen but also fuck hype for Zen. You're self inconsistent. Fuck off
>>54696171
>shit eating retard can't help but dig a deeper hole
Bulldozer wasn't just bad, it was terrible. Energy per op is horrendous, and its made even worse by the insane leakage displayed from being early silicon on GloFo's bugged 32nm PD-SOI process. The Zambezi family was originally meant to target clocks near 5ghz inside of a 125w TDP. The FX 8150 hit a 3.6ghz base clock. Astoundingly short of its projected target frequency range.
Later Vishera improved a bit, but still could only compete in one specific subset of workloads vs intel's Sandy Bridge and Ivy Bridge i7s. Heavily threaded integer bound ops that didn't involve a lot of cache hits. Not only were front end issues holding the arch back, but the low bandwidth and high latency of the L2 and L3 brought considerable performance penalties of their own. Bdver optimization guides have specific rules just for avoiding cycling ops through the cache.
Half of Steamrollers IPC gains were lost when L2 latency increased even further in a trade off for increasing bandwidth for APU performance. You know how they tried to fix this in Excavator? They cut the L2 in half to decrease access time, not certain workloads show a performance regression because L2 is hit too frequently despite being somewhat faster.
None of this is good, and there is no excusing it.
On Zen in this thread I have been nothing but consistent because I stick to posting facts, something you have no command over. Su said it herself when she stated Zen Opterons would address 80% of the market. They're going to be irrelevant in 20% of workloads that make heavy use of FPU ops.
Zen isn't going to be a home run, and nobody should delude themselves into thinking otherwise. You bitching because people aren't foaming at the mouth with rampant fanboyism is disgusting beyond words. People with reserved expectations are completely level headed. You are not.
Kill yourself, child.
>>54696344
This is a perfect example:
http://forums.anandtech.com/showpost.php?p=38198889&postcount=996
In X265 video encoding an Excavator core is 88.8% slower than a Haswell core per clock. A 40% uplift over this isn't worth getting giddy over.
Hyping this up only leaves a bad taste in people's mouths when the chips launch and reviews come out. Exactly like what happened with Bulldozer.
>>54693042
I like her because she's not an attention whore. Unlike those sjw woman "CEOs" whose whole purpose is to meet a quota and bring companies down. Lisa is humble and confident in herself.
>>54696454
Su is actually an accomplished mathematician and engineer, she even has worked in marketing before, she isn't comparable to any tech SJWs. With that kind of pedigree shes among the best of te best in Silicon Valley. People with real accomplishments don't rely on crutches like gender.
>>54691389
Anyone been able to deduce the die size based on this image?
>>54696493
hard working taiwanese versus speculative americanomercials
you know taiwan runs goods and the us runs the price
>>54696539
Nope, theres no reference to go off of.
My own personal guess has always been around 150mm2 and I'm sticking to it.
>>54696344
>They're going to be irrelevant in [...] workloads that make heavy use of FPU ops
Does this mean they'll be shit for games?
>Sandy Bridge
>Summit Ridge
These codenames are phonetically similar.
AMD confirmed for only targeting Sandy Bridge performance. It's over. Zen is a big fat flop.
>>54696567
I don't know of any games which are explicitly FPU bound so I can't qualify that.
Summit Ridge isn't going to perform worse per clock than Vishera or any current APU. Thats the only way I can answer your question.
So should I wait for zen instead of buying a 5820k or a 6800k when it comes out? Not that it'd necessarily be better.
>>54692039
neat
>>54692083
Guess which CPU Jim Keller first worked on when he hopped back on the AMD train?
>>54696567
FPU OPs are generally run on the GPU given that you're not Bethesda and implement x87, which was deprecated some 20 years ago.
>>54697468
He actually worked on Excavator for a short amount of time before starting with Zen.
I'm not sure he even touched the Cat-cores as these were pretty damn decent on their own already.
>>54697504
>I'm not sure he even touched the Cat-cores as these were pretty damn decent on their own already
He worked on both Excavator and Puma. I'd say Zen has more based off of the Cat cores because Keller knew it was a good starting point.
>>54693494
>FM3
That's probably faked. By March 2015, AM4 was the only nomenclature AMD was using for their new unified socket.
>>54693662
Wait a sec, will each core cluster have their own memory bus?
Does that mean that each core will only get a single channel? Or will both cores get dual channel?
Neither of those makes sense, so I'm hoping that those are connected via a common bus between the cores.
>>54693887
They've been innovation pretty gud senpai, moreso than Intel I dare say
>>54696171
bulldozer WAS that bad. especially with all the FUD and hype leading up to it with AMD claiming it would be the best serial architecture we had seen yet.
>>54697556
No, FM3 has been an internal development name for AMD for years.
That slide is legit. AM4 has never been used internally as its the consumer name, not the codename.
Even FM2+ was called FM3 for a long time.
>>54697556
early leaks said FM3 with unified socket for cpu/apu, amd could've easily just used fm3 internally and changed it to am4 later down the line for marketing/clarity purposes.
>>54695347
It's like you like being fucked in the ass with no lube
>>54693042
>A+MD
A+ post, anon.
>>54697662
Commies get off 4chan
>>54693988
half of those are inferior copies of intel tech, the only good thing on your list there is amd64, otherwise we'd all be stuck with itanium or plain old x86.
>HBM
HBM is stolen technology, it's equivalent to HMC except they ripped out the horizontally connected memory, probably because it wasn't ready to produce at the time HBM was 'designed'
>>54697744
I'm not a commie, I'm capitalist by heart my man
>>54696344
Neck yourself neckbeard
You flaunting buzzwords and meanlingless insults only serve to harm and defame the community
Fucking pompous piece of shit
I bet you're in your thirties and still living out of your mother's basement feeding off of her welfare check, you neckbearded autistic fuck
Go outside and apply that bullshitting ability with an actual job
>>54697819
>being this much of a butthurt retard
Reddit is that way >
>>54696344
>>54695886
>>54695696
These are the most autistic comments I've seen on /g/
And this is /g/ we're talking about here
>>54695886
>it's that "i'm a cpu engineer, really i am" troll again
>>54696344
kek
this is pathetic
@54696344
Came here from reddit
Wow, what a wanna-be nerd
>>54695696
>>54695886
Glad you didn't pop out of my pussy
>>54697649
Which makes sense as Kaveri has both a DDR3 and DDR4 memory controller.
The DDR4 side has never been used though.
Seems like FM2+ wasn't actually supposed to exist, but something couldn't be finished in time so it existed as a stopgap.
>Kaveri dies would likely have been sold for FM2 and FM3, thus the dual controllers...
>>54695029
>I didn't post it so they must be shills.
I'm genuinely interested in this shit.
Also still waiting for a standalone Xeon Phi for my esxi box, intel will surly deliver.
>>54697884
are you in London
>>54697894
There were patents pointing to a variant of Kaveri utilizing GDDR5, or at least some variant of it. Kaveri and Carrizo share the same 3rd party memory controller.
The Stilt has done a good run down on it, and why hes so dissatisfied with it.
>>54697905
Hope you've got deep pockets. Even the low binned Xeon Phis have been over a grand. New cards are going to be ridiculous.
>>54694032
they already confirmed polaris for macbooks, so probably high
>>54698057
I'd say they have the best chance of ending up in the new Mac Pro.
>>54697949
>New cards are going to be ridiculous.
I don't want a co processor. I want a phi (C)PU, you know something I can boot off of but with 72 cores + 2x hyper threading for 288 logical cores and the 100 Gbps Ethernet fabric and the 36 pci 3.0 lanes. They are fucking jewing us on ram though, the 16 gigs of L4 is nice and all but only letting us have 384GB of system ram really sucks ass with this many cores.
I want to offer at least 72 dedicated core vps's per 1u rack space.
>>54693042
Can I call her mom?
>>54698057
>>54698077
Googling AMD + Apple shows some october 2015 rumors of Zen being used by Apple. But where? imac? mbp?
>>54691389
YESS
cant wait to swap to zen intel's shitty chipsets piss me the fuck off, not touching them again
>>54693042
10/10 would let save company
>>54697819
Kill yourself retard. You just got utterly obliterated and outed as a fool who knows nothing of the shit you blab authoritatively about.
>>54698439
Just ignore it. Someone who samefags that hard when they get told either has real mental problems or they're a real shill. Given AMD's tight finances I doubt they're paying for shills.
>>54693042
She's an engineer. A motherfucking engineer. About fucking time AMD got an engineer as a CEO.
Is 8 cores going to be enough? Intel can easily do a 12 core i7 for $999 if they feel like it.
The 8 core zen will have to sell for $500 USD at most. The server ones might be more interesting, if they can make a 16 core MCM version for $1000 or so it would sell well.
>>54693042
I SHARR BRING HONOR TO FAMIRY
>>54699555
with that post number you certainly did
>tfw I need to upgrade because my mobo died on me and Zen is still 5 months away
I'll have to fall for the skymeme but I hope AMD catches up this time so my next upgrade can be Zen.
>>54699555
>>54699566
>>54693662
The two 4-core clusters are supposed to have a new interconnect fabric between them, an article following the project suggested 100GB/s territories.
Having the clusters offset like that is the first suggestion this isn't exactly Zen.
It also completely goes against any marketing slides seen about zen comparisons.
Each quad-core cluster should share half the total L3$, with the interconnect bridging the L3$ units running at half-bandwidth. I imagine the uni-directional bandwidth is 51.2 GB/s, twice as fast as HT3.1
>>54691389
the new AM4 socket will be rectangle?
Problem is has AMD fixed their branch prediction and cache miss/latency issues?
AYYMD HOUSEFIRES
AYYMD HOUSEFIRES EVERYWHERE
Question, why are caches always so big on the die while their size hasn't changed all that much(6-8MB L3) for quite a while
Aren't caches susceptible to die shrinks as well?
>>54700641
Cache units are large (generally relative to their connected logic units) arrays of what is basically RAM. They need high power lines to provide a stronger, wider signal. Cache traces (trace = wire in a chip and on a board) can not be shrunk as much as logic units, and logic units themselves have size bottlenecks with their own power and data traces.
Reading up on the power advantages of HBM would tell you all you need to know.
>>54700570
Are you retarded?
>>54694397
/thread
>>54694394
It does offer 6 real cores with hyperthreading though, for a reasonable price. :^)
>>54693042
She looks like cooking mama got out of the kitchen.
>>54693988
>AMD has more innovative IP on the market than any of their competitors
Although I agree with your other points. I don't think you know who IBM are. AMD would have the lionshare of decent patents that pertain or have any relevance to gamen. But thats it. They're a top tier comapny with a monstrous legacy, that tripped down a hill in 2008 and kept snowballing till now.
>>54699253
>Intel can easily do a 12 core i7 for $999 if they feel like it.
no they can't, price of being independent and HUGE at the same time
their fabs not as efficient as they would want to
>>54700550
No, it definitely is Zen.
>>54699601
Same thing happened to me last week. My old core 2 machine bit the dust
Except instead of falling for skymeme I just got the cheapest mobo cpu combo fry's had. I still intend to upgrade to zen
>>54702701
Seems pretty retarded to have to pay @$100 extra just to have Zen, sempai
>>54703379
I could afford it , plus, I don't intend on keeping them. I know a guy I could sell it too even if I only make half of what I paid back.
>>54697750
>HBM is stolen technology, it's equivalent to HMC except they ripped out the horizontally connected memory, probably because it wasn't ready to produce at the time HBM was 'designed'
Are you retarded?
Stacked chips have been on the horizon for at least a decade, and memory was always gonna be the first thing to get vertical integration.
And that's pretty much where the similarities end.
HMC uses potentially multiple fast and narrow serial buses per module for daisy-chained or mesh networked fanout, while HBM is designed for a slow, massively wide parallel bus over a silicon interposer to save IO power.
>>54691389
Just beat my Fiery Bridge on more cores and my next build will be yours, AMD. Intel needs a wakeup call.
>>54704626
>Are you retarded?
He very probably is. You can't even count that as shitposting when the average poster making those comments actually believes it. This board is just full of retards.
>>54705327
Idle hands.
>>54705587
fuck off pajeet
>>54702303
Proof that it isn't some throwaway company stock image?
Better phrased, proof that it's legitimate.
>>54700641
>Cache units are large (generally relative to their connected logic units) arrays of what is basically SRAM*
SRAM is made up of flipflops which allow it to retain its state indefinitely as long as power is not disconnected from it. DRAM which is what we use for system and graphical RAM is 4-6x smaller than SRAM of equivalent capacity. The catch is that DRAM doesn't hold its state for long. DRAM is basically a transistor and a capacitor, the capacitor holds the transistor open while its charged and when it's empty the transistor closes.
>>54705891
It isn't a stock photo, and its a totally unique die like nothing else ever made. There is no other CPU in existence with a die that looks like that.
That is the Summit Ridge die.
>>54705957
And I quote
>"What is basically RAM"
>RAM
>Not specified as DRAM, VGRAM, or SRAM
It was a dumbed down response, thanks for noticing.
>>54705969
Whoa! Nice statement. Care to back that up with evidence?
>>54705994
>I'm a tech illiterate retard who can't recognize a die shot
Hilarious.
Yeah AMD just made a totally fake dummy die for use in their investor presentation on a segment specifically about Zen. Its totally fake guys.
>>54706032
Die shot is nonsensical compared to everything that is not an AMD APU or Bulldozer/Piledriver which I have seen for the last fifteen years.
>>54706032
I wouldn't put it past AMD to release faked images of a non-existent product, but this is pretty late in the development cycle for them to do that.
If it's not Zen, a Construction, or a Cat chip, then I don't know what it is because it's definitely not Intel
>>54706173
There is nothing "nonsensical" about it.
>>54706248
You have no idea what you're looking at here
We can't even see how the two four-core sets are stitched together (where is the L3 interconnect), and this is possibly the most inefficient layout to connect two quad-core CPU modules (why are the blocks of L3 not as close as physically possible? Why are there cores in between the sets of L3?)
This is a graphical mockup using Excavator modules (Excavator being narrower than BD due to HDL libraries and using smaller but faster caches)
>>54706576
You're such a tech illiterate retard its adorable.
Front end features do not carry signals, that is BEOL. Interconnects are back end, just like power delivery. If you can't understand something this incredibly simple then there is absolutely no hope for you.
There are no cores in between the L3, that is almost certainly the cache interface similar to the Jaguar compute unit.
Look at the scale in the retard mock up you just made. Look at the peripheral IO and memory pads on the Bulldozer die, now look at them on the Summit Ridge die.
That isn't even a picture of Excavator modules. Jesus Christ you're legitimately retarded.
>>54706659
You might actually be retarded. Physically, provably deficient.
>>54707309
You made a total ass of yourself and got blown the fuck out.
Stop trying.
>>54698548
>A motherfucking engineer
>https://www.youtube.com/watch?v=bmrqPJigiVc
>She became an Electrical Engineer because she thought it was difficult
>Born and raised by strict Asians
How can Intel compete
>>54707320
Only in your fantasy land.
You'll choke on your idiocy in three weeks.
>>54707629
You literally just said a Bulldoer die was a picture of Excavator modules.
Nothing more needs to be said.
>>54707660
Yeah, a bulldozer shot I compressed to Excavator size using both Steamroller>Piledriver and Steamroller>Excavator comparisons as a guide because it (BD) was the only die shot I had that was directly physically comparable to this supposed "Zen" shot.
How fucking dense are you?
>>54707867
You just freehand scaled a full Bulldozer die down in ms paint, that is totally arbitrary. You don't even know what the module size on each node is for the BD family of chips.
Now you're just shitposting to save face like true tech illiterate retard fashion.
You don't know the difference between FEOL and BEOL
You don't know how interconnect fabrics work
You don't know how caches work
You get completely blown the fuck out and respond with
>b-b-b-but ur retarded even though I can't defend myself any more
>>54707969
>freehand
No, I measure down to the pixel from official sources, calculate the size ratio and finally the exact size in pixels, and resize the image.
>shitposting
Ok, kid.
>Don't know the difference between Front-End and Back-End
Ok, kid.
>All this other bullshit
If this is what Zen is it's completely shot out from the start. The layout Is. Not. Good. The MT performance will suffer. Cache heavy code will suffer when data bounces from one L3 to the other.
And finally, randomly showing a Zen die shot months from announcement, with no plans made public to even show the thing off at a scheduled date, is so outrageously wrong it hurts.
AMD has not operated this way at all this year or last, and they won't start with Zen.
You're incapable of piecing together facts shoved in your face.
>>54708077
>making assumptions based on literally nothing
>further proving you have literally no idea how interconnect fabrics work
32nm Bulldozer and Piledriver are 18mm2 minus L2
Excavator is 14.48mm2
You didn't pixel count anything, you're just flat out retarded. Tech illiterate retards always try to desperately save face when called out.
>>54698548
hate to break it to you but she's been running the show for a long time now.
So, should I upgrade from this Xeon to an 8c Zen when it comes out, or should I stick with my original plan to go to a cheap 12c HW-E Xeon?
>>54699601
I still have an Intel Core 2 Quad Q9450 from 2008. Should my next CPU be Zen? Don't let me down /g/. I only do this once a decade.
>>54709472
zen isn't likely to perform better than sandy-ep, so no.
>>54710272
yes. wait for skylake-e if zen flops.
>>54693662
link?
