It's confirmed, Vega is indeed a new or slighly tweaked architecture and not just big Polaris.
AMD has also confirmed a Vega 11 to replace Polaris 11.
Order of succession (according to cost and diesize)
Fiji -> Vega 10
Hawaii -> Polaris 10 -> Navi(2018)
Trinidad -> Polaris 11 -> Vega 11
Tonga stays where it is for now or Vega 11 might actually be a Tonga replacement according to diesize, not confirmed since no point of comparison.
Polaris 10 and 11 are confirmed for a June launch.
Vega 10 and 11 is confirmed for a Q1 2017 launch.
tl;dr
Polaris 10 high-performance GPU fighting off 980's sucessor
Polaris 11 low-midrange GPU fighting off 960's successor
Vega 10 enthusiast GPU fighting off 980Ti's successor
Vega 11 midrange-upperlow GPU fighting off whatever there at that time.
In short, if you want to replace your 980Ti's/Fury's you'll have to wait for next year.
Also the 'big' dies won't probably go over 450mm^2 at all since FinFETs are quite young to both TSMC and AMD.
But whoever manages to make a bigger diesize that gen wins it.
Polaris 11 has been confirmed around 120mm^2 already a few months ago.
Polaris 10 will probably be around 240-290mm^2
Vega 10 as mentioned will be around 400mm^2+
>AMD fags on suicide watch
>>53518430
Does this mean I should just go with either the gtx 970 or r9 390 now for ~$330 in preparation for VR? Amd doesn't look like it will be targeting that market in the coming months.
It's also hard to determine if liquidvr is better than what nvidia is offering without any benchmarks. :(
What do?
>>53518521
Don't buy shit now, it's the worst time to buy.
Both GP204 and Polaris 10 will be launching in a few months, and they'll rip apart both the 390 and 970
>>53518526
That's to be doubted.
AMD won't improve much other than efficiency.
And it will cost you much more than you would gain from.
With this, Nvidia doesn't see any threat and doesn't push for better hardware.
>>53518546
Performance comes from efficiency, what do you think 2.5x perf/watt means? 2.5x performance at the same wattage, or 2.5x less wattage at the same performance.
But GPUs aren't black and white like that, you'll likely see 40-60% more performance at less wattage.
>>53518560
>you'll likely see 40-60% more performance at less wattage
In your wet dreams maybe.
You will get 10-20% + performance of a R390X with a little bit less power demand for 600$.
That's how it will go.
>>53518583
No it won't, you'd get that from architectural improvement on the same node, but not with a two node jump.
You don't seem to realize that just that two node jump will increase the amount of transistors in the same area by two, coupled with FinFETs that lower leakage significantly that the current GPUs suffer from.
It's not joke when people say that this is the biggest jump since the HD4000 series.
>>53518608
>HD4000 to HD5000 series
Sorry, typed too fast for my brain to keep up.
>>53518583
>In your wet dreams maybe.
Hawaii is 40% faster than Tahiti at the same wattage.
Fiji was 30-40% faster than Hawaii at a lower wattage.
This was on the same node.
60% perf at lower wattage is an understatement.
>>53518546
what is node shrink?
>>53518643
A meme
The same company that rebrands their hardware for years will not replace a completely new GPU 6 months after release. This is bullshit.
>>53518758
Obviously you can't read because there's at least a year and a half between each upgrade in the OP.
>>53518758
D E N I A L
>>53518777
>June 2016 -> Q1 2017 is a year and a half
>>53518430
It's like your explaining the plot to some complex movie trilogy.
>>53518636
>Fiji was 30-40% faster than Hawaii at a lower wattage.
No, the 390X is only about 10% slower than the Fury.
>>53518789
Now I'm pretty sure you can't read for shit, the June 2016 GPUs aren't replacing Fiji.
They're replacing Hawaii and Trinidad
I fucking told you AMD fags that Vega is a new architecture. The graph was about new architecture's planned release.
But you faggots kept saying "N-NO! THAT'S JUST A POLARIS BIG DIE!".
What do if i want to replace my dead 7870???
>>53518800
OP clearly wrote that Vega 11 replaces Polaris 11.
>>53518797
Stop shitposting.
http://anandtech.com/bench/product/1513?vs=1559
Theres better islands than Trinidad.
>>53518820
>Tonga stays where it is for now or Vega 11 might actually be a Tonga replacement according to diesize, not confirmed since no point of comparison.
Taken into account, Polaris 11 -> Vega 11 might not be how it actually goes.
Also, they're lowend GPUs, why do you care?
>>53518831
>http://anandtech.com/bench/product/1513?vs=1559
I only see average from 10-20% gains.
And the recent benchmarks comparisons with AoS , Hitman and so on are even less than 10% difference.
>>53518831
>anandtech
>>53518796
>your
so whats the likely release date for vega 10? i was really hoping that polaris 10 was basically going to be vega 10 in terms of specs but now i have to wait even longer to upgrade.
So i have to wait till 2017 to get a good midrange card like 380 now? FUCK THIS SHIT.
>>53518831
>using anandtech unironically
>>53518877
Early February or late January.
When AMD releases in Q1 they always do at those times.
>>53518583
AMD is trying to drop prices under that $600 mark. Thats the whole aim of Polaris.
>>53518877
>>so whats the likely release date for vega 10?
February or March 2017.
>>53518880
That's what happens if there is no real competition on the market.
AMD fails to deliver, Nvidia doesn't see the need to deliver anything therefore as they are already controlling the market.
>>53518875
Dude, can't help it, I'm too stoned.
>>53518848
It's more than likely that Vega 11 is a Tonga replacement than a Trinidad replacement which got released in June anyway.
So pretty much a top to bottom FinFET lineup in half a year.
Not bad, for a company with so little money.
God dammit I don't know what to do.
Do I buy Polaris 10 in June or do I buy Fury/Fury X right now/in june
>Vega 11 midrange-upperlow GPU
I dont get it. Give me example of such a GPU right now please.
>>53518908
>buying Fury X in June when Polaris 10 will be around its performance, with new features and better power consumption and without a VRAM bottleneck
For what purpose
>>53518916
Tonga PRO
>>53518908
Id wait for june.
>>53518923
Very funny.
>>53518918
God, I'm getting wet when I imagine a 250mm^2 GPU beating a 600mm^2 in all metrics.
It's like Evergreen and Fermi all over again.
The future is fucking bright, what a time to be alive.
Expect
Polaris 11 ~$100 replaces R7 360 and R7 370
Polaris 10 ~$200 replaces R9 380
Vega 11 ~$300 replaces R9 390
Vega 10 ~$500 replaces R9 Fury
>>53518930
It's a lower midrange GPU, what's very funny?
>>53518918
But it will also be HBM1 and I'm playing at 1080p (smoothness over resolution everyday)
>>53518949
Who? Polaris 10? It will be GDDR5 and GDDR5X at best, HBM, even first gen is too expensive for that price bracket.
>>53518936
yay playing that new cis tomb raider in 4k and while checking your privileges
>>53518961
>implying I play kusoge
>>53518944
Can someone confirm. OP doesnt agree with you.
>>53518959
Polaris 10 is GDDR5, AMD confirmed it 2 days ago
>>53518970
He's wrong, Polaris 10 replaces 390/X
Polaris 11 replaces 370-360
Vega 10 replaces Fury X
Vega 11 replaces 380/285
>>53518944
Polaris 10 ~$200 replaces R9 390
Vega 11 ~$300 replaces R9 390x/Nano
Vega 10 ~$500 replaces R9 Fury/Fury X
>>53518972
they confirmed it was -not- HBM2
>>53518988
Guess ill have to go with Polaris 11 to replace my 7870. Wonder if ill gain any performance benefits...
>>53519002
>At this time we have only publicly demonstrated a GDDR5 configuration of the Polaris architecture.It’s important to understand that HBM isn’t (currently) suitable for all GPU segments due to the current HBM cost structure. In the mainstream GPU segment, GDDR5 remains an extremely cost-effective, efficient and viable memory technology.
Only thing demonstrated so far are Polaris 10, 2 days ago, and Polaris 11, over a month ago.
Both are GDDR5, and that makes sense since they don't need HBM.
And certainly won't have it considering their price bracket.
>>53518988
AMD said that one of the main goals of Polaris is to make VR more affordable.
The 360 or 370 replacement will never be fast enough for VR and the 390 replacement will be as expensive as the 390.
Only the 380 replacement could possibly have the right perfomance and price.
>>53518988
I DANT WANNA WAIT TILL VEGA 11!!!! FUYCK YOU AMD!!!!!!!!!!
>>53519040
Buy Nvidia then.
Or shell out $100 for a notably stronger GPU.
>>53519048
But i dont want nvidia, theyre scumbags! Also we know nothing about new nvidia cards right now. I just want 380 tier card!
>>53519060
You're not getting it because Tonga XT is a new GPU and they're not replacing it that soon.
Meanwhile everything else aside from Fiji are GCN 1.0-1.1.
Which is obsolete, featurewise and in dire need of replacement.
>>53519028
AMD's full statements refers to
>Higher-end market segments
>segments
The R9 390 is part of such segment. Saying that it won't have HBM without the smallest amount of doubt is pure speculation if you're only using January's and yesterday's statements
>>53519078
I don't know what more you need, two GPUs have been shown off so far and both have been GDDR5, from AMD's mouth.
You'll get HBM in highend with Vega XT and Vega PRO at the start of 2017.
HBM makes no sense for these cards since HBM is expensive and HBM2 is not ready.
If they're using GDDR5X, even better, there's no bandwidth bottleneck and you won't have a 4GB softcap
>>53519105
There doesn't have to be bandwidth bottleneck with regular GDDR5 either.
A 512bit bus and 7000MHz, heck, even 6500MHz memory would give more than enough bandwidth, AMD has been using much lower clocked memory than Nvidia but a wider bus.
>>53519060
Why not get Polaris 10?
>>53519284
Too poor, it'll probably cost $330
>was planning on buying GP200 or whatever AMD's Fury X replacement is this year in September
>have to wait for next year
Fuck me, I'll wait, I'LL FUCKING WAIT
AMD HAS NO GPUS & NO HBM2
AMD IS FINISHED & BANKRUPT
>>53518994
God, I hope so. Especially when I didn't pick up a 390 when they were going on sale for $230. It's a racket right now.
>https://www.youtube.com/watch?v=iilqtDkeIBE
Waveworks looks incredible to be honest.
>>53518521
Wait what? In capsaicin they state that they are mainly focusing on vr. In fact, they have saturated the vr market iirc
So AMD is finally throwing us poors a bone with polaris11? Their lowend gpus have been trash for such a long time now it made no sense to buy AMD.
Any thoughts on performance? Maybe ~ gtx960?
>>53521332
Polaris 11 will be around GTX 970 / R9 390 levels of performance
Polaris 10 will be around GTX 980 TI / R9 Fury X levels of performance
>>53521519
That's hard to believe. Am I mistaken in thinking polaris11 is being aimed at the lower end?
>>53521332
Are you a moron?
370>950
380>960
380x>970 (at 1440p) (3.5 isn't enough at 1440p)
390>970
390x>980 (980 is the worst price/perf card atm)
Fury<980ti
Furyx>980ti (At 4k)
>>53521019
It would be even more incredible if it didn't run like shit on other GPUs
>>53518430
>In short, if you want to replace your 980Ti's/Fury's you'll have to wait for next year.
That all depends on how fast Polaris 10 and w/e Pascal NVIDIA come out with are. For all we know Polaris 10 could be faster than a 980Ti/Fury X. It probably will be, I doubt they'd launch a new high-end GPU that's slower than old shit already on the market. It probably won't be an amazing leap, since they'll keep that for Vega though.
AMD demonstrated the Polaris 11 against the GTX 950 so expect them to be on the same price level.
Polaris 10 is supposed to make VR gaming affordable.
Vega 10 replaces Fiji.
No information about Vega 11 yet but it should be between Polaris 10 and Vega 10.
Guess I'm sticking with my 680 for another year.
>>53521698
It will be 10-20% faster at best, there's not much of a reason to upgrade unless you can overclock them by 30%+
>>53518583
Yeah, AMD are totally going to release high-end cards at $600 that get BTFO by a fucking 980Ti that's already on the market.
>>53521733
>680
>>53521735
OC headroom should be quite promising. The GPUs will be lower powered, which means there's a lot of headroom in terms of heat/power consumption to push the clocks and overvolt on typical aftermarket coolers. OCing should be decent at least unless the architecture is completely shit at it.
>>53521762
The trend in the industry is to slowly move away from overclocking, or at least make it as difficult as possible, I don't have much hope of the nextgen GPUs overclocking very well, especially as they're FinFETs
>>53521646
>370>950
Care to prove that?
>>53521790
I'm not sure what you mean exactly. AMD is pushing GCN balls to the wall ever since Hawaii, not much surprise that there isn't a lot of OC headroom there. The latest thing NVIDIA released, the 980 Ti, is a fucking miracle OC card, so I'm not too sure what you're saying.
If you're talking about CPUs, Intel's mainstream suffers heavily from their crappy TIM instead of being soldered, but GPUs don't even come with heatspreaders nowadays, so I doubt that would be a real issue. Intel isn't moving away from OC, as much as the market is suffering from a serious lack of competition. CPU clocks also seem to have gotten very close to some hard limitations, but you'll notice that newer Intel CPUs are generally capable of reaching the same clocks as older ones, so it's not like the smaller nodes are incapable of high frequency, once you remove the thermal bottleneck due to the shitty TIM.
>>53521627
No but considering it's a two node shrink it could be that substantial
![perfdollar_1920_1080[1].png](https://i.imgur.com/uXjN9L6.png "perfdollar_1920_1080[1].png")
>>53521829
Are you retarded son?
>>53522713
Sorry i was too rude. Didn't want to.
>>53521829
Does anybody knows where can I watch some of the GDC talks? I'm relaly interested in some of the talks from today but I can't find anything.
>>53521627
I get the impression that there will be cut down versions of the chip to market lower end
>>53518988
This guy gets it.
>>53519878
wha
are you angry about something
>TFW I have a Fury X
Feels good
>>53523142
>4gb
cuck
Same chip bit smaller
>Suicide Watch
>>53523233
Not HMB mem
>cuck
>>53518583
>>53518430
AMD has literally been rebranding the same architecture for 3,5 years now. They are not going to survive this postponing. The nvidia monopoly that will follow will be bad for us all.
>We have a Polaris 10 demo today!
>plays for 5 seconds
>dies
>Pajeet: WOW AMAZING THANK U FOR DAT
>>53524940
>marketing lackey
>good at vidya
It's a good showing because that buggy as fuck games needs a 1080Ti to even reach 60FPS.
>>53518583
>doesn't understand that smaller chips means more chips on a single wafer so they are actualy cheaper to produce
>>53524958
After the stream I finally know why amd is in the position they are in.
>>53518430
>replacement of 390 will basically be a 390 on a smaller node with all the performance and power advancements it will bring, might even be coupled with HBM1
If it's around 300-400 bucks it sounds like an awesome card
>>53525006
It's GDDR5 or GDDR5X, AMD confirmed it isn't HBM for Polaris 10.
>>53525045
>AMD confirmed it isn't HBM for Polaris 10.
Source?
>>53525081
That's a post, not a source
So they're remanufacturing current cards on a smaller process?
>>53525123
>So they're remanufacturing current cards on a smaller process?
It doesn't work like, idiot.
