POLARIS 10 CONFIRMED FOR 234mm^2
ITS OVER NVIDIA IS FUCKING FINISHED, THEY NEED a 50% LARGER DIE TO GET 10% MORE PERFORMANCE WITH GP104
http://videocardz.com/59487/amd-polaris-11-and-10-gpus-pictured
It's Fermi all over again
so which one should i buy AMD OR NVIDIA???
not going to wait over 3 months so ITS NVIDIA!!!!!
>>54320925
You'll still wait 3 months anyway since both Nvidia and AMD launch at Computex
Typical, Nvidia blew their architecture load with Maxwell it was supposed to be a 20nm part but that got canceled, they repurposed it for 28nm, Pascal is Maxwell 3.0 since they can't make an entirely new architecture that fast.
I don't get how this matters at all. Is needing to make the die larger a problem?
>>54320956
Larger die means more defects, less chips per wafer, higher price, higher power consumption.
A larger die that's only 10% faster but 40% bigger than its competitor is quite literally Evergreen vs Fermi tier.
>>54320971
Okay, I only thought that it meant less efficient use of space on the card or something.
And I guess the GPU pictured here is small, and that Nvidia information is from some other source?
pretty sure pascal will pwn the fermi 9000 anyway. nvidia always does.
>>54320979
Nvidia's GP104 has already been measured a while ago at 317mm^2
http://wccftech.com/nvidia-pascal-gp104-gpu-gddr5x-leak/
>>54320989
Thanks, I haven't really been following Nvidia's news because I'm already committed to AMD anyway.
There's actually a chance we might get 3 GP104 and Polaris 10 GPUs if the yields are low enough, and from what I got so far they are.
>AMD returning to small dies again instead of chasing retarded reticle limit dies
THANK YOU BASED AMD
Now AMD can do what it does best, transistor density per mm2 they are unrivaled.
>Zen
>Polaris
>m.2 SSD
faster than light
>>54321025
>not U.2
Pleb
>>54321021
This, Vega will at most go up to 400mm2, AMD doesn't need 600mm2 dies, they can be 10% slower than Nvidia but be much cheaper and overclock like mad.
>>54321030
>not 4.U
Subhuman
>>54321030
oh nice, another connector, exactly what we needed.
>>54321052
Intel knows best.
>>54321078
>Desigend
lal
http://www.gamespot.com/forums/pc-and-av-hardware-909394553/amd-going-back-to-small-die-strategy-thoughts-33021274/
Are you people retarded? (rhetorical question)
Neither Polaris 11 nor Polaris 10 are aimed at the highend segment, which should point out their performance range.
They will be slower than a R9 390X.
Nvidia is releasing cards soon that are gonna replace their high end cards.
>>54321201
>They will be slower than a R9 390X.
Apparently they're faster than a 980Ti mate. Lower tiered, but faster.
Also GTX 1080 will come with GDDR5X.
AMD has no card against this until Vega.
>>54321241
Desperate wishful thinking
You are retarded to believe they are faster than 980Ti, which makes them way faster than their own highend cards, the Fury series.
Seriously, do you people even have a brain?
>>54321254
I never said I actually think they're faster than a 980Ti, just said that they're apparently faster than a 980Ti.
Also, you provide no reasons to believe otherwise anyway. You're just claiming it's impossible because it would be faster than a Fury, which is a Non sequitur.
And also, it doesn't need to be faster than a 980Ti to be faster than a R9 390X.
>>54321254
Could you post any source?
>>54321283
Moron
AMD themselves said that the Vega series is gonna replace the Furies.
Now do the math.
>>54321201
GP100 only releases next year.
>Polaris 10/11
>confirmed to be aiming at low end/midrange market
>retards hyping the cards up to be way faster than the current flagships
Gaymer kiddies, not even once
>>54321297
There's no math to be done because I'm missing variables. New generations of cards can be faster than older generations of cards, even if they're lower tiered. All your saying is that Vega will be faster than Polaris, which I'm not saying anything about (just to clarify, I think that as well). This has no relation to Polaris' speed, at this point in time at least.
>>54321316
>>Polaris 10/11
>>confirmed to be aiming at low end/midrange market
Source?
>>54321201
you stupid fucking cunt, how can you assume that?
>>54321254
highend next gen cards are as fast or faster than previous flagships, this has been ATi/AMD's and Nvidia's doing for years, what exactly is surprising here?
Neither AMD or NVidia is gonna release a Fiji/GM200 replacement until the first quarter of 2017.
>>54321283
polaris 10 is speculated to be nearly as fast as a 980 ti
>>54320902
What's with that shitty picture?
Fuck me, here is some sources on all the bullshit going around:
Source on Polaris 10 and 11 being lower/mid tier cards: http://www.overclock.net/t/1598134/vc-amd-officially-confirms-polaris-10-and-polaris-11-market-positioning
Source on Polaris apparently being as fast as a 980Ti: http://www.overclock.net/t/1598515/game-debate-rumour-amd-polaris-10-reportedly-offers-near-980-ti-performance-for-300-usd
New source on Polaris 10 being more in line with 390X performance (though if you read the latest posts in the thread, it mite be the M430 that is being benched, which would mean new consoles and laptops with Polaris in them will be pretty fast): http://www.overclock.net/t/1598845/gfxbench-polaris-10-benchmarks-found
Take from this what you will, its still all up in the air, but I can guarantee you now that 1070 and 1080 will be quicker than 980Ti, by how much is debatable, but they are shipping with GDDR5X and pretty high clocks out of the box, so I am tipping it will be something around 25% faster (still a bit dissapointing really, but whether they then OC on top of that will be interesting).
>>54321425
Fair enough. Thought it said faster, but guess it's just near.
>>54321619
>Source on Polaris 10 and 11 being lower/mid tier cards
That just says "mainstream desktop" really, which just means not enthusiast tier like the Fury.
>>54321794
They've been rumored by multiple sources now to be the 480-490 replacements for awhile now, or maybe even 470/480 and Vega will be 490-490X (That to me makes more sense, since I doubt AMD will use a Fury moniker or anything stupid like that again for awhile now).
>>54320902
>234mm2 @ 14nm FF GloFo
Alright, so how fast is this bad boy gonna be?
>>54321794
iirc they make more than low/mid cards than high-end, which doesn't come close to the sales figures of low/mid cards. Not sure as I said but it's worth considering. If they make more money from low/mid range cards then an extremely viable mid-range competitor is what they'd need much more so than a high-end card.
>>54321201
>Neither Polaris 11 nor Polaris 10 are aimed at the highend segment
Except Roy Taylor literally confirmed that they are just a few days ago. Keep up, cunt.
http://wccftech.com/amd-polaris-mainstream-vr-market/
Priced around the same as the R9 390 and will beat it in performance whilst consuming half the power. The only thing they're not targeting is the ultra high end, which will come next year with a HBM2-equipped Vega chip. Nvidia won't be releasing a new Titan/Ti until then either.
when can we get yet again nice sexy single slot cards?
there are so many pci express ports on motherboards today and so very little use for them
will this round of amd chips and dx12 embed multi-gpu rendering mode make us go us back to the good old days?
IT'S OVER, AMD IS FINISHED & BANKRUPT
>>54320971
>>54320902
>>54320979
and everyone else talking about chips size vs transistors count.
What people miss is that a larger die with smaller transistors means more space and mass to handle thermal spikes for better clock to load performance. Think of the long term life cycle performance of such a chip. It is more then just shrinking and shrinking, as there are other factors depending on the end application.
>>54321362
there is none because it simply isnt true. Polaris 11 is a midrange card for the average user/gamer whatever while the big chip is supposed to be somewhere above Fury X/980Ti while consuming significantly less energy.
>>54320902
DELETE THIS
>>54320929
Ohh man, that's going to be good.
>>54321241
Nope, a Polaris 10 was benchmarked and scored 85% from a stock clocked 980 Ti and 90% from a Fury X. The gap between the supposed 1070 and the 980 Ti was within 95%.
AMD BTFO yet again.
>>54321362
http://arstechnica.com/gadgets/2016/04/amd-polaris-will-be-a-mainstream-gpu/
From Roy himself
>>54321052
>another connector
>>54321052
>Intel knows best
It's because SATA Express only had 2 PCIe lanes and enterprise customers wanted more bandwidth for 2.5" drives.
m.2 has 4 lanes but isn't a form factor suitable to continuous heavy load. u.2 drives can suck down 25W, while fast m.2 drives like the samsung 960 thermally throttle at under 5W.
>>54323461
U.2 is just a rebranded mini-SAS HD connector.
>>54323487
yeah but with different pin assignments IIRC.
enterprise just wanted 4 NVMe/PCIe 3.0 lanes with the least amount of hassle and just made their own shit up without waiting for slower standards bodies.
U.2 looks likely to murder everything else in the data center market, given support from both drive and server manufacturers.
>>54323310
mainstream = $300 or below
AMD's strategy is to specifically target these price points that actually sell in high volumes.
Nvidia is making a bigger chip so they can proclaim that the 1080 is the fastest single GPU in the world and look good in benchmarks, but at the lower price points they will not be competitive.
>>54320909
>Fermi
Such a shit gpu i hated it
in fact went straight from my 285 to a 6970 to a 770
had a 290 but it sucked ass back in 2013 launch so i got rid of it.
4gb isnt enough these days anyway, i really cant wait for polaris
>>54320902
>http://videocardz.com/59487/amd-polaris-11-and-10-gpus-pictured
le hardware conspiracy theories
AMD is being pretty non-competitive in the high-end for a few months, but the mainstream cards will likely be power efficient enough (and inexpensive enough) to make their way into OEM vendors' hands. By wattage, it seems that the performance will probably be 980ti tier, seeing as that the Fury Nano is already close the kind of efficiency needed to scale to 980ti performance at the same wattage. However, that still means that these cards aren't VR ready, hence why AMD calls it "mainstream". Not incredibly excited for Pascal in comparison though, I'm pretty hype for AMD to regain the mainstream marketshare.
>>54322751
Except with the renewed focus on power efficiency both teams are pretty far below what you would be worried about with power draw. I would be surprised to see either of the top end mainstream cards jump above 75-80C under full load if they're both going to be drawing less than 200W.
Although, if Nvidia fucked up their FP64 cores again we could see another Fermi...
>>54323777
I'll certainly be replacing my 670 with this stuff.
>better support over the years
>better performance because of that support
I liked my 670 for the couple years it got good support, but now it's shit.
Daily reminder that:
>8800 GTX ran over its TDP and overstressed PSUs
>Mobile GeForce 8000 series de-solder themselves http://forum.notebookreview.com/threads/gpus-igp-nb-desoldering-themselves.582158/
>Fermi GF100 hot chips yield 1.8% success https://semiaccurate.com/2010/02/17/nvidias-fermigtx480-broken-and-unfixable/ http://www.zdnet.com/article/nvidia-fermi-gf100-gpus-too-little-too-late-too-hot-and-too-expensive/
>Nvidia pays developers to inject unnecessary amounts of tessellation into games http://www.extremetech.com/extreme/173511-nvidias-gameworks-program-usurps-power-from-developers-end-users-and-amd/2 [https://www.youtube.com/watch?v=IYL07c74Jr4&feature=youtu.be&t=1m46s]
>GTX 480s caught fire http://forums.evga.com/My-GTX-480-SC-just-shot-fire-out-of-the-top-fan-intake-m626592.aspx
>GTX 590s caught fire demo: https://www.youtube.com/watch?v=sRo-1VFMcbc
>GTX Titan had no drivers until 3 months after launch
>NVIDIA launched drivers that melted GTX 780s
>GTX 960 beating the Titan Black and GTX 780 Ti because of nerfed drivers
>970 ram nerfing http://www.anandtech.com/show/8931/nvidia-publishes-statement-on-geforce-gtx-970-memory-allocation
>Law suits against Samsung and Qualcom for questionable stuff
>Nvidia using non G-Sync technology on G-Sync branded laptops
>Nvidia speed up batman demo to make it look more fluid https://www.youtube.com/watch?v=zsjmLNZtvxk&t=24 listen to the thugs voices
>Windows 10 drivers brick laptop displays https://forums.geforce.com/default/topic/862417/geforce-drivers/windows-10-official-353-62-drivers-are-killing-samsung-and-lg-notebook-lcd-display-panels/1/
>Now we had DX12 test, granted it's just one DX12 test, but it's DX12 nonetheless... Nvidia sucks big fat cocksicle on it, and how do they react? They blame the devs.
>http://www.theregister.co.uk/2015/07/28/windows_10_update_nvidia_driver_conflict/ Release Win10 drivers being flat broken
>>54320925
maxwell has no dx12 support no reason to buy it
>>54321201
they've stated the 490X is near 980ti performance
>>54325382
Daily reminder? Are you on crack?
>>54321316
it only has to be about 20% faster to match the high end cards at 4K which shouldnt be hard given they are moving from 28nm to 14nm
>>54325382
>implying Nvidia hasn't learned from all of its past mistakes
You're forgetting that ATI/AMD had the same growing pains around the same time. GPU technology is still rather young and in development. Neither companies are dumb enough to fall for those same mistakes again
>b-b-b-but muh greedy corporate Nvidia
If you think Nvidia would fall back on those mistakes willingly, then you're dumb and a bag of rocks. AMD has less money, less manpower, and less time to refine their own designs before they put them on sale. They have the worst track record for supporting their own product line than 3DFX. If anything, AMD is most likely to repeat their mistakes because they simply can't afford to be as thorough as Nvidia.
>>54325822
>You're forgetting that ATI/AMD had the same growing pains around the same time.
AMD has never bricked cards or displays.
>Neither companies are dumb enough to fall for those same mistakes again
Nvidia has released at least 4 drivers very recently that bricked or damaged cards.
>>54325938
>AMD has never bricked cards or displays.
AMD fucked up their drivers a while back and fried cards.
https://www.youtube.com/watch?v=ruCr6OXbK5E
>>54325938
>AMD has never bricked cards or displays.
Yes they have, in the last year there have been a dozen reported instances due to their shitty Crimson software. Fans wouldn't work or running a high refresh-rate monitor would cause the cards to use an abnormal amount of electricity at idle.
Not to mention that ever since GCN was released, AMD's GPU reliability has sank like a rock whereas Nvidia's reliability is stronger than ever with Maxwell.
>Nvidia has released at least 4 drivers very recently that bricked or damaged cards.
No card was proven to be broken or bricked. Those were exaggerated reports from childish buyers who over-reacted when their driver crashed and they BSOD. In fact, at least one of them was banned from the official Nvidia forum for spreading false information that the new driver killed his card. Nvidia was able to prove that he was lying with the intent to damage their business.
>>54320902
>ITS OVER NVIDIA IS FUCKING FINISHED, THEY NEED a 50% LARGER DIE TO GET 10% MORE PERFORMANCE WITH GP104
you mean both AMD and nvidia produce their hardware at the exact same factory on the same exact hardware??
holy shit
but why does nvidia always cost twice the amount?
and please don't answer this question with "muh architecture" because its proven to be only 5% better on average, with AMD actually beating nvidia is many games and especially in GPU compute
please explain why
>>54326077
It's the way it's meant to be played.
>>54325980
>>54326028
No, the 3rd party overclocking software people used instead of the overclocking software in AMD's drivers did that.
AMD's drivers never directly fried any cards, and no cards were even proven to be fried.
>>54326028
>implying 320.18 never happened
http://modcrash.com/nvidia-display-driver-damaging-gpus/
>implying 353.62 never happened
https://forums.geforce.com/default/topic/862417/geforce-drivers/windows-10-geforce-drivers-are-killing-samsung-and-lg-notebook-lcd-display-panels/1/
>Not to mention that ever since GCN was released, AMD's GPU reliability has sank like a rock whereas Nvidia's reliability is stronger than ever with Maxwell.
You got a source on that? I want reliability numbers from every GPU manufacturer from GCN 1.0 onwards, of course, since that is your claim.
>>54326159
>You got a source on that
http://www.pcworld.com/article/2052184/whats-behind-origin-pcs-decision-to-so-publicly-dump-amd-video-cards-.html
https://www.pugetsystems.com/labs/articles/Video-Card-Failure-Rates-by-Generation-563/
Both sources are from mass builders of custom gaming PCs with nearly a decade of experience. They know their shit.
>>54326226
>2013
>2014
Yeah, that's what I thought. You can't provide a modern source, only a source from over 2 years ago.
Also, Puget only uses ASUS cards, not cards from a variety of vendors which most people use, and the Origin article uses the same numbers from the Puget article and never actually gives failure numbers from Origin.
>>54326346
The 300 series is just a rebranded 200 series, so why should the reliability figures be any different? They are the same exact GPUs with the same exact manufacturing process made by the same exact people and put under higher stresses. If anything, those cards should have even worse reliability.
>async
>dx12
>lower frametimes
>proper double precision
im an #AMDMissile this year
>>54326401
Because, and I don't know if you know this because it's totally a super super secret fact of the manufacturing industry that absolutely nobody knows, as time goes on your reliability goes up because you learn to manufacture things better as you manufacture the same things over and over and over.
If what you're saying is true, then Nvidia's failure rate for the 400/500 series should have been through the roof even at their EOL.
>>54325382
>>GTX 590s caught fire demo: https://www.youtube.com/watch?v=sRo-1VFMcbc
Why the fuck did the fan stop?
>>54326501
How else are they going to break your card and sell you a new one? It's the way it's meant to be played.
So basically it's worth it to wait?
>>54328334
Yes
AMD is doing a total refresh of its lineup starting with low to mid range (Polaris) to high end and enthusiast (Vega).
>>54325382
>http://www.theregister.co.uk/2015/07/28/windows_10_update_nvidia_driver_conflict/ Release Win10 drivers being flat broken
at this point I've completely abandoned Nvidia when I've done nothing but buy Nvidia since i started pc gaming
FX5500-->8800GTS-->GTX260-->GTX560TI
now Im getting an AMD gpu when polaris releases
Ive already abandoned the (OY VEY) intel market and only go AMD cpu, will be happy when they release ZEN to go with mostly AMD in my system
>>54326418
>proper double precision
says who?
maxwell all but completely gutted fp64 capacity, and Fiji did the same.
fp64 capacity only makes sense on GPUs that will be sold in both gamer and professional/enterprise/HPC variant cards, otherwise it's mostly just a waste of transistors and energy.
I'd personally be happier with Polaris, med. Pascal, or even Vega with minimal double precision support.
GP100 being <=4k shaders at 1.3-1.4(turbo)GHz is basically sandbagging for the high-end gamer market this year.
GP200 or whatever with 6k+ fp32 shaders in 2017-2018 will be a drastically better consumer product and one that Nvidia could have easily released a year earlier if they weren't trying to milk the 16nm process gen. for as long as possible.
I was planning on getting a Vega, but if it pulls this shit too I have no qualms in skipping another gen.
>>54329739
>either company can bring maximum reticle limit chips to consumers with first to market products at any acceptable price, yield, or quality
You're out of your mind. There's a reason why GP100 with full dp64 is going to price above 10k per card in the professional space, and it isn't solely because of CUDA performance
>>54330572
GP100 and Vega can't come out in volume until H1'17 because the gooks can't get HMB2 out the door fast enough.
The extra 6-8 months beyond Polaris/GP104 releases should be plenty of time to get 500+mm^2 die yields to reasonable levels on the Samsung/TSMC process nodes, especially if they go with copious fusing/binning.
Nobody really gives two shits if their fancy new card will have 6200 or 5800 shaders given that last gen was stuck in the 2.5k-4k territory.
8 core zen APUs with SMT and HBM2 on die fucking when?
>>54331763
2019.
>>54331763
>HBM2 on die
I don't think you understand what "on die" or "HBM" mean, anon.
>>54320925
buys amd polaris, 2018 finds its no longer supported in Ubuntu 18.04LTS
>>54329714
If you think NVIDIA has driver issues, just wait until you experience AMD! :)
>>54323293
>a Polaris 10 was benchmarked and scored 85% from a stock clocked 980 Ti
it scored 99.4% of a 980ti>>54324511
>>54326077
because nvidia actually gives a fuck to optimize your bought card to every new game that comes out.
AMD card - bunch of tiny chinese calculating faster then nvidia but shit bricks when it comes to new games and optimization.
>>54332019
on package then faget
>>54332433
and after the game is a week or so old? most people wait to buy games during steam sales anyways
hell look at fallout 4 patch 1.3 AMD is in a pretty sizable lead for the 390 vs 970
>>54332634
Nvida user buys gaymes full price within couple days of release
AMD users buy gaymes on sale a year or 2 after release
Sounds right
>>54332674
>a week or two
>translates to this anon as a year or two
yes unless the game works within 24 hours of launching it's not worth playing at all; media only has worth when it is perceived as new, not on it's individual merit
>>54332701
>top kek
Steam sales would not happen a week or 2 after release
>>54320902
>THEY NEED a 50% LARGER DIE TO GET 10% MORE PERFORMANCE
That's literally what they have been doing since forever. Making super big dies that perform 10% better but cost 50-100% more than AMD/ATI cards.
>>54332734
i know that, but the general idea is that after a patch/driver update AMD usually beats nvidia. nvidia only has the track record of being the best on day one, and for the last year they've even had problems keeping up with that
>Polaris 10 vs Pascal
>Freesync vs Gsync
When June/July rolls around and I see some benches I can make a decision.
Hopefully the 490X performs well and is priced accordingly, then I can grab the MG279Q monitor. I don't really want to pay $300 premium for the PG279Q just because of the Gsync module.
Here's to hoping
>>54320902
NVIDIA already committed harakiri when it bought into the self driving car meme
>>54333697
Yeah i need a new IPS screen my TN is driving me nuts
>>54334207
Yeah but it's also 2560x1440, 144 Hz, and framerate sync with all the adjustments like tilts and swivel
>>54333697
yeah, I havent decided between that same monitor or Samsung UE590
>inb4 not ips
I saw it on display at best buy. its fucking good enough
I love how some of yall are prissy about the cards' theoretical performance gains, that it wont be a big jump from current to new but when the gtx 900 series was the same deal the world couldnt stop masturbating to Nvidia
>>54332433
Oh look, someone who hasnt used or kept up with Crimson thinks they know what they're talking about
>>54334226
Yeah will be quite a step up from 1920x1200x60
>>54334366
the world masturbates to nvidia even when nvidia fucks them in the ass, they love it
So if i want a new card before ~november , i can get a 970 now right?
No way these polaris\pascal cards are going to appear on secondhand markets by ~november right?
>>54321814
hitman was maxed at 1080p/1440p never going below 60fps, that's all we know, likely vsynced, so all is speculation beyond that point.
look at what gpus are able to do this and that's the minimum you should expect.
these gpus got massive fuck off revisions to them while keeping largely the same shader architecture, as it was everything around it that was letting the gpu down.
its sad, but we have literally nothing to base assumptions off of spec wise, for either nvidia or amd.
>>54322751
bigger chips mean FAR more expensive chips.
last i remember, a single wafer was something along the lines of 50 grand regardless of what you did to it.
>>54323777
um... amd just put 2 furys on one pcb, its more powerful than a 980ti and uses 20 watts more power at load, and less at idle.
unless something else is going on that i'm missing, even current amd is more power efficient.
here is the thing, all nvidia has right now is the high end, they cant shrink maxwell for a rebrand as that would loose much of the efficiency due ti it being node/process specific tweaks, all nvidia can do with the new process to out do their old cards is just make a more powerful gpu, and this requires a fucking stupidly large chip, whereas amd given the hitman demo just brought high end gaming down to a mainstream price of 250-300$
>>54334004
self driving cars are the future, how far out is a good question. god knows if i could do other shit in the car other then drive, i would be fine liveing a good 1-2 hours away from work, a 4 hour long comute every day would not be nearly as painful as it would be today, it would allow me to get a house in the fucking sticks, well, for me it would allow me to get a house in a cheap neighborhood with good internet, apposed to a small as fuck appartment that i pay to god damn much for, if i could literally do anything else but drive.
fuck i don't get motion sick, i could watch tv/movies/play a videogame in the car to work, and back from work and when i get home, do shit i can only do there.
>>54320902
Holy shit it's just GCN and yet nVidia is BTFO
Occasional reminder that AMD bans anyone who gives worse reviews on their cards compared to nvidia from future amd events and from review samples.
>>54332433
>he thinks Nvidia still makes good drivers
>>54332765
the only reason they have that track record is almost always their involvement with a dev and locking render paths from other devices forcing inefficient work arounds.
>>54336268
No, 300mm wafers generally cost around $4000-7000, depending on how bleeding edge the process is.
>>54320902
That won't matter much unless they have drivers :^)
>>54338055
We should start a fire to get their attention
>>54336539
>He doesn't knows the Nvidia subreddit is moderated by AMD shills.
Not long ago a thread that said xxx.xx driver was causing problems was pinned by the mods, a bunch of people started questioning the OP and trying to troubleshoot his issues and he refused to post proof or get help every time, followed by a crapton of people posting their pc configuration and reporting having zero issues.
Those same mods also moderate the AMD subreddit, allow GPU troubleshooting threads on the Nvidia subreddit but ban anything of the same nature on the AMD subreddit.
>>54334004
They are making more money than ever.
>>54323646
>>>54320909
>4gb isnt enough these days anyway, i really cant wait for polaris
4GB isnt sufficient enough? I'm playing at 1080p everything on high/ultra with the majority of current triple A titles and I only have 2GBs of VRAM. if youre attempting anything higher than 1080p then fair play to you sir.
Im new to GPU stuff, can someone give me the short of what we could expect from Polaris and Pascal?
Will it be a big leap forward or will it be another incremental 5-15% performance improvement with lower heat and power draw? I would think that it would be a bigger leap because amd and Nvidia are halving a transistor size which they've been stuck on for years.
>>54339281
well nvidia will release their pascal before amd releases polaris so, if u gamer then go for the nvidia.
>>54339314
Yes but is there any speculation on how big the jump in performance will be?
>>54321254
Not the other guy, but with die shrink + new compression, this could very well be possible. Its happened before during the 4000 series HD. It could happen again.
Anyone in this thread bought a 390 series card with the free game? How was your experience with redeeming it? I've been waiting for 2 days for their shitty customer service to check my proof of purchase so I can have my damn game.
>>54323611
Polaris 10 (490) will be the $300 mark, 11 (480) will probably be the $200 mark.
Since Nvidia is launching 1080/ti variant, AMD will probably launch their own Vega, aka 4K to compete with nvidia's 1080p. Literally cucking the nvidia cucks.
>>54332224
Nvidia has driver issues, just this year, they had two different drivers WHQL (tm) where they killed GPU. Their latest drivers are complete shit with anything windows10.
AMD has mainly been releasing stable beta releases monthly and its working out just fine. One exception was when they moved to the new crimson software and fan controller setting wasn't properly coded. That was fixed a day or so later.
>>54339387
this is literally what they tried with the fury series and we all know how shit they are compared to the 980 ti in 90% of games.
>>54339486
Fury wasn't bad tbqh. It was merely the 980ti that cucked them. The only series in the 900 series where nvidia really outshined them.
>>54321201
>(rhetorical question)
:^)
>>54339387
>Polaris 10 (490) ... 11 (480)
You're so off base, it's not even funny. 10 will be up to the 480X most likely, 11 will be low tier and laptops (470 at best).
Don't you fags ever get tired of arguing over which double decker sagmonster is going to give you more FPS in Crysis 11?
>>54339570
>crysis
This isn't video games, this is technology.
Also better GPU mean better browsing. Hardware required to render the large webpages and the smoothness it brings is due to modern graphics card.
So its necessary to get the best card for a good price inorder to browse 4chan.
>>54339655
>Also better GPU mean better browsing. Hardware required to render the large webpages and the smoothness it brings is due to modern graphics card.
Not him but any GPU made within the last three years will suffice. Even Intel, with their poor designs and insanely bad directX/OpenGL support, now have good enough IGPs.
>>54320902
>lurk /g/ for years
>it's over for nvidia
>nvidia are finished
>nvidia on suicide watch
>nvidia btfo
>still waiting for that glorious AMD showstopper
>pubes going grey
>will I live to see AMD glory day?
>>54339655
>So its necessary to get the best card for a good price inorder to browse bloated web 3.0 garbage.
Fixed for you. 4chan runs fine on even the shittiest bay trail atom netbook.
>>54340174
No. The desktop market is already on it's way downhill, even if AMD makes up a bit of lost ground while Intel/NVidia retreat, it will be in a proportionally smaller market.
>>54340174
lol in that time amd's market share has probably decreased instead of this massive global domination ayymd shills keep going on about every year
>>54338454
>claims someone else posted some BS with no proof
>claims some BS with no proof
ayyyyyy
>>54340422
>damage control incoming
>>54340483
>DX12 support on Fermi
Sure thing m8
>>54340483
How assblasted one has to be to create that image? These gpu threads are hilarious. You man children never fail to impress me with your red vs green shenanigans.
>>54336437
didn't it happen only with techpowerup?
>>54329714
I am sorry to break it to you but every AMD processor built after 2013 has OYVEY inside
you can't get away from it
>>54340636
http://www.eteknix.com/things-go-from-bad-to-worse-for-reviews-of-amd-radeon-fury-x/
http://www.kitguru.net/site-news/announcements/zardon/amd-withdraw-kitguru-fury-x-sample-over-negative-content/
>>54340684
there's a whole reddit forum thing full of amd fanboys damage controlling about this
>>54340684
>known Nvidia shill sites get assmad AMD doesn't give them free cards to test
>get mad and type down their shitfits because they can get paid to bash the FuryX like Jen Hsun paid them for
Give me a fucking break
>>54325382
Man i am so glad i got the 770 its been literally bulletproof.
Probably their last good card apart from the 680 and 980
3 years of solid performance chucked out the window because of no drivers for my card and now no DX12 support after they said it would have it.
Sure it has it in Software but it runs like shit and so does the 970
Weirdly enough the 980, ti and titan run dx12 alot better by comparison so some serious bullshit is going on but you can forget AMD
>2013
>290s imfamous for literally frying themselves and taking the whole pc with it
i can attest to this my 290's memory crapped out and took my motherboard, ram and psu with it in 2014 just a month after i got rid of it.
>No decent drivers for the 290 and shit DX11 support until 2014+
>390 literally a overclocked 290x sold at full price
>Fury only has 4gb
>No GDDR5X Fury over 4gb
>Duo pro only had 4gb of memory per gpu
common /g/ both are pretty fucking shit right now and while the future looks alot better for AMD i doubt nvidia will pull anything good out of their ass until next year at the very least
Pascall is literally just a 980ti with a die shrink and more cores = shit DX12 performance no hardware async AND obvious gimping of older cards in drivers.
Polaris and next years refresh might be good/
>>54340784
>literally bulletproof.
It's a graphics card, not military armour.
>>54340778
>websites which have given amd very favourable reviews in the past and constantly shit on nvidia for 3.5gb are banned from receiving review samples because they understandably called out amd for releasing more rebrands with the r9 3xx
>amd get all buttflustered and ban them from events and future review samples
this is why i'll never buy amd. disgusting company.
>>54340831
Still a great card.
shame it got gimped i was planning to use it for a while longer but with nvidia fucking up every major driver release since 2014 i just cant stand it anymore.
i remember when they actually used to fucking optimize their drivers for all cards including new and as old as one to two generations back but i havent seen that since the late 2000's
ATi was never the same under AMD either the driver support in 2015 was fucking hilariously bad.
>>54336325
>how far out is a good question
not in our lifetimes
ever used google maps?
subsidiary of the largest corporation on the planet and they cant even get navigation right
fucking thing will not even recalculate your route once you take one turn different from the suggest route
how do they ever hope to get self driving cars when they can't even into navigation?
it's never going to happen
>>54338466
well then their clients are throwing money down the toilet then, they will realize it eventually and stop buying into this meme and then their sales will drop off
R9 490x will compete against the GTX 1070
R9 480x will have no competitor until GTX 1060
GTX 1080 will have no competitor until Vegas.
AMD better get the 480X right this time.
The R9 380(X) was a shit card.
Oh, hello guys! Just passing by the remind you why nVijew is done for good.
nVidia BTFO part 1
https://www.youtube.com/watch?v=0ktLeS4Fwlw
nVidia BTFO part 2
https://www.youtube.com/watch?v=aSYBO1BrB1I
It's fucking over to them.
>>54341342
>he doesn't know this has all been disproven
>>54341342
>spend years claiming that Gameworks Just Werks™ and AMD just needed to make better GPUs to deal with the retarded amounts of tesselation Nvidia was forcing onto developers
>Nvidia fanboys will now spend years bitching that developers need to specifically go out of their way to make their $650 ti cards better than $300 AMD cards
>>54341424
show proof
Poolaris in the loolaris
>>54338551
>1080p
2010 called, they want their monitors back you fucking pleb
>>54341492
his whole argument revolves around developers optimizing for dual gpu setups within consoles which will let amd produce and sell dual gpu graphics cards (on single interposer obviously) and those console optimizations (via dx12) will then carry over to amd's new graphics cards and allow them to compete with nvidia in the high end market for much cheaper. he even goes on to say that nvidia won't survive because of all of this occurring.
now we both know that the next consoles will be coming out soon. whether they're just simply upgrades to the ps4 and xbox one or they're going to be released as a whole new console gen will remain to be seen but the fact he is that neither of these consoles will be using dual gpu apu's from amd. this means that for the foreseeable future, 5+ years, amd won't be able to release these dual gpu desktop cards because developers will have absolutely no incentive to waste their time optimizing for it (because the consoles wont have dual gpu). this whole business move for amd will have to take off asap otherwise they'll keep dropping in value and nvidia will keep stealing more and more of the discrete gpu market off amd. they would have to shake up nvidia literally RIGHT NOW with this dual gpu dx12 stuff but unfortunately for them they can't do it for another 5 years at least and by then nvidia will have their own solution/answer to this.
kek
>>54341717
>his GPU die is over 300 mm2
>>54341744
how is this relevant? i've been using ati since amd bought them.
please kill yourself you deluded shill or just accept the facts that nothing amd can do will make them dominate the market.
>>54342288
who cares about dominating the market, all we want is competition and price war along with delicius 14 nm polaris
>>54338551
Playing on 1080p with an 8GB card and some recent titles (Arkham knight, ROTTR) have used over 6GB...
I suppose more VRAM isn't needed though, that's why we're still using 256MiB like we were in 2010 for 1080p.
>>54342342
Rainbow Six: Siege, 1080p high res textures everything maxed 6+GB
>>54332224
AMD hasn't had driver issues for years now...
>>54332592
interposer, maybe?
>>54324511
hory shit
>>54324511
whats even the point of Fury X lmao
>>54333697
also HDR with AMD cards
>>54342659
So shills will buy them and keep them afloat while they spend all their time not caring about PC hardware and working on console hardware.
>>54334004
they want to diversify, it's natural, they're about to loose a large percentage of their PC market
>>54342665
>also HDR with AMD cards
lolol
> HDR
> HDMI 2.0b
HAHAHAHA
HDR is not supported until HDMI 2.1. Go read a fucking spec once in your life.
>>54340483
>DX12
>Nvidia GPU
Pick one cause the jews at Nvidia won't let you have both
>>54342702
If you cheap out sure..
> not going for >= 980 Ti
>>54341717
>now we both know that the next consoles will be coming out soon. whether they're just simply upgrades to the ps4 and xbox one or they're going to be released as a whole new console gen will remain to be seen but the fact he is that neither of these consoles will be using dual gpu apu's from amd.
The ugrades might be coming soon, but they're not the next consoles. You're talking about something else.
The upgrades will only have a slightly faster GPU yeah.
>>54341530
underrated meme
>>54342690
>what is displayport 1.4
>>54342777
a connector that nobody uses
>tfw new GPUs are still about 2 months away
>>54342803
>le goalposts
>>54342857
I know that feel, Anon. Feeling empty that my wallet is full.
>>54342857
>tfw my OC'd 980 Ti in SLI will still BTFO out of anything AMD has to offer up until 2018
Upgrading as soon as the GP100 cards come out.
>>54342857
>tfw top tier GPUs (from a non-shit company) are still 8 months away
>>54342913
try 2017 familia
>>54342913
Until you turn on DX12
>>54342957
Vega 10 will be shit. Calling it now. GP100 will cakewalk all over the motherfucker and AMD will have to go back to the loo to figure out a way to beat the monster that is Pascal+HBM2
>>54343027
>dieshrunk Maxwell + HBM2
I'm sure AMD is scared shitless.
>>54343098
If they aren't, then why aren't they releasing a card to compete with the 1070 and 1080?
They're releasing a shitty $200 card instead because they know that they'll fall flat in the face of actual competition.
A confident AMD would have launched a 490X and 490 to answer to the 1080 and 1070.
>>54343149
>They're releasing a shitty $200 card
They're releasing a $300-350 and $150-200 cards.
>>54343149
They're targeting different markets, and they're waiting for HBM2 for their high end cards
>people think Nvidia will release the 1080Ti/Titan before next year.
Hahah oh wow
>>54343149
>$300 for a similar performance of a $650 card
>yeah amd is done
I think i'm missing something.
>>54343149
This card had something of a $350 MSRP(and this was before prices dropped when Fermi launched)
Price means little, AMD is competing with smaller dies again like they should have done for the last several years.
>>54343027
>GP100 will
You realize that GP100 is a pure GPGPU chip for enterprise, right?
it shot its transistor load on fp64 and bigger register files, not more shaders.
~4k shader @ 1.3GHz-1.4GHz(turbo) is only marginally better than overclocked 980Ti or even Fiji for fp32 stuff.
You had better hope that GP104 or 102 or whatever ships with >3k shaders, or else Nvidia won't have a part really upgrading to.
Do you guys think the Nano will drop in price despite AMD waiting till next year to release their high end cards?
>>54343457
It's already $450
>>54342904
>tfw poorfag
>can't afford a brand new GPU right now anyway
>waitfagging for at least another 4 months
>>54343457
Yes
>>54333697
I bought the viewsonic xg2401 for its freesync. Worth the price for a tn performance monitor.
Adaptable sync technologies need to be included on lower price monitors, otherwise it will continue to be niche hardware. Freesync over hdmi 2.0 looks promising.
>>54339369
I got the new hitman and farcry primal from a newegg promotion. I waited till my card came in the mail before redeeming the codes, but I got them pretty much instantly. Sorry to hear yours is not as fast.
>>54320902
> CONFIRMED
> Estimated Die Size
Nah bruh.
But if we look at resistors size, then yes, it's a small chip. Let's see if AMD can deliver.
>>54343287
Even if AMD releases a 5870 tier card, most people will still buy from nvidia even if nvidia's cards will be inferior. I can't really explain that though.
>>54343287
Still have that card. It was glorious! Keeping fingers crossed that Polaris will deliver like this mother fucker did years ago.
>>54345648
i have the promo codes. its when i enter them on the amd rewards website it makes me provide extra info for some reason. i've sent the proof of purchase form quite a few times with no reply.
>>54345993
It's called brand loyalty and fanboyism.
Guess AMD won't be making 550mm2+ GPUs anymore, they'll probably counter whatever GP100 derivative for consumers is with a 400mm2 Vega 10
>>54347261
It won't even be 400mm^2, it will be smaller.
Polaris 10 has 2560 shaders, Vega 10 has 4096, that's a 60% shader increase and for that you don't need more than a 50% die, especially once the GDDR controller gets replaced by HBM one and not needing to duplicate other stuff.
Vega 10 will be around 360-370mm^2
>>54347309
>360-370mm^2
>with HBM
AMDelusional
It will be about Fiji sized and the yields will be terrible, as per AMD's usual standards.
>>54347339
>that little amount of shaders
>needing 600mm2 on 14nm FinFET
>>54347339
>Fiji
>AMD's usual standard
Delusional, Fiji is an experimental chip for AMD and which they'll not repeat in any short term.
Reticle limit dies aren't worth it.
>>54347361
14nm is not 14nm
It's more like 20nm with Samsung's process. 4096 shaders will need at least 70% more die space.
>>54347339
Fiji directly scaled down to 14nm would be 180mm2. HBM's memory pads are much, much smaller than what GDDR5 uses. The inclusion of HBM on package in absolutely no way indicates an inordinately large die.
Tech illiterate shitposters need to stay in /v/.
>>54347419
Polaris 14nm != true 14nm, dumbass. Only Intel makes true 14nm dies.
>>54347413
>>54347450
>I mindlessly regurgitate things without understand any of it
Gate length on Samsung's 14nm LPP is 14nm. Chipworks has dissected A9 chips as reference. Back end scaling has nothing to do with front end feature size.
Tech illiterate shiposters should stay in /v/.
>>54347413
Please stop being retarded, Polaris 10 has 2560 shaders in 234mm2, Vega 10 has 4096 shaders without the large memory controller/phy, just increasing the ALU count alone by 60% doesn't need 70% more die area, that's ridiculous.
Anyhow, you're still there floating around 380-415mm2 even with 70% more, which is absolutely nothing compared to 600mm2 still
>>54347461
Desperate AMDshills are coming out of the wood works.
They can't come to terms with their retarded company going under from poor business decisions and shady business tactics in the past. Remeber TressFX?
>>54347495
AMD isn't dying until 2020, you'll have to hold your sore butt until that time.
>>54320902
What really matters is they have those motherfucking 10%. As long as it's there, there will be people willing to buy it.
>>54347545
Just overclock it if those 10% mean something to you.
>Q2 2016
>we still don't know the ROP count or L2 cache size of Polaris 10
>>54347590
>GCN
>overclocking beyond 5%
top kek
meanwhile, Pascal is an overclocker's dream because it's based on Maxwell with refinements towards operating at higher frequencies
>>54347495
>even more tech illiterate shitposting
This is a 14nm Exynos 7420. Samsung's gate length is 14nm.
>>54347261
>GP100
>600mm^2
>300W
>Consumers
There will not be any consumers chips that large until 2018
>>54347632
>you
>not retarded
Only thing that doesn't overclock is Fiji, Tahiti, Bonaire, Pitcaim, all overclocked great, Fiji is already overclocked to compete with GM200
Also, 10% Hawaii overclock means more than a 15% GM204 overclock, since AMD's architecture has higher IPC
As stated before, Fiji was a mistake, AMD will never make reticle limit dies again unless OEMs request it.
>>54347687
Derivative I said, Nvidia can rip out the FP64 units out of the GP100 and the TDP and diesize will drop notably.
>>54347726
>Only thing that doesn't overclock is Fiji, Tahiti, Bonaire, Pitcaim, all overclocked great
What is this sentence even? Fuck.
>>54347770
I'm writing from bed so not much thought goes to my sentence structure, you can still understand just fine. Please deal with it.
>>54347765
It would still be way too large to release to compete with Vega. Just look at 28nm and how long it took for big consumers chips to arrive there.
>>54347815
No, I really do not. I do not know what overclocks and what does not. Just put a fucking period in there or something.
>>54347831
Fiji does not.
Rest do.
Honestly if you actually followed GPUs over the years you'd know what chips overclock and what do not.
>>54347831
>I do not know what overclocks and what does not
Here
>Fiji: buy a 980 Ti
>Hawaii: buy a GTX 970/980
>Tahiti: buy a GTX 960
>Bonaire: buy a GTX 950
>Pitcairn: buy a GTX 750 Ti
>>54347856
I'm sorry that I did not?
>>54347870
Should you be arguing about this topic then?
>>54347907
This topic being the thread's topic, or what I brought up (your sentence)? I see no reason no.
>>54347868
you mean:
>AMD: buy literally anything else
>>54347928
The overclocking topic, but die size fits under that as well, smaller dies do overclock better in general so a topic starting with diesize would swerve to OCing sooner or later.
>>54347968
>The overclocking topic
Then no, but I'm not arguing about that anyway, so why even ask then?
>but die size fits under that as well, smaller dies do overclock better in general so a topic starting with diesize would swerve to OCing sooner or later.
I disagree with your implication that you need to know which previous GPUs could overclock well to be able to discuss die size of new GPUs.
>>54347994
I wasn't arguing with you about die size, I was arguing about you about what overclocks and what does not.
That is if you are the one who said that "GCN can't overclock more than 5%"
If you're not, I apologize.
>>54348029
>That is if you are the one who said that "GCN can't overclock more than 5%"
Clearly I am not. I thought this would be obvious at this point.
>>54348051
Okay then.
Does anyone know the size of a 256bit GDDR5 memory controller?
ATI BRAND REVIVAL WHEN?
>>54348093
Really big, around 40mm2
>>54347726
We'll most like see xbox hueg dies by AMD again once we hit another 5 years on the same node ala 7nm
>>54348244
A "big" 7nm die would only be 150mm2 though, and larger wafers are going to be in use by then so they're going to be stupid cheap.
>>54341078
google maps always recalculates, far better then my stand alone gps,
on that note, already have self driving cars, and they have gotten into no accidents of their own, only accidents of others.
we will likely see something happen consumer side within 10 years, they wont be low end cars, but god knows i would be willing to pay a fuckload of my time commuting back, especially if i can live somewhere far cheaper.
>>54348302
at 7nm, there is no where else to go except for parallel processing or a new material.
>>54348302
True, if they're using EUVs the reticle limit of the fab might change, 150mm2 is still too small, I wager somewhere around 450mm2 before cost and defects go lunatic
What happened to moving to 450mm wafers? Intel was talking about it a few years ago that they're ready and there's still no word.
Does ASML not have the machinery ready or the fabs just lazy?
>>54348409
Lack of consumer demand actually, ASML stopped producing 450mm equipment because the fabs calculated a small die cost saving(10-20%), so it's not worth the initial investment. Unlike the 50% cost saving per die when the move to 300mm happened
>>54348409
>>54348511
Back in 2013-2014 the lingering Common Platform Alliance started getting cold feet on 450mm. After a while they all basically said they were "shifting investment focus" or something along those lines. Though it was understood that 2018 would be the year when they finally put it into place after much pissing and moaning.
They've got to do it eventually, and anything that lowers the price of commodity chips is going to end up on the table.
>>54343149
>then why aren't they releasing a card to compete with the 1070 and 1080?
polaris 10 is in the $300-350 range which should be competing with the 1070, taking the 970's spot
only way that wont happen is if nvidia goes full jew and prices the 1070 as a 980 and the 1080 as a 980ti
>>54348907
Which they probably will, only to cut prices if AMD manages to put out something competitive.
Threads's been up for almost 2 days
>>54347495
>remember tressfx
>in the age of game works
Yes I remember amd developing an open source rendering technology and then proceeding to lay the framework for an open source cross platform api to compete with directx
>>54320902
To be fair... they could now afford to use bigger die due to HBM being much smaller than gddr5
>>54349777
Polaris won't use HBM.
>>54349833
he was talking about pascal, which probably won't have HBM2 right now anyways because there's not enough of it
>>54348968
i only doubt it because otherwise nobody will buy the new cards
if people were gonna drop $500 on a card they would already have 980's; the only way new cards sell is if they offer more performance for less money
the people who buy the $300 cards wont suddenly buy $400 cards just because a lower option isn't available
if anything the 970 will get lowered to like $250 and sell like fucking free beer and hookers
>>54325382
>Mobile GeForce 8000 series de-solder themselves http://forum.notebookreview.com/threads/gpus-igp-nb-desoldering-themselves.582158/
I FUCKING HATE THIS ALL THINKPADS WITH GPU'S SHIPPED WITH THOSE POS AND KILL THEMSELVES
FUCK NVIDIA
>>54325822
>Neither companies are dumb enough to fall for those same mistakes again
Both have fucked up the same things more than once
>>54325980
>AMD fucked up their drivers a while back and fried cards.
That was literally third party overclocking software and it pretty much never fried any card
>>54350732
>I FUCKING HATE THIS ALL THINKPADS WITH GPU'S SHIPPED WITH THOSE POS AND KILL THEMSELVES
mine still works :^)
but really I bought it refurbished in 2013; it's probably already broken and repaired
>>54350749
yes and ironically nvidia actually did this like 3 months ago lmao
>>54340174
>still waiting for that glorious AMD showstopper
Literally happened twice in this decade
It seems you're pretty new friend
>>54350789
>using the smiley with a carat nose
>>54351614
>using the >using the smiley with a carat nose
>>54350837
It did?
>>54320929
Isn't Computex May 31st? Only a month
>>54351807
Paper launch, you could at best hope for two weeks for the real launch, since its new fabrications for both companies it might take a month or so
>>54352130
Ah, gotcha
>>54349009
7nm, possibly 6nm is the limit of silicon due to quantum tunneling, unless this is solvable it becomes to much of a problem passed this.
graphene is what we will likely move to, probably going back a good number of generations on die shrinks, but getting 50ghz on air cooling along would make up for it.
>>54350603
depends entirely on what else is available. remember, the 700 and 900 series nvidia burned a fuck ton of bridges.
if amd puts out a fury/980/980ti ball park card in the 300 or less range, which they could, then all bets are off on nvidia.
hell, look at the cern leak for the zen, 2 16 core cpus make up the 32, amd is gearing up for multiple chips on one die, navi may just well be multiple smaller gpus on one die... possibly one strong gpu with multiple smaller just for the sake of backwards compatibility.
>>54350789
nvidia does that whole gpu killing driver every few years, the last gpu driver killed the display along with the gpu if i understand what happened correctly.
and nvidia never needs 3rd party software to do this.
only amd gpu that i saw that died, died on fire due to the chipset that was controlling it, some jack ass tried to pass it off as amd did it when it was a motherboard chip that fried it self, stated on fire (intel by the way) and took everything in the pci slots with it.
>>54345993
Sad, but true
>>54347495
You've been proven wrong, quick point to something else!
>>54353572
Nvidia have the marketing. They've already been sending out these briefcases to reviewers with an encrypted message which hints towards a Friday reveal, which also coincidentally is the day DICE are revealing battlefield 5. They're probably going to show the gtx 1070 running battlefield 5 at 4k 60fps maxed out or something. They're already trying to win over all the casuals for these next generation of graphics cards whereas AMD are doing practically fuck all with their Polaris marketing.
>>54353740
Source
http://www.guru3d.com/news-story/nvidia-starts-the-hype-teases-pascal-with-order-of-10-puzzle.html
>>54353095
in the future consoles AMD will build 2-4-8 smaller GPUs into the socket which will force game devs to build their games ground up to run efficiently in Linked GPU multiadapter mode which once it happens means that scaling GPU power will no longer be about building larger dies but yielding the most performance per mm2 of space and per watt and then just fabing that chip and slapping however many of it you need on to an interposer with HBM to scale up performance. This is what navi will be and this is why Nvidia is getting out of the gaming GPU buisness and moving towards high DP workstation GPUs for tesla and others.
Volta the cards after pascal might actually be the last GPU's nvidia ever makes for gaming
>>54353740
>FPS game
talk about easy mode I bet even polaris10 with its 256-bit bus can manage that
of course they wouldnt showcase a next gen open world game that fucking eats resources and bandwidth which what actually matters when comparing gpu's
where's muh 490/490x
>>54353740
inb4 paper launch just to preempt AMD's reveal at Computex.
Being first is good marketing, even if you don't have any product to ship to customers.
>>54353893
this is why AMD won. People dont understand that the Mantle API that AMD created was a paradigm shift in the gaming industry. It isnt winning by gimping games through gameworks its winning by opening the programming base allowing the devs to utilize hardware in ways it never could before.
This API on consoles and forcing devs to optimize for multi-adaptor is going to finally replace SLI/crossfire but treating multiple GPUs of identical spec as a single larger GPU with multiple command processors and memory regions that can communicate and share resources.
AMD said they want to power VR thats 16K resolution in each eye and running at 240hz with realistic color mapping and HDR lighting this is how they will one day build the GPU thats gonna do that.
the new GPU's will be two rows of power conduits and capacitors sandwiching a rectangular interposer with a series of small GPU dies and HBM modules meshed along its length.
>>54353893
>it's another adored tv shill
None of what he's saying will happen. It's all been disproven. You can't just say stupid things like that without looking at the bigger picture. For shills it looks like the "master plan" but for anyone with a brain that isn't deluded by this amd marketing knows that this will never take off for multiple reasons and nvidia will never be leaving the gpu market like adored says they will.
>>54353941
Who cares. Casuals don't care about that. They'll just see some nvidia card running a game at a hyped resolution and will probably go out and buy nvidia because of that. Not everyone is educated when it comes to graphics cards, or even consoles for that matter. I know people who really believe the 'ps4k' will actually run games at native 4k.
>>54354125
the fuck is a ps4k
>>54354079
You mean that youtube runt? Who cares for his opinion, why is he even brought up here? This isn't fucking /v/, take your e-celeb niggers there.
>>54353893
>>54354051
It's that deluded kid from the other day. He does it for free guys.
>>54354129
Not him but apparently sony is coming out with a new PS4 that will have beefed up hardware inside to allow it to do VR.
I have to go look up the specifics though.
>>54354134
Go back to bed.
>>54354129
>>54354167
https://youtu.be/j-3SJCzIQZw
>>54354125
Well, it might, with severely reduced settings and framerates, if this was a real thing. That's several doubtful guesses followed by something that's almost certainly not the case.
>>54323876
But I see things like this being important with 80W setups, and wouldn't be surprised if it still mattered as low as 60W to some extent. We have to compensate for chip size vs wattage, because if we drop the chip size faster than the wattage then the core chip temperature can climb very fast and high.
Also while the computation per watt has actually increased thanks to better technology. Those gains are not as big as marketing would have people believe. Some of those gains are actually thanks to software optimization, which has huge potential over most hardware right now.
And nearly all of these low power consumption chips are artificially limited to create improvement in other areas. Because the hotter the chip gets the more energy it takes to compute which means more energy goes to heat, thus is the nasty feedback loop every electrical system has to deal with. Well unless we get super conductors, which is why these improvement in conductivity direction has been critical to get these real gains we have now.
(Much like an acceleration governor on a car, sudden boom in fuel economy at cost of acceleration without much change in hardware) We are really running in to very hard physical limits and we are tweaking around the issues rather then admitting it.
>>54353893
issue is backwards compatibility
unless there is some hardware feature to make everything work as one, it will seem like 2/4/8 gpus, each one with various amounts of ram...
imagine a gpu, with 4 dies on it, each one is just under a 280 in power, but they scale so fucking good, its above a 980ti
in a pre dx12/vulcan (ie all games) that will run at best as crossfire, now imagine each gpu has 2gb of ram, that may be 4 way crossfire, but its still only 2gb of ram. it would run like shit.
whats more likely is one gpu that is fairly beefy, mixed with other more minor gpus because i will tell you now, no one will buy a gpu for future games if it cant run the current ones for shit.
we need more info on how navi will work before i get excited because it could very well be a massive step backwards as we go forward.
Polaris and Pascal are both garbage.
Vega fucking when?
>>54354079
look at the 32 core zen, its 2 16 core apus stuck together... honestly, i see cpus and gpus going this route, at least on amd's side.
remember the 3% yield, if gpus worked then the way they can now, that would have never happened.
>>54336268
We are talking about adding roughly $40 to the end chip cost to get some of the highest performance gains the market can yield.
Sure it is expensive, but that is not much for the higher end markets, especially with what you get for that extra money.
>>54354367
no, the bigger the chip the lower the yields, the more expensive the chip is to produce, the more cost it is put on you.
right now you can get pascal cards, 8 for 140,000$ if i remember the price right, this shit costs a fuck load more then you think.
>>54354344
I've been saying this all along. If you look at the bigger picture, this multi gpu shit will be a failure because it could potentially prevent older games being played because of shitty sli/xfire optimizations within those games.
Also don't forget that Nvidia gets the majority of AAA game sponsorships for the PC. I can boot up any random 10 AAA games from the last year and most of them will have some kind of Gameworks feature, whether it's full blown PhysX or just HBAO. These fanboys think that Nvidia will get blown out of the market because of these multi gpu consoles and their games being ported to multi gpu amd desktop cards but if nvidia get most of the sponsorships (which we know they do) then these devs will be forced to optimize for those nvidia cards. It doesn't matter if the devs don't want to do it, if nvidia is offering cash for the sponsorship then they'll have to do it. Look at games like hitman. The devs have come out and said async was practically pointless because of how much work they put into optimizing for async only to get a 5% performance increase. They only did it because AMD forced them to with their sponsorship of the game. If it was up to the devs they wouldn't have wasted time with async because they'd know nvidia has 80%+ dGPU market share and wasting time with it won't provide any massive benefits for the majority of dGPU consumers. Money talks though so they had to do it.
Am I autistic if I buy exclusively AMD GPUs simply because Nvidia's GTX garbage can only support 4 monitors without SLI?
>>54354468
How do you justify having over 4 monitors?
>>54354440
I love how every amd faggot seems to think that nvidia can't play the same game. If devs start supporting multi-gpu better through dx12 and vulkan, they'd be all over it cramming as many dies as possible into something they could sell as a premium
>>54354501
How do you justify having over four dollars?
>>54354367
I dont think you realize how it works. As chip size gets bigger, the odds of an error that fucks the chip irreversibly goes up exponentially with each square millimeter added to the die area. On top of that, the larger the dies get, there is a inversely proportional number of dies possible from each wafer.
Intel's 18C Haswell-EX die is so large they cant even get 100 possible chips out of each wafer. As each wafer runs about $4000-$5000 USD, and out of that maybe 20 dies out of 88-90 may be viable. Just to recoup the cost of the wafer itself, each viable chip, full fat or not, has a base price of about $250. Then there's materials (keep in mind, there's a number of rare and/or toxic materials used in semi-conductor construction, and that's not including the gold used in every single chip's packaging), processing, power consumption for the equipment used to manufacture the chips, cash that has to be paid to employees, R&D recoup money, all of these costs that MUST be covered before the margins are added so they actually make a fucking profit off them.
There's a reason why AMD's Fury series of GPUs were so fucking expensive: They're huge, and the yields consequently sucked.
Its also why we gave nvidia so much shit during the Fermi launch, they made this huge die on a new immature node, and the yields were beyond terrible. They couldnt sell any full-fat GF100 chips to consumers because the very very few that they got (1.7% yields for viable dies), if they even got one from a wafer had to go to Tesla cards.
>>54354501
I would try to explain it but I'm afraid you could not possibly understand.
To oversimplify it's basically my fetish.
>>54354440
Nvidia has roughly 60% of the install base counting workstation.
>>54354611
http://wccftech.com/nvidia-amd-gpu-market-share-q4-2015/
DICE dropping AMD?
No way, or Nvidia paid them very many moneys.
This feels sad, really.
>>54354658
>source specifically says shipments
>I said install base
baka desu senpai
Also MR is far more accurate on shipments than JP
>>54354687
These big developers will be able to predict how the market will pan out and will base their marketing decisions on that. DICE are obviously are predicting nvidia will make them more money and will probably expand their player base more than AMD could have.
Activision did a similar thing when the current gen consoles came out. They predicted ps4 will be the more successful out of the two consoles and ended up ditching MS for Sony, which has clearly paid off because they're getting a fuck ton of sales with all those 35+ million ps4 users buying up their overpriced CoD DLC.
>>54354389
That portion of cost in relation to yield loss is more directly linked to the fabrication processing then materials. Which is why the percentage of fabrication costs have been climbing as material cost have been falling. The newest and smallest fabrication size has the lowest yields as the technology is not mature and making anything that small is hard.
So making a die bigger, would mean more complex fabrication and huge yield losses dramatically driving up cost. Just as you say. But that assumes that space is packed with additional transistor. If the transistor count is relatively lowered then that also lowers fabrication complexity, it still will be complex as the transistor is still small but lowering the relative amount of them means there are less things that can go wrong during fabrication. In addition that "empty" space could be marked with tool alignment markers (much less complex then a working transistor) making the fabrication process easier allowing increased yields.
It is counter intuitive but the economics makes sense. As you are basically paying for more (blank) wafer material to get better thermal control, so conductivity and process performance can be gained. (and blank wafer costs have dropped a lot thanks to growing demand with the solar energy market sharing some of the material market leading to a boom in supply)
>>54354538
see my other reply
>>54354802
>>54354580
Try me.
>>54354817
Except that doing "blank" space like that completely nulls out one of the major reasons to go with smaller transistors in the first place, that being you can get more chips-per-wafer. You're wasting perfectly good silicon for no reason. Whether its a 250sq mm 2B transistor die completely packed or a 300sq mm 2B transistor die with a blank area around it, the center still has 2B transistors of which all must be functioning properly.
Its also nulling out another reason why we go to smaller transistors: clock speed. A smaller more compact die means a signal going from one end to the other takes less time to get there (even at .7c the propagation time from one end of the die to the other is a non-zero number), which means the clock signal cannot tick again until ALL ends of the clock tree have received the pulse. If the die is physically made larger, it takes longer for that clock signal to get from the clock input, through the various PLLs that control the on-die clocks, and then to the flip-flops they need to get to.
Yes, architecture has something to do with it (looking at P4 with its ridiculous clocks and double-pumped ALUs), but ultimately its up to the density of the transistors, which means the transistors must be packed together as much as possible without them interfering with each other.
And as for the alignment markers, they're already there. Look around the edge of this Westmere-EX die. You see all those tiny little squares, especially on the very left, right, and bottom of the die? those are the alignment indicators.
>>54337899
In my fab a 8'' wafer is $25-50k in ICs.
>>54354440
first off, while i agree a multi gpu gpu will fail at older games, this is only for the time being, and could be fixed for the most part with proper crossfire drivers/support.
if the gpus have 4gb of ram per gpu, or even just 2 have a 4gb pool, there you go, damn near any game made to this point will be good to go.
then you have every console is using amd cpu and amd gpu, and the consoles will move to whatever is most efficient.
nvidia, while capable of trading blows with amd, doesn't want to anymore, preferring high volume high price pro cards/applications.
you may see nvidia tag around in gameing passed what that guy said, but likely not with more then one chip sku,
there will be the titan, the ti which is slightly cut down, the 80 which is cut down more, and the 70 which is severely cut down, all of which will just be test beds for pro cards.
>>54354538
just so you know, the 28nm was mature, the yields of the fury weren't bad, and the card itself wasn't expensive, it only seems that way because nvidia sacrificed the titan s price point to counter amd
other then that you are correct in everything.
>>54354982
Actually I am adding the "blank" space mainly to increase clock speeds, and would be spaced between transistors not around the die itself. Don't misunderstand, dropping the distance traveled allows things to get there faster. But I am looking to increase speed and increase conductivity to more then make up for that extra distance.
Also markers aliments around the perimeter are used, but by spacing them through the die the distance to each focus check point is smaller which helps. Though admittedly how much it helps is probably debatable given more modern manufacturing (old accuracy vs precision debate) .
>>54320925
any card you get within the next 3 months won't be HBM2, so why bother?
I'm sitting on a 980, waiting for HBM2 cards to come out (probably early-mid 2017). Shit can't come soon enough, but anything without HBM2 is missing out on this iteration's best feature.
>>54355333
coming from a 780 i'd take something with gddr5x for a while, was close to buying a 980ti until i started following the 'new shit' news & rumours
>>54355125
It likely wont help at all in terms of accuracy considering that right now they can make 14nm processes using existing (albeit) modified methods. On top of that, running markers through the transistor area would eventually run into the problem of the Photoresist and later the traces of the metal layers getting in the way. On the edge, they are always visible because there's nothing fucking there.
And adding space between the transistors would only help in the fact it would lower the interference between the actual gates, which, considering the fact they work at 14nm w/ the current densities, means that it would not help much at all. The core problem still remains, making billions of transistors that small and consistent is enormously difficult, and silicon is starting to reach the limit to what its capable of. Chip manufacturers need to move to Gallium-Arsenide and later something carbon based if we're to keep going smaller and faster.
>>54355076
>just so you know, the 28nm was mature, the yields of the fury weren't bad, and the card itself wasn't expensive, it only seems that way because nvidia sacrificed the titan s price point to counter amd
I assumed the yields were terrible simply by the fact the Furyx2 was coming with a third fully assembled Fury package in the box. You dont exactly do that unless there is something spectacularly wrong with a large number of the dies that were being put out by the fab, mature node or no.
>>54355484
>Furyx2
what is this about a third one?
i honestly know nothing about the furyx2 other then its a fucking amazing gpu when power is concerned, it has 2 gpus in the power envelope within 20 watts of 1 980ti
outside of that i know nothing else.
>>54355333
because who cares if it has hbm 2 or not, if the cards coming out, at least amd's offerings are as good as the demos show, and sub 300$, they may be well worth an upgrade to, i don't get the highest end gpus as i cant justify the cost, i also wont be gaming at above 1080p, as a 4k monitor for me would be around 100-120 dpi,
>>54355484
I think the flaw isn't in the fab but the interposer stage.
>>54355526
It comes with another Fiji die in the box for display purposes
>>54355553
>i don't get the highest end gpus as i cant justify the cost
in which case you want the most bang for your buck.
HBM2 doesn't just affect the resolution you can game at. It defines how fast data can be manipulated by the card. That shit is important.
but ok.
>>54355526
For Furyx2 cards that arent part of a VR box, as part of the packaging, there is a third, fully assembled Fury package, HBM and all.
I also question AMD's reasoning behind the $1500 USD price tag.
>>54355563
I dont see how the interposer is the issue, its made (IIRC) on the 65nm node, which has long since become a mature node, and is passive to boot (no transistors). Yeah I can see the TSVs for the HBM being an issue, and the fact the die for the interposer is fuckhuge by itself, but as a passive component on a very old and mature node I really dont see how the interposer itself is an issue.
>>54355484
As I said the markers gains are debatable, so they are not were the real gains are going to come from. Also there are other ways to mark things that don't use metal layers.
It would do more then lower interference, it would deal with heat issues. Given my previous posts get such replies, I will try simplifying it even more by suggesting people think of it more as a very powerful integrated heat spreader.
As for limits, this does not magically break them. Only better aligns the chip to the typically usage profile. Which will improve end user experience. Making anything 14nm will always be hard.
As for materials, Gallium-Arsenide has proven to surpassed silicon. But carbon will never be a base because of its proprieties, but it could act as a highly conductive layer to get gains be addressing heat issues. The government already uses a diamond layer for some special application chips just for that very reason. I suspect we may someday see that eventually as diamond prices continue to fall.
>>54355734
>As for limits, this does not magically break them. Only better aligns the chip to the typically usage profile. Which will improve end user experience.
See, the problem is, while this could, in theory, work for consumer and mobile chips because they have the size to spare and because they'd actually gain something out of it, chips on the other end of the spectrum, the >500mm^2 behemoths, they cant afford to widen the space between transistors, as they're already huge as it is. Lowering transistor density while keeping the area the same means that the chip would (obviously) have less transistors, which would more likely than not translate to less performance, as entire blocks of logic are cut away. Going to multi-chip solutions on the same package presents its own issues.
Data centers always want more density w/ less power, which is why we've reached the point where there's dozens of cores per server. Developing a dense node for enterprise and then making a non-dense node for consumers simply doesnt make sense. You're wasting money on R&D that could be put towards other things.
As for Diamond, yeah, i had to check on that. Diamond prices are also irrelevant simply because, if needed, they could build the desired carbon lattice structure directly.
>>54354798
On the other hand it will be hilarious if those to events are separate and DICE will run Bf5 on Polaris.
>>54356063
its too much of a coincidence for them to be on the same date. it makes more sense for nvidia to use the first AAA title this year as their marketing tool. battlefield is way more popular on pc than it is on console so nvidia want to grab those extra consumers. they'll probably throw in a free copy with all nvidia cards like they did with the division.
>>54355970
Well it actually applies to even for the big industries as well, if core tech is aligned properly. The IBM z13 actually appears to incorporates many the same principles, but people (including myself) aren't as familiar with such new mainframe setups, given we live in the age of severs. But obviously the more targeted the market the more strain on the R&D budget. That is were chip grading and over lap makes sense, something that is already done. The real problem is the true market costs of maintaining such an industry. (as we have seen the the iPhone drop, endless yearly refreshes can only do so much for such expensive stuff, partly why I think a small number of chip types with more focus on software is the future as software companies cost much less to maintain)
As for diamonds, they are really cheap to make. GE actually built a machine to make diamonds. But De Beers has the global diamond market so jacked up with their monopoly that even the technical markets are only a fraction of what they could be. But if people want it badly enough that artificial barrier will break.
>>54355575
hbm 2 will not be on ANY low end card, nvidia will push it to the highest end if it even comes out on consumer cards, for amd, its possible a refresh of the lower end line up would happen, but i doubt it.
i honestly don't see the memory as important unless the card is bandwidth starved and sense we know noting about any of these cards in benchmarkable situations, the whole wait in hbm2 is retarded.
>>54355649
2 extremely fucking highly binned fury's, remember, this is 2 gpus that in total are 20 watts higher then a 980ti.
2 furyxes at original price was 650 each so 1300 for two, then you add in its two on one card, and that its a niche... and there is the extra 200$
remember, when nvidia did this with the titans, they charged 3000$ for it, that would be like if the fury x2 cost 1950$
>>54356230
>2 extremely fucking highly binned fury's, remember, this is 2 gpus that in total are 20 watts higher then a 980ti.
>
>2 furyxes at original price was 650 each so 1300 for two, then you add in its two on one card, and that its a niche... and there is the extra 200$
>
>remember, when nvidia did this with the titans, they charged 3000$ for it, that would be like if the fury x2 cost 1950$
Ah, I see. Still a bit on the expensive side, and way too frickin late to be any good. If they had released it between Fury/FuryX and Nano, they'd be in a much better position.
I'll admit, I'm kinda interested in one for my rig just for the novelty, but it'd be wasted on a 1080p screen, and I'd have to spend an addition $150 or so to get a PSU that could support the beast, as my Seasonic G550 doesnt have a third PCI-E power cable.
And speaking of the Titan Z, I looked them up on fleabay just for kicks, they're still about $2500 for the cheapest ones. Da fuq are these people smoking?
>>54356334
>I'll admit, I'm kinda interested in one for my rig just for the novelty
senpai I feel this way about getting a pair of nanos and building a custom water loop for them and cpu
but it's hundreds of shekels i could put to better use elsewhere
>>54356891
Same, $1500 would be much better spent on upgrading the system core, and I could still have a significant amount left over to get a GPU that is significantly better than the GTX 660 that I currently have.
One of the things I've been semi-seriously considering is going from an E5-2690 (SB-E, 8c/16t) to an E5-2690v3 (HW-E, 12c/24t) and a decent board. I dont really overclock anymore, and I definitely could use the extra cores/threads. 2690v3 is also relatively cheap, and isnt jacked up through the fucking roof like the v2 version is.
>>54356334
they do not run on consumer drivers meaning games stutter like hell
Too bad it will suck and produce shit frame rates.
>>54356966
>going from an E5-2690 (SB-E, 8c/16t) to an E5-2690v3 (HW-E, 12c/24t) and a decent board. I dont really overclock anymore, and I definitely could use the extra cores/threads.
Not a bad move desu, I should do the same but I'm coming from a 3930K
I've got a 3770K system sitting unused that I should do up for vidya instead of using my work pc,can clock it to the moon and watercool it instead. It's in a roomier case for that, too
![ASRock-M2-to-U2-Adapter[1].png](https://i.imgur.com/gcQz6LC.png "ASRock-M2-to-U2-Adapter[1]")
