It's fucking over, AMD is going all out, HBM on Zen, mainstream APUs with 4 cores and GPU with HBM, performance CPUs without GPU with 8 cores.
https://twitter.com/blueisviolet/status/682619647372296193
>Integrated GPUs will now not be shit
>270X performance on the iGPU
Finally I can have a laptop that's not 7 pounds, but has some graphic horsepower while retaining battery life.
I assume iGPUs will go up to 1200 shaders for the first gen.
>>52175761
Maybe we can see AMD take back the igpu crown. That's if they don't go bankrupt in 2016.
Gonna be a long year for
>>52175761
whats an x1?
>>52175947
Shitbox one, that home entertainment system with 2 games.
>>52175805
The problem will as always be TDP.
Hope AMD gives us a way to unlock TDP over 95 because when the cores and GPU run at full tilt one of them is going to suffer performance.
Now the question is, is HBM better than Intel's eDRAM?
>>52175805
Sorry to break it to you buy current apps could be used in lightweight laptops but companies refuse to make them that way because jewel $$$.
Plus a lightweight app laptop can't be marketed as ultra book which is the new fad.
>>52176047
No, not even close
>>52176560
Why do you think so?
>Onion
Will it be ogre for Intel?
>>52175761
I just bought 150 shares of AMD at $2.90 a share yesterday.
I've lost $4 so far.
Hopefully this means another bump in their stocks. I wanted to invest when it was $1.60 a share... but I listened to half of /g/ that thought they would fall through and I'd lose my ass....
Fucking shills, man.
>>52175980
The PS4's SoC operates in a 95w envelope. That 28nm planar chip has 20CU, 18 being active clocked at 800mhz, and 8 Jaguar cores with a big sloppy interconnect sitting between them. They could fit all of that in a cheap 28nm chip, they'd have no problem fitting an IGP that size along with 4 Zen cores in 14nm.
The throttling issue that desktop APUs have isn't even one of TDP, its a behavioral one. Desktop Kaveri chips will throttle back under heavy CPU+GPU workload, but they're at no risk of exceeding TDP. If happens even if you undervolt the chips. Its something put in place for a locked mobile chip that just didn't get disabled for the desktop parts. Shame because it affects all benchmark performance figures.
>>52176560
>>52176047
HBM must have some advantages over edram, else there is no point for amd using it. edram has been available for years already
>>52178302
Ignore the retards.
eDRAM is fast, and can provide high(bi directional) bandwidth, but it can't offer the capacity really required by a GPU. When you're working with a small buffer you end up constantly wasting cycles flushing, and over writing data.
Intel's 22nm Iris Pro equipped parts had 128MB of eDRAM, and compared to the rest of the die its absolutely massive. Scaling to 14nm reduced over all die size, but the eDRAM proportional to the rest of the logic is still quite large. Intel is putting themselves in a position where they can't win out long term since you can't continue this trend. They plan on upping their L4 cache size to 256MB in future chips, and that package is going to be considerably larger than everything else on package.
>>52178390
Why doesn't Intel just use HBM or HMC, which they already have in their new Xeon PHi's
It looks like a win on all fronts, buffer size, packaging, power consumption, physical size and probably even bandwidth, I don't know the latencies of either but it shouldn't be all too much different, but it shouldn't really matter since GPUs were made to deal with high latencies.
Intel will be releasing Broadwell-e in a few months.
>5% ipc increase
>significantly lower power consumption
>$380 for Hexacore, $580 for octocores, $1000 for decacores
How can AMD even compete
>>52178531
moar cores for moar money!
amd will only win if they price zen 100$ below skylake if they match skylake price, then gg amd people will still just buy intel due to at least knowing it works
>>52178522
They may switch to HMC or some other memory technology in the future. They likely chose eDRAM since deploying HMC would have been a larger undertaking. Would have required a new socket and motherboard.
HMC modules use a separate controller to simplify signal routing, and if you put them on the motherboard then everything has to be designed around that. If you put them on the CPU package like how HBM is used, then you'd need to create a much, much larger package, and you'd need a new socket to fit that which would still require a new motherboard design.
>zen apus with hbm
HOLY FUCK
>>52178589
It will definitely be in the same ballpark in terms of price. AMD cannot afford to undercut nor can they win an undercut war. They'll price it the same as superior competing products and hope people will be dumb enough to buy it.
>>52178602
Also worth noting that HMC uses a lot of power. The HMC Consortium quoted a 4 die 128GB/s module as consuming over 11w.
>>52178647
>>52178589
The high end Raven Ridge chip will probably be around $250, maybe more.
I expect 8 core Summit Ridge to be priced around $500 for top binning.
Summit Ridge is competing against the i7E, Raven Ridge is the mainstream chip competing against the i5s and i3s.
>>52178685
>I expect 8 core Summit Ridge to be priced around $500 for top binning
lel you can buy 8 core sandy/ivy xeons on eBay for $100
>>52178647
They are changing the game and you suckers ate sorting
>>52178729
No one bases the price of new products on what used things sell for.
>>52178761
Yea no shit. Got any other wisdom to drop on us Great One?
>>52178729
What does that have to do with anything? I can buy your mom on eBay for 100$
>>52178788
Remember to Zen because reasons
>>52175925
Honestly Intel would probably just throw them just enough cash not to die so they don't need to split due to monopoly laws.
Or they'll try and bullshit and say ARM is a competitor and them being the only x86-64 maker is somehow not a monopoly
>>52178390
Well no shit, it's another billion transistors...
>>52178894
Monopoly aren't illegal
>>52178924
I'm gonna be a bitchain baller gpu pomp
>>52178945
American education?
>>52178973
if u start a fag company now the gov won't break u up
>>52178945
Right on brah. I got it for Christmas
>>52175761
fuck I guess one day soon a SoC with built in GPU will have better performance than a system with discrete GPU. That low latency, high bandwidth.
>>52179197
Never, since there's simply not enough room or TDP OR Yields to fit 600mm^2 of GPU on a SoC.
And latency will never make up for that huge loss in overall power, especially if something faster than PCIe comes along.
>>52179197
Nvidia
>>52179197
A larger form factor will always allow you to draw more power and dissipate more heat. New interconnect fabrics will slowly erode at the latency between distant processors in a system as well.
An integrated solution will never outperform dedicated discrete components.
>>52179260
You're on the right track, but only half right. Large die yields are not a factor when working with an MCM.
>>52179336
Why do you assume an MCM? GPU and CPU physically share the die, what's the point of splitting them up? For yields? How do you deal with the fixed function hardware on the GPU without losing efficiency?
>hbm for apus
Did they really have to do this though? The 80$ a8 7600 already kicks intels ass price/performance wise. This is like executing someone with a cannon.
>>52179412
Because AMD is going to be employing MCMs for all of their enterprise APUs. Multiple dies on an interposer, there is no technical hurdle in doing this.
The concept is called die stacking, and they'll eventually be fabbing each component on a process specifically tailored to the given IC. AMD and GloFo have given half a dozen presentations on it in the past couple years.
>>52179896
AMD is circling the drain. Nuking intel from low earth orbit in a manner like this is about all they can do to remain relevant. If they don't slap intel's shit this time around then they're going to fade into oblivion in a matter of years.
>tfw /g/ doesn't believe in AMD
>>52178894
Intel would not save AMD. If AMD goes bankrupt, everyone is fucked.
Monopoly laws are pretty relaxed in the US. All Intel will do is make the claim that they do have competition via ARM, which exists in a shitload of devices already (pretty much all mobile devices). That alone would prove they're not a monopoly. And at that point they will then throw all their weight full-force into the mobile sector, and desktop CPUs will stagnate for the next ten years until some meaningful third-party development happens.
>>52179896
So? You still get better performance with an i3 + a dedicated gpu.
Why do people have to be so fucking poor, holy shit.
>AMD announces X
>"people" educate the public how good it is, goes on a lengthy discussion about how sophisticated and state of the art it is, and general hype
>it ends up being slower than Intel/Nvidia products from years ago
Anybody noticing this?
It happened with bulldozer, trinity, bobcat, and hbm
Am I in a legitimate paid shilling thread?
>>52180023
Yes.
Currently you get better performance.
But with a smaller die, more efficient CPU design, that leaves a lot more potential for the GPU, and with HBM on die, bandwidth would no longer be an issue.
And we've yet to consider what is capable once AMD start putting their yet to come Arctic Islands 14nm GPU on their yet to come Zen based APUs.
Nvidia or housefire
>>52180058
This is honestly the last chance we are giving AMD. Zen flops and most of us cry our woes and accept intel as our lord and saviour.
To be fair though, AMD wouldn't suck so much if devs stopped being lazy niggers who force their games to run on one, maybe two threads efficiently (looking at you arma 3).
>>52180058
>AMD announces X
>People rationalise and think about future potential and how it will play with other technologies
>Shitposters parrot it
>Other shitposters latch onto it
>"hype" starts building
>More shitposting
ftfy.
>>52180058
People on /g/ like to joke about Jews calling gentiles "goy", but it's true.
On the sabbath, older Jews would call for a "sabbath goy" (basically a gentile that would do work around the house in exchange for money) to light their fires, move things, etc. As a child of 7 or 8, my grandfather (who was also jewish) would heed their call and do things for them to take their cash. He didn't really care because he was already an atheist by that age.
I liked my granddad.
Anyway, not at all related to the thread but I thought it would bring some levity to the discussion since AMD is in such a bad place right now. Their balls are in a salad shooter.
>>52180058
>>it ends up being slower than Nvidia products from years ago
Which ones are those? Pretty sure the r9 290x is comparable to *current* generation novidya cards. Are you getting AMD and Nvidia confused?
>>52180058
Ah to be young and retarded, imagine what would happen if you actually knew what people here were talking about.
>>52175761
Zen will be on APUs in 2017-2018. It will be to late again.
>>52180347
Clearly says 2016 there.
But I'm guessing Q4 2016.
>>52178009
Should have bought them at $1.6 familia
>>52180002
Microsoft wouldn't be happy if AMD went down.
Trying to sell proper Windows on expensive stagnating Intel monopoly x86 against the rising tide of becoming ever cheaper and more powerful ARM platform that only their most core party software runs acceptably on would cede a lot of ground to Linux.
>>52180347
Raven Ridge is due Q1 2017 according to the client roadmap in the leaked slide deck.
>>52180367
The image is from a phd thesis, not a technical document from AMD.
>>52180347
Nah, at worst we can expect Zen to arrive at Q4 2016. It could be possible we will see some integration of Zen around Q2 2016.
>>52180413
ARM is garbage. This abomination should have never existed.
>AMD still can't compete on CPU market
>AMD still can't compete on SoCs
>Mobile devices (laptops, tablets, phones) with AMD are the rarest
Oligopoly on oxygen support is just as bad as monopoly. I hope AMD dies soon and Samsung / Qualcomm joins into battle.
>>52180522
And they join, with those license? Dumb shitposter.
>>52180413
>Microsoft wouldn't be happy if AMD went down.
That doesn't matter. Microsoft is increasingly irrelevant and they don't have the pull they used to. You'll notice that despite Microsoft wanting a viable ARM platform, they still base their Surface tablets around Intel x86 processors.
Again, if AMD goes bankrupt, you are all fucked. At best they will be acquired by Microsoft or Samsung and you'll end up with retarded shit like Exynos and big.LITTLE, most likely embedded and closed up to their own products. At worst AMD be parted out and you end up with Intel and Nvidia fucking you in the ass for the next decade.
>>52180522
I see laptops with AMD APUs all the time.
>>52180533
Hopefully there will be new small firms that joins in the market.
>>52180522
Do you seriously think that AMD ceasing to exist will magically make Samsung and Qualcomm chips less shit? Because things don't work that way.
>>52180561
Sure Roy.
>>52180571
WHO
WILL
GIVE
THEM
THE
LICENSE
YOU FUCKING IDIOT
Do you have any FUCKING idea how much money it takes to get to what Intel, heck even AMD are at now with x86?
Hint: small firms can't dream of that much money.
>>52180002
Intel would loose the x64 license
>>52180571
There's an awful lot of "hopefully" in your post for someone wanting something as awful as Intel and Nvidia's main competitor to go under. I don't think you understand the barrier to entry that their market has. Smaller firms? The reason AMD can't compete in the first place is because they can't afford research and development costs like Intel can. How is a smaller firm with fewer resources supposed to compete? Qualcomm is and has always been a joke. The only possible entity that could compete is Samsung, and you wont see x86 out of them.
>>52180634
Not if AMD goes bankrupt. And even if they did, you'd just end up with fucking Itanium, and I'm sure you'd be happy then.
i believe in amd
>>52180634
If AMD goes down Intel keeps all the IP they licensed from AMD. They have a cross-licensing agreement. It works the same for AMD but Intel isn't going bankrupt anytime soon.
>>52180599
The Chinese
>>52180571
>new small firms that joins in the market.
Wait a minute, we will end up with AMD and Intel for centuries forever?
>>52180707
The Chinese are not innovators. Their entire industry is gear around taking existing products and manufacturing them as cheaply as possible. That's why you see cheap tablets and phones from them, they don't really have to do anything. They just take designs that are already on their manufacturing lines, tweak them, and then rebrand them.
>>52180707
>chinks
>being innovative
Just stop
>>52178973
He's right though. Monopolies aren't illegal. Abusing your monopoly status is what's illegal. It's a distinct and legal difference.
>>52180712
For x86 that's pretty likely. Or worse we might end up with Intel and basically nothing else. AMD's x86 license doesn't transfer to other companies and Via sure as fuck isn't going to compete with Intel anytime soon.
The only way to fix this is to move to an OPEN ISA, moving to proprietary ISAs like ARM will just eventually lead to the same shit.
>>52180754
>amd/nvidia ceo is Chinese
>silicon valley is literally a chinatown now
whitey want to hold us down
>>52180811
ARM is wholly different from Intel though. Their whole thing is licensing to as many companies as possible. They make their money in licensing not selling chips.
They're not even comparable.
>>52180811
I can't wait for the tech modern dystopian future
>>52180823
It's easier to find a few Chinese people that are capable of leadership (i.e. them getting into C-level positions in western companies) than it is to restructure Chinese society to not be based around cheating and intellectual property theft. Those CEOs may be Chinese, but the employees that are making the innovations sure as hell aren't. And anyone that has been on the engineering track can tell you that Chinese students are usually bottom-tier and maintain an artificial success through rampant cheating.
>>52180909
>muh anecdotes
statistics don't lie
>>52180978
They sure don't, anon. They sure don't.
http://money.cnn.com/2014/07/01/pf/college/chinese-students-cheating/index.html
>>52180978
Chinese students cheating their way through is a pretty serious issue.
What if, listen! What if the west stole everything from China? Like, everything happened before Cold war was a hoax made up by the West?
>>52180865
I'm assuming that sometime in the future someone is going to try and buy the rights to the ARM ISA and associated IP. It doesn't seem likely at the moment but there's a lot of companies getting invested into ARM it seems likely that at some point one of these companies is going to find it easier to acquire ARM and thin out their competition over time.
>>52181056
>yfw applel starts buttfucking intel
>>52181022
>>52181028
Wow, I didn't know it was this serious. Now only if there's a department or something that handles the hiring process to weed out the useless it would be of a great resource. We can call it people resources or something.
Got any ideas fellas?
>348mm2 @28nm
>draws 100w~
>would be 105mm2~ in 14nm
>power consumption would be cut in half
One package of HBM providing 128GB/s could feed 20CU moderately clocked with only a slight memory bottleneck. With 4 Zen cores and new PowerTune IP they could fit all of this into a 65w package for desktop. A 35w laptop chip would still be an absolute beast, low end discrete graphics for mobile would get totally rocked.
>>52181339
Human resources is almost always staffed with people that have no idea what the requirements for a position are. Have you ever actually interviewed candidates for a tech related position before? I have, and all HR does is take the requirements that you send them, compare it to the candidate's resume, and forward the candidate on to you if they see the correct buzzwords. Incidentally, that is exactly the type of system that Chinese people are great at gaming.
>>52181398
Jesus, that's horrible. What if they start gaming their work and actually release good products? How do we stop those evil Chinese?
>>52181439
Good fucking luck, you're more likely to just have them rip off another company's IP and get you involved in a lengthy litigation process.
http://www.telegraph.co.uk/news/worldnews/asia/china/10132391/Riot-after-Chinese-teachers-try-to-stop-pupils-cheating.html
>>52180132
Bohemia Interactive have not done shit with the engine since OFP. They make a shitton of money of VBS and the costumers using VBS either dont care about 60fps or use machines that are powerful enough to run it.
>>52181476
Alright, how can we protect ourselves from Chinese applicants?
>>52181519
You don't have to. They have their purpose, they're good at making things in mass quantities for cheap. As long as they stay in China you can profit off what they're good at. It's when they start trying to come to western countries and bring their shitty cultural ethics along with them that problems arise.
It's not a problem for westerners to solve. It's a problem that the Chinese need to solve. Until then, Chinese workers will be treated with heavy-handed prejudice. But it's unlikely that they will fix it themselves since they're more content to just riot when pushed to perform ethically.
>>52178563
>>52178531
because if they do keep the same price sku they are replacing, 8 cores will beat out the 6 in multicore, and here is a fun fact, no one gets the 6 or 8 core for their single core performance.
now what im interested in is what they will charge for motherboards, as amd will probably beat out intel there too, im not paying 300~$ for a motherboard for a 300~$ cpu.
>>52180522
yea, and the only way to get an amd cpu now is to get a fucking prebuild at best... yea... no.
I thought desktop Zen wasn't getting any iGPU. On the topic of APUs, Intel has been making them all along. Broadwell really closed the gap in the graphics department, but it's still way too expensive. AMD really needs to hurry the fuck up since right now Intel doesn't give a shit with things like Kaby Lake being announced. Hopefully Zen isn't complete shit and can at least offer Haswell performance per core. An APU with fast Zen cores and a 14nm GPU with HBM2 would be incredible.
I really really hope amd gets its shit together im still running on my phenom 2 1090t because my system still runs most applications and games i throw at it. Ive only had to upgrade my GPU twice since my build. Whats the deal with HBM is it supposed to be that drastic of a performance gain?
>>52183264
apus aren't going to be at launch, just the 8 core non apu.
>>52183349
no, well... yea its a performance gain, but the real gain comes from the size you can make now that you don't need gddr5 taking up so much fucking space.
I wouldn't mind using an ARM desktop CPU as long as it could compete speed-wise with a Haswell i5. The major Linux distros are all ready to go with ARM, so that's not a huge deal.
>>52183264
Zen isn't the name of a chip, it is the name of a core architecture.
Summit Ridge is the 8 core CPU with no IGP.
Raven Ridge is the APU.
>>52183349
>Whats the deal with HBM is it supposed to be that drastic of a performance gain?
HBM is just memory, it confers no performance by itself. What it does is allow the shaders in a GPU to be adequately fed information to process without any hang ups.
Current APUs have 8CU, and the fastest DDR3 available at a reasonable price can't provide even half the necessary bandwidth. DDR4 makes a fairly substantial improvement, but theres still a big bottleneck.
A single HBM module would allow AMD to put an IGP like the one used by the PS4 into their desktop chips and feed it adequately. Thats where the performance gains would come from.
>>52183391
sounds good so we might see a shrink in gpu sizes hopefully i like compact builds but i also like airflow so i run mid towers having to pull off some wiring voodoo to get everything to fit in my case while not looking like ass and having gangster gpu lean is a pain.
>>52183477
tiny cards, at least as they are now, are a premium product, see the nano.
once nvidia gets in on hbm and if they don't go full jew, we may see smaller cards, but i don't think people who demand performance give two fucks about the smallest case.
>>52183516
i know but its not a huge deal breaker for me but it would be nice to see cards shorter then 10.5 in again so some of the guys with prebuilts arnt screwed. Thats what has kept alot of people i know from getting into pc's because their $200 purchase doesn't fit nevermind the psu exploding.
>>52183462
That's what I meant with the desktop Zen. I tend to forget the ridges.
>tfw no monster exascale super processing unit
>Doesn't have 24 cores.
Why even bother if it i'm not getting the bleeding edge in cores?
>>52183349
HBM is DRAM but its on the same interposer as the processing unit. Being so close it has extremely low latency relative to most other memory but it also has the advantage of being stackable(more memory in one space) plus the bus used to feed the processing unit is extremely wide which gives HBM a huge amount of bandwidth. Graphics is greatly affected by bandwidth so more is pretty much always a solid improvement
They have to give me a good reason to choose their 8 core in favor of pic related. Moreover when Broadwell-E is this year, giving us not only a faster version of the 5820k, but an 8 core version that isn't the highest end. The latter will definitely be in the $700 range though.
>>52184006
It doesn't really matter if you aren't heavily AMD or Intel. But AMD forcing Intel to step up their game is a good thing for us all. Intel might have to pull all their stops on Broadwell-E to make it be better than Zen.
>>52184043
or they flex their monopoly muscle again get slapped with a billion dollar fine and amd dies... i believe they will do this before they ever try to compete again.
>>52184064
Well, then they get the AT&T style breakup. Or they will be forced to license x86 freely.
>>52184097
think that would happen in america today?
how many real options do you have for internet?
how many 25 down 4 up do you have?
i have 1, my alternatives are 3 down .2 up and dialup.
>HBM on Zen
So is this a scheme to try and get you to buy new RAM?
>>52184217
Run me through the mental process that led to you typing out that meaningless jumble of words. I'm curious.
>>52184252
If HBM is so much better, why wouldn't you want to replace regular system RAM with it?
>>52184305
HBM is fused to an interposer on the CPU package. You do not buy it separately, you could not ever install it yourself.
>>52184315
So what's the point of it? They're just using it for the cache?
>>52184327
Its there to feed the IGP, genius.
>>52184336
So.......completely useless on the average desktop found on /g/
>>52184350
Yes, but for the two idiots who use the iGPUs, it'll give incredibly fast memory.
Which, as shown by Fury/FuryX is completely worthless.
>>52175925
>amd
>bankrupt
can we please stop this meme
>>52184350
Tons of people use APUs, they're the best selling chips AMD ever produced. An APU with a considerably powerful IGP has even broader appeal.
Don't try to act clever when you make an ass out of yourself.
>>52184406
If the Zen CPUs fail they won't go bankrupt because of the console and GPU markets, but they probably will bow out of the x86 CPU market which is bad for anyone who owns a PC.
>>52184483
>console market
Kek
The profit margin on the consoles is razor thin. They earn cents for every console sold.
500 million consoles? Yep, that's 500 million cents of profit.
That'll patch the 12500 million cents they're bleeding every quarter.
And as for the GPU markets- they've got like 12% market share. The past two generations has seen AMD's market share plummet from ~40% to 15%
Bought.
>>52184483
A failure for an IC would be not bring a return on investment, and thats impossible here. Even if they scrape by on consumer sales, they're still going to gain enterprise marketshare.
>>52184549
>The profit margin on the consoles is razor thin. They earn cents for every console sold.
Their financials prove otherwise, you shitposting child.
>>52175850
aren't they pretty much locked at 8CU, 800ish Mhz to maintain a 65w load currently, and that's without GDDR5 busses and chips?
16nm finfet would see roughly a 12CU (768 shader), 1Ghz use the same power with no change to the memory subsystem.
I assume that HBM equipped APUs will be no more than 640 shader with 1GB of 500Mhz (1000MT/s * 128 (one HBM interface in bytes) = 128GB/s)
It would be (a GCN 1.3 or hopefully Polaris graphics unit) roughly equivalent to the 7790's 896 shaders.
>>52184627
See: >>52181341
>>52184377
GCN was bandwidth starved, and now it's cache starved.
They still saw big gains from it in power savings.
>>52179260
It's entirely possible to mate a high end CPU with a high end GPU chip on an interposer, linked with a very high bandwidth interconnect. HBM solves the area scaling problem of GDDR5 and simplifies such a thing considerably.
It's also in the realm of niche fantasy, so high in the clouds that you'll never see it except, maybe, in very special HPC applications in a few years.
>>52184707
Die stacking has been done for years, there is nothing fantasy about it. AMD's future enterprise APUs will in fact be MCMs. All of their Opterons in recent history have been MCMs as well, and there is little difference as far as the package is concerned.
When you don't have any knowledge of a given topic you really shouldn't make baseless tatements about it.
>>52184006
Those Haswell-E chips are humungous and expensive. You could fit 16 Piledriver cores on that chip.
>>52180770
The problem is that abuse includes throwing money in all directions until the DoJ/FTC case against you goes away, or never gets filed in the first place.
>>52175761
The 270x is still a damn good GPU. Having one built in would be nice, but I doubt the APU's integrated graphics will work as well because no GDDR5. Still, with DDR4, it could be a really solid option
>>52184823
Look at the picture again.
>>52184641
Die shrinks don't work that way though.
You'd either get that massive area reduction, at the cost of power usage and clocks (performance),
Or you'd get the power reduction with the same performance, sacrificing smaller die size.
>>52184742
>die stacking
Not at all what I was talking about. die stacking is mostly used in memory (SSDs), and layering processing elements (CPUs) has not yet been commercially viable.
I did say that such a "chip" is in fact possible (the point was mating a 600mm2 GPU to a large CPU, when 600mm2 is the limit for producing working parts). Just not feasible, because hardly anyone would buy it.
>>52184845
HBM? I don't get it. Would it be included with the APU as it's own personal memory or what?
>>52184854
>Die shrinks don't work that way though.
You need to stop making all of these stupid assumptions. You have no knowledge on any relevant topic, all you're doing is talking out of your ass.
Take a good hard look at the attached pic then apologize for trying to correct someone far more knowledgeable than yourself.
>Not at all what I was talking about. die stacking is mostly used in memory (SSDs), and layering processing elements (CPUs) has not yet been commercially viable.
Yet again you have no fucking clue what you're talking about.
3D stacking is not die stacking. Die stacking is a specific term used in the industry to describe manufacturing on MCM, particularly with components made on different processes.
The WiiU's MCM is a die stacked part.
Go get murdered, or just voluntarily stop posting here. You single handedly are making this board worse.
>>52175761
We need a socket am1 update
>>52184965
Probably not happening unless AMD repurposes some Beema dies. Looking at the sales for Carrizo-L I don't see that happening. Their new low power part Stoney Ridge is DDR4, so it has no chance of ever winding up on AM4.
>>52184919
>You have no knowledge on any relevant topic, all you're doing is talking out of your ass.
I work in the chip fabrication industry.
I am a regular on a few high level design forums.
You're just throwing out meme images with no explanation of how the fundamentals of design behave.
DIE SHRINKS are:
>the smallest they can go
sacrifices leakage and attainable clock speed
>tuned for higher performance
sacrifices the ultimate size of the shrink and power draw characteristics
>tuned for power
sacrifices ultimate performance and the shrink size
>mixed for a benefit of the three
In one case, you lose performance for size and leakage
In one case, you lose (some)* size for the same performance at less power
In one case, you lose (some*) power savings for the same performance at (some) size savings
In one case, you lose size and power saving*s for performance orientation.
In many cases, the aim of a shrink is to maintain performance while lowering size and power, which have to make tradeoffs*.
Note that *(some) does not mean a regression compared to the larger node, but instead refers to less than the theoretical lowest figure compared to a shrink designed for that particular metric.
But please, keep telling me what I think I know.
>>52185019
>so it has no chance of ever winding up on anything but AM4*
fixed
>>52185020
That "meme image" is taken directly from a technical presentation by Global Foundries.
You're a terrible liar butthurt about being called out. A complete tech illiterate retard and shit eating autistic child.
14nm LPP provides over a 70% area scaling advantage over 28nm HPP. It does this while providing massively lower leakage, and greater frequency. These are empirical facts. Your childish shitpost tirade will never change this.
Try actually educating yourself instead of posting laughable bullshit. Or don't, the world would literally improve if you died tonight.
https://www.youtube.com/watch?v=SPQvgzL_DSc
>>52185073
Oh fantastic! A foundry's advertising image surely tells the whole story!
Surely they don't neglect the subtle facts that when choosing for a particular metric (size, power, or performance) you have to sacrifice a bit of the others to embellish the target!
Surely!
Keep being a /g/ pleb, little boy.
>>52184568
if i remember right in total, they were looking at either 1.2 or 12 billion dollars this console cycle... its all of jack and shit once you spread it out over 5+ years.
>>52185102
>get completely blown the fuck out
>i-i-its just marketing!
You couldn't tell the difference between sLVT, LVT, and HVT if your life depended on it. You demonstrably don't understand something as simple as BEOL interconnects and how it affects area scaling either. You're so absolutely pathetic that you tried to lie about working in the semiconductor industry to win an argument on a topic you know nothing about. That is hardcore childish autism.
>>52185144
>Buh-bu-bu-but I was told that with this thing, comes all these other things
>What do you mean when I focus on one the others don't improve as much?
>couldnt tell the difference between low, medium, and high
You're ten times the nigger I thought you were. My job is literally testing designs for their operational thresholds within given min/max parameters.
>don't understand back end interconnects
O am I laffin
also
>trusting Global Foundries
That's worked out for the company which is the topic of this thread, so well, for so long. Hasn't it?
t. fag from Chandler, AZ
>>52185262
>even more childish autistic shitposting
14nm LPP is Samsung's process, you not knowing this is just another hilarious strike against you. GloFo is offering parity at Fab8 with 3 of Samsung's facilities, they're all offering the exact same process. Trying to slight GloFo over this is just stupid.
You're not going to fool a single person, its more than apparently how desperately clueless you are. You do nothing but shitpost like a /b/tard when confronted by sourced empirical facts. You're just another NEET dipshit who stumbled into a discussion he couldn't keep up with and began shitposting to save face.
>>52185316
>trying to slight Glofo
Yes, I agree it was underhanded. Laugh a little.
>not going to fool a single person
I hope to educate them. The world isn't miracle fantasies and indeed any choice has drawbacks from other areas.
>sourced
you have never, aside from one investor marketing image, provided any sources
>empirical
My knowledge coincides with the general consensus of people in the industry. You're a tryhard hot breathed faggot.
>began shitposting
I handle this every day. I live and breath transistor qualification.
>save face
We have no identity here, nigger
>shitpost like a /b/tard
4chan is 4chan. May we turn the mirror back on yourself?
>>52178009
Hold out bro, it'll go up
AMD = quantity over quality
Intel = meme
>>52185384
I directly liked the technical presentation that the slide came from, you pathetic lying autist.
Claiming to work in the industry and demonstrably proving that you know literally nothing about the industry is akin to claiming your dad works at Nintendo.
>>52185419
>investors analyst presentation
>no products in the wild
>pathetic lying autist
One of these things is not like the other...
.>>52185384
>transistor qualification
>>52185447
Continued shitposting isn't helping you at all, liar.
That isn't a presentation for investors, it is a technical presentation held for the engineering staff of IC vendors. They hold these events in a room full of technicians that are sent there from companies looking to adopt the process. Again, if you worked in the industry you would know this.
The more you lie the more pathetic you are.
>>52185522
>wahh wahh someone isn't spouting the obscene bullshit I want to spread
>>52180754
>>52180766
their being innovative,
at making money.
