NVIDIA Volta with GDDR6 in early 2018

Wait GDDR6 is a thing?

I thought GDDR5X was made because there was going to be a big gap before 6 so they pushed a high spec 5 instead.

>>60045921

The flagship will get HBM2, not sure on 4096 bit

>>60045921
"Early"

File: amdvega.jpg (156KB, 633x758px) Image search: [Google]

156KB, 633x758px

>>60045921
NOOOOOOOOO

>>60045921
And another paper launch? God I love novidia.

>>60045959
that's rich coming from amdrone still waiting for vega

>>60045972
But I have novideo card.

>>60045959
Paper launch?

>>60045921
bbut HBM2 was meant to GOAT..????! it was developed by highly skilled engineers at AMD, wtf Nvidia???!

>>60045986
Pascal was launched eons ago with GP100.

like it matters anyway Nvidia is just going to dump marginaly better tech because why not milk the monopoly like big pharma or google or amazon does, shits like the style now.

>>60045995
It is GOAT. It's just costly.

so AMD is gonna be TWO fucking generations behind nvidia
what a time to be alive

>>60045997
vega was launched december 2016 for enterprise

>>60046015
we have no idea what vega is yet

>>60046016
And MI25 is already here. Much much faster than GP100.

>>60046021
It's Maxwell from AMD.

>>60046024
Citation

>>60046021
>wait for vega while nvidia is pushing out gddr6 and double clock speeds
yeah dude, I am totally waiting for vega

>>60046056
It's /g/ so fucking google it.

>>60046056
>at compute

AMD always was faster at compute.

File: 3F837CF700000578-4437298-image-a-51_1492951348448.jpg (64KB, 634x423px) Image search: [Google]

64KB, 634x423px

>>60046024
Bullshit

kill yourself AMDead

>>60046066
>vega 2017 HBM 2
>volta 2018 gddr6

yup, just wait for volta

>>60046068
>>60046072
I need a citation and not your AMD cringe leak please

>vega will be slower than 1070/1080/1080Ti
>Nvidia will then release Volta which will be faster again
>AMD will fall further and further behind
Step up your game AMD, you're stagnating GPU tech by not being competitive.

>>60046087
>I can't count
Okay, Rajeesh.

>>60046098
B-b-but 2xvega pro duo :^)

>>60046098
Ramesh pls.

>>60046098
>will be
>will be
>will be

hey, I can do that too

>>60046024
I'm the england queen

>>60046111
I need the source of what you said and fuck off with out Tflop meme from amd

>256bit GDDR6 card will have same amount of memory bandwidth as HBM2 2 stack AYYMD HOUSEFIRES with significantly lower cost
>384bit GDDR6 card will have more bandwidth than AYYMD HOUSEFIRES

AYYMD IS FINISHED & BANKRUPT

AYYMDPOORFAGS CONFIRMED ON SUICIDE WATCH

File: poo.png (312KB, 595x654px) Image search: [Google]

312KB, 595x654px

>>60046111
>Rajeesh
stop projecting poojeet

>>60046024
>And MI25 is already here. Much much faster than GP100
>t. amdrone

Stop

>>60046149
>pure compute power is a meme
Are you braindead? I mean you're an Indian, but how can one be so retarded?

>>60046131
>The Quadro GP100 is also the first instance of many technologies coming close to the consumer front in the form of HBM2, NVLINK and over 20 TFLOPs

>Radeon Instinct MI25 - With a Vega GPU which is listed at 25 TFLOPS

enterprise specs never lie.

>>60046156
>>60046159
All the Pajeets.

>>60046176
GP100 is 9.7tflops FP32 afaik.

>>60046176
Where are the test? Where is the (((MI125)))? These numbers from AMD mean NOTHING.

>>60046184
>MI25 Announced – Features 12.5 TFLOPs of FP32 Compute Performance.

>>60046200
those numbers from nvidia mean nothing.
I can do that too.

>>60046200
>enterprise specs mean nothing
End yourself, Rajeet.

>>60045921
4k introduced?

>>60046184
>Lying about things you don't know

>>60046246
>specs mean nothing
Okay, Ganesh.

>>60046246
>The GP100 GPU's based on Pascal architecture has a performance of 10.6 Tflops of FP32 performance and 21.2 TFLops of FP16 performance.

he wasn't that far off

>>60046211
>>60046215

We have tests and benchmarks of Quadros and Teslas with the GPxxx but nothing from AMD. What is the price of the '''MI25''''?

>>60046280
>compute numbers mean nothing for compute cards
Okay.

>>60046280
you are retarded, enterprise specs never lie because if they do company gets fucked for false advertisement by corporation lawsuits
enterprise market is not forgiving, you can't ride your customers on 3.5Gb

You guys are missing the important thing here

Will it be called the 11xx series or something retarded like 20xx?

>>60046409
No one cares about the name.

>>60045921
Bounty bars are delicious and I miss them like heck in Japan.

Snickers are fairly boring but about the only candy bar they have. A few places have Kitkat Chunkys, but they're also nothing next to a bounty bar.

Which reminds me, there was supposed to be some recipe to making your own bounty bars. I should look that back up.

>>60045921
>Volta
>Pascal with GDDR6

mark my words.

>>60045932
http://vr-zone.com/articles/gddr6-memory-coming-in-2014/16359.html

>>60046432
I fucking care about the name if they make it the 20xx ill have to change my computers name to the TARDISt200 and that is a HUGE version jump

Is HBM just too difficult to scale up in comparison to GDDR?

Should it just be limited to constrained environments like laptops, tablets, etc where there is no upgrade path so there is an actual benefit in saving space over everything else?

>>60046481
No, it's just slapping more GDDR is cheaper.

>>60046150
16gbs gddr6 vs 512gbs+ hmb2 stack
Ye nah

>>60046493
Slapping more GDDR means more traces though.

>>60046526
Aaaaand that's precisely why NVIDIA gives zero fucks. Its all up to vendors. And slapping more traces is still easier than working with interposers.

>>60046526
GPUs are already 6/8 layered PCB with or without HBM

>>60046539
You still need space to get the traces from the GPU pins to the other layers and then bring them back. It would be an absolute pain in the ass to try and get like 16 chips connected to a single GPU, which is why they limit to around 8.
Most likely this Volta card will have 8 2GB, nothing really new in regard to the number of chips or the traces required.

File: AMD_Hawaii_GPU_RAM.jpg (327KB, 1000x583px) Image search: [Google]

327KB, 1000x583px

>>60046608
Add more layers, it really isn't as hard as you think, PCB cost only goes up with number of layers not the traces.

>>60046083
Navi is hbm3

>needs 2 fucking phases just to modulate the GDDR*
>uses 3 times the power of HBM*
>prevents SFF ala the Nano
>higher latency


No thanks, HBM2 on $600+ GPUs, keep your GDDR on mid range cards

>>60045921
> wait™
Yeah nah
I'll just pick whatever next 120W card offers significant features improvement over my current card. 150W is pushing it in my tiny brick with bad airflow, while 120W can work fine with case fans on minimum.
GTX960 raw performance is fine enough for 1920x1200 anyway.

>>60046730
Are you stupid or what?

https://www.zotac.com/my/product/graphics_card/zotac-geforce-gtx-1080-mini

https://www.msi.com/Graphics-card/GeForce-GTX-1070-AERO-ITX-8G-OC.html

>>60046801
>GP104
>still bigger than the Nano

>>60046817
Only HBM allows to make truly smOll high-end cards.

>>60046837
That's the point?

GDDR, even 6 which is still 9 months away is inferior to HBM2 in every single aspect besides price, and I don't really give a fuck about price when I'm throwing over half a grand on a GPU

>>60046874
And it will be inferior to HBM3. GDDR is cheaper though.

>>60046911
That's what I said.

>>60046874
Capacity maybe? HBM2 tops out at 16GB at the moment, though that seems enough even for a compute monster like P100.
Also 8GB stacks are coming out sooner or later, 32GB will be enough until HBM3

File: emib-anticipate-norm-yield-blue-16x9-graphic.png.rendition.intel.web.416.234.png (23KB, 416x234px) Image search: [Google]

23KB, 416x234px

Intel seems to have found the answer to the high interposer and TSV cost in EMIB.
Which is rumored to star in Kabylake-G late in the year with HBM2 and a Radeon GPU

It's like lego honestly, these APUs aren't using custom buses for GPU<⨪> CPU communication.

>>60045972
I thought Vega cards existed, but they were being deployed as workstation or grid computing cards?

>>60047062
>Radeon GPU
kek now intel igpu will have abysmal drivers too

>>60045921
But will it support async compute?

>>60047062
http://hothardware.com/news/intel-kaby-lake-g-series-leaks-with-heterogenous-multi-die-design

>>60047072
http://www.eteknix.com/amd-unveils-vega-based-radeon-instinct-mi25-gpu/

>>60047074
Can't be any worse than the inability to run Diablo 2 or Warcraft 3 without freezes every 5 seconds on a Intel HD4000

>>60047077
>async compute
here comes the retard regurgitating amd marketing buzzwords

>>60047107
>async compute does not exist REEEEEEEEEE

>>60047062
>Intel seems to have found the answer
yup, using AMD patent is the answer.

>>60047062
>and a Radeon GPU
No.
Intel have supposedly licensed AMD iGPU related IP after the very same deal they held with nVidia expired a few months ago.

>>60047119
It's just a hack to mitigate amd's abysmally bad gpu architecture.

File: 1479618237612.png (325KB, 522x593px) Image search: [Google]

325KB, 522x593px

>>60045932
Yes, and so is HBM3 also coming next year.
Up to 64GB per stack and multiple-terabytes-per-second of bandwidth.

>>60047148
GCN is godlike though.

>>60047167
yeah sure
that's why they need (((async))) because amd cheaped out on ROPs and half their shaders stay idle

>>60047203
That's an arch design. GCN leverages compute. And it aged better than anything novideo.

>>60047203
>cheaped out on ROPs
aren't you confused there? nvidia the one that cheaped out on ROps.

>>60047235
>what is fury x
>>60047218
>aged better
you men finally get thir heads out of their asses and managed to delive somehow acceptable drives which still have terrible OGL support

File: RX580-guru-3d-1.png (732KB, 1364x6707px) Image search: [Google]

732KB, 1364x6707px

>>60047203
>A $550 290X
>say as the 390X in the attached image
>from 2013
>still +/-5% of the $650 980Ti at 1440p from 2 years ago
>regardless of async compute being used or not.
>bad architecture reeeeee!!!!!

Now I bet you'll go post day 1 980Ti release benchmarks of the old GCN drivers, faggot.
GCN is just a plain superior architecture to anything Nvidia has released the past 6 years.

>>60047275
>amdpoojeets are so retarded that they fail to quote the right person
colour me surprised

>>60047275
>Hawaii STILL kicking ass
>keklers in the trashcan
Motherufucking hilarious.

>>60047329
I quoted the right person, dumb fuck. Delete your embarrassment.

>>60047275
Well you need your own personal nuclear reactor to give enought juice to the 9000000000000000W
r9 furry xxxxyxxxyxx.

1070 at least does not need custom loop with nitrogen and 2000W ps

File: 1491034527753.jpg (46KB, 622x647px) Image search: [Google]

46KB, 622x647px

>>60047335
>not tested: Fermi, because it'd be far too embarassing. But go ahead and keep testing the 2011 280X/7970.

>>60047368
He's talking about Hawaii, not Fiji you dumb Pajeet.

>>60047368
>waaahhhh the best cards ever from years ago are less power efficient that newer cards!!!!
Keep buying a new card every year, cuck.

>>60047275
You're fucking retarded. Just bite the bullet.

>>60047358
literal imbecile right here

>>60047387
Actually that could be cheaper in some places than to keep energy vampire.

>>60047373
Well I mean, they can't test any GTX680's anymore. nVidia fucking killed them all with driver updates.

>>60047423
That's kekler. 7970 competed with late thermis.

>>60047423
They could still use the last supported driver.
And yes, 7970 came out soon after Fermi did. It's understandable why they stopped driver updates.

>>60047107
It probably still doesn't have a hardware scheduler. :^)

>>60046313
>>60046291
How much cost the MI25?

>>60045921
>early 2018
So exactly one year or more.
Why would i care with Vega comming in like 1-2 months?

>>60048621
Because vega is problably slower than pascal

>>60048648
Vega will probably be somewhere between 1080 and 1080Ti for less money, i see no problem.
>a card comming out in over a year from now will be faster
Well no shit.

>>60048648
Based on specs alone it's just as fast as the 1080, and this is without taking into account that Vega will be the biggest architecture change since GCN. So yea, we'll see who's slower.

>>60048724
>specs alone
Specs alone put it above the Titan Xp. Where do you guys get these numbers from, your ass?

>>60048742
>>60048664
https://www.youtube.com/watch?v=Ku5OMWYVKSs

From a video released by AMD itself.

It's showing performance below what a 1080 can do (1080 runs similar framerates at 4k ultra, not 4k high).

Granted that is without optimised drivers, but considering AMD thought it prudent to show this I don't think it will be very far from it's full potential.

I do believe the flagship Vega chip will be faster than the 1080, for AMD's sake I certainly hope so.

It runs with HBm2 and the die size is about 80% larger than that of the 1080, meaning it's alot more expensive to manufacture. so it better beat it by a significant margin

>>60048768
I meant nighmare over ultra

>>60048768
Fuck, you're so retarded.

>>60048784
What a fantastic retort

>>60046469
>ill have to change my computers name
kill yourself

>>60046654
>mfw wait for navi
the jump from my 980ti is going to make my dick explode

>>60048833
Why don't you wait 10 years longer?

>>60048768
Didn't know the vid but conaidering it's from early January i'd take it with a grain of salt.

>>60048793
Waste of time.
You think a GPU that's 80% larger, has 50% higher TFLOPs, absurdly lower memory latency, greater memory bandwidth, and all those similar optimizations that Maxwell and Pascal had over Hawaii/Fiji like tiled rasterization and primitive discarding, not to mention new shit, is only going to be "slightly faster" than the 1080.

Easier to just point out that you're retarded and wasting valuable post space on /g/ with your technical illiteracy.

I'm waiting for volta so that vega will be even cheaper by then...

>>60048866
Did you watch the video?

You know, the actual proof of the performance of the card as demonstrated by AMD?

Also the fact that you are comparing TFLOPS across different brands of cards and even worse different architectures just demonstrates your complete lack of understanding of GPU technology.

>>60048884
The Vega card was underclocked and yet still had 30% higher minimum FPS, you tard. And that was on Fiji drivers.

>Also the fact that you are comparing TFLOPS across different brands of cards and even worse different architectures just demonstrates your complete lack of understanding of GPU technology.
The fact that you think Pascal is 50% ahead there shows your lack of understanding. RX480 does not match the 1060 by having 50% higher TFLOPS. It's about 26% higher.

>>60048742
shader count, clockspeed, that sort of stuff

>>60045921
>First things first, how do we know it’s Volta? Obviously, we don’t know, but since Vega is not here yet, chances for Vega successor in early 2018 are close to zero.

ITT retards endlessly slapping their dicks around with hateboners at the thought of AMD getting crushed despite them being a tiny company in comparison to both of their rivals.

Seriously the amount of retards that will immediately start jerking off if it looks like AMD is about to get shit on is fucking pathetic. You should all just kill yourselves.

File: AMD-Vega-GPU_2.jpg (46KB, 630x354px) Image search: [Google]

46KB, 630x354px

>384bit GDDR6 768GB/s is more than HBM2 512GB/s max 2 stacks
>256bit GDDR6 512GB/s matches HBM2 512GB/s max 2 stacks while being much more cheaper and easier to implement

AYYMD IS FINISHED & BANKRUPT

AYYMDPOORFAGS CONFIRMED ON SUICIDE WATCH

>>60045921
so HBM ded?

>>60047107
Kill yourself.

>>60048935
>HBM3 at 1-2TB/s is more than 384bit GDDR6 768GB/s
NVIDIA IS FINISHED AND BANKRUPT

Nvidiot doesn't even know that 1TB is 1024 GB.

File: 9bb6b7df-6be9-45fe-a2d1-6e4580a309f1.jpg (40KB, 700x402px) Image search: [Google]

40KB, 700x402px

>>60048956
Look at this dumb retarded AYYMD faggot, HBM3 won't even be available until 2019 earliest, more likely 2020

JUST WAIT(TM)

>>60048849
>muh theres no point in waiting!!!!!!!!!!111!!1!!
post

why don't you kill yourself? It is not hard to create a metric for when it's feasible to upgrade your GPU; for me I need at least a 100% increase in in-game performance overall at a price of under $400. I expect GPUs in mid-late 2018 to deliver ths.

>>60048976
You do realize HBM2 can do 1TB/s, right? Of course not, because you're an idiot.

Your JUST WAIT Nvidia GDDR6 coming next year is still slower than HBM2. The speeds for that chart is for single stack speeds.
And GDDR6 isn't Nvidia proprietary technology. Samsung is making it. The RX600 series next year will likely use it as well.

File: doitfgt.jpg (85KB, 843x722px) Image search: [Google]

85KB, 843x722px

>>60045921
>NVIDIA Volta with GDDR6 in early 2018
>First things first, how do we know it’s Volta? Obviously, we don’t know

Everyone who even replied to this with "AMD IS DOOMED LOL" needs to slip into a noose.

>>60049008
You're beyond dumb, you can see Vega shit only has 2 HBM2 stacks so it will NEVER ever have more than 512GB/s memory bandwidth

JUST WAIT(TM)

>>60049023
Yes, likely 512GB/s on Vega, compared to 484GB/s on the 1080Ti.
JUST WAIT until Q2 2018 at the earliest for Volta only to get BTFO by Navi.

>>60048935
>much more cheaper and easier to implement
You do realize this wouldn't affect the price with Nvidia at all?

>>60049046
Navi doesn't even have HBM3 because HBM3 won't even start production until 2019/2020, so it has no memory

Dead on arrival, TOP KEK
AYYMDPOORFAGS really have no brains

>>60047107
AMD_(((OPEN)))_ASYNC_COMPUTE(void *poo, void *poo)

//Cannot be implemented in Intel, Nvidia, Qualcomm, Arm, imagination Tec.

>>60046464
>Volta
>Pascal dieshrink with GDDR6

fixed.

>>60045921
AMD and INTEL doesn't support PCIEx 4.
#POWER9_FOR_DESKTOP.

>>60049357
Nvidia has been working on Volta for like 5 years.

It better not just be a dieshrink with GDDR6.
Their slides claimed it'd be their biggest performance jump from one architecture to the next ever.

>>60049440
>Their slides claimed it'd be their biggest performance jump from one architecture to the next ever.
really?

>>60046200
MI25 is a pure compute card, it's already on the market, it has no display output port.

>>60049177
All of those already support it, just not on GPUs

Well Intel does if you consider Xeon Phi a GPU, which it originally was going to be.

>>60045921
>tfw vega will probably, at best, match 1080 speeds

being an AMDrone is suffering lads, kill me

>>60049527
>Implying this is any different with AMD marketing slides

Remember how they teased Polaris with 'Poor Volta' and it ended up being a budget mid-range card.

lmfao

>>60049569
I think at best it will match the 1080ti, but I expect it to in between 1080 and 1080ti level.

Considering the fact that Volta will hit soon after that I doubt it's enough

>>60049575
>Remember how they teased Polaris with 'Poor Volta'
no

I remember how they advertised its power efficiency and at least delivered on that front. Too bad it's still weaker than the fucking 290x.

>>60049587
>Volta will hit soon after
Why is every retard in this thread repeating this rumor like it's fact

>>60049595
don't you know?
>a year later is SOON

>>60049598
Are you talking about Vega?

Which compets with high end Pascal but is nearly a year later?

File: power_average.png (33KB, 500x690px) Image search: [Google]

33KB, 500x690px

>>60049592
>AYYMD HOUSEFIRES
>power efficiency

TOP KEK

>>60049592
nvm, they teased Vega with poor volta.

Still, I think that won't happen, people are doubting if they will be able to compete with high end Pascal, let alone volta

>>60049604
I'm referring to this
>Summer 2016 - Pascal
>Summer 2017 - Vega
>Summer 2018 - Volta

File: Girls[1].png (490KB, 449x401px) Image search: [Google]

490KB, 449x401px

>he doesn't wait for GDDR6 and bought a 1080ti

>>60049616
If Volta is really just a dieshrink + GDDR6, Volta is only going to be merely catching up with Vega a year later rather than blowing it the fuck out. It'll be like GCN and Keplar all over again.

>>60049628
Pascal is 'just' a die shrunk Maxwell.

Yet it's mid range card (1080) is 25% faster than the Titan X of the Maxwell series.

Your post literally makes zero sense.

>>60049357
>Pascal
>Maxwell dieshrink

fixed

>>60045932

No it was just more effecient to milk you that way.

Nvidia has been using hbm2 from amd in their high end enterprise shit like the 6k cards for productivity.

Its all about fucking you anon. The jig seems to be up though graphics cards have gotten to the point that you can get the mid tier from last cycle and be missing nothing.

File: AMD-Vega-architecture-1000x505.png (218KB, 1000x505px) Image search: [Google]

218KB, 1000x505px

>>60049628
>Volta is only going to be merely catching up with Vega a year later

cut this marketing bullshit out,there is Hope vega will dominate but stop hyping so much without any benchmark.

File: vlrPf7s.jpg (23KB, 450x646px) Image search: [Google]

23KB, 450x646px

>>60049622
>he didn't wait for Titan X(pascal) and bought a 1080
>he didn't wait for 1080ti and bought a Titan X(pascal)
>he didn't wait for Titan Xp and bought a 1080ti
>he didn't wait for pascal refresh and bought a Titan Xp

>tfw polaris successor will have gddr6

>>60047142
EMIB is their own invention.

>>60047062
lmao Intel is licensing that from AMD, you fucking retard.

>>60050390
http://www.intel.com/content/www/us/en/foundry/emib.html

>>60050317
>>60050434
I sincerely hope EMIB is not another R&D failure from Intel.

>>60050493
It's a good technology if they're already using it so soon, Kabylake-G is late this year

>>60050504
>so soon
Looks more like a desperation move.

>>60049637
The performance gains can be attributed to the much better process, 28nm to 14nm is a three generations jump

>>60049637
A 1080 is not a mid range card, it costs $600 for God's sake.

>>60048768
UUHHH

>>60047074
Intel has always had abysmal drivers in every single product they have

>>60050548
No, they actually had patents on this for a while.
Even Intel knows its unfeasible to have huge monolithic dies.

Shit talk Volta, how can amd save face if they can even beat pascal lineup

>>60047445
>tfw Thinkpad with thermi runs like shit on anything that uses the GPU thanks to shit tier novideo+jewtel drivers
It's bearable on GNU/Linux using only the i915 drivers and disabling the dGPU though

>>60050586
Yet they still throw moar ringbuses. Intel is weird.

https://lists.freedesktop.org/archives/mesa-dev/2017-April/152733.html

Full OGL 4.5 support for Vega, it's close.

>>60050582
That's wrong, though. Intel has great drivers. Even their open source Linux driver is better than the proprietary AMD and Nvidia drivers.

File: Alot.png (113KB, 600x450px) Image search: [Google]

113KB, 600x450px

>>60048610
It's a enterprise card.

>>60049606
LOL!!!!!

RX 580 consumes more energy than the GTX 1080ti.

>>60050726
It's not, Intel on Windows is a total shitshow, they literally can't run DX9 stuff, and DX10 and newer stuff is way too intensive for most of their iGP's
God forsake you if your stuff is older, since they usually kill support within 2 years
The meme of the Linux drivers it's just a meme perpetuated by people who do basic stuff on their machines and never run into problems, but their drivers are shit specially on OGL, just ask the mpv devs

>>60050737
And?

>>60045921
im not waiting until 2018, will make my decision after e3 and vega

>>60050695
>can finally run wolfenstein stable at 4k
lads...

File: Radeon-Pro-Duo-Polaris.jpg (159KB, 1600x1068px) Image search: [Google]

159KB, 1600x1068px

>The Radeon Pro Duo is equipped with 32GB of ultra-fast GDDR5 memory to handle larger data sets, more intricate 3D models, higher resolution videos, and complex assemblies with ease. Operating at a max power of 250W, the Radeon Pro Duo harnesses a total of 72 compute units (4608 stream processors) for a combined performance of up to 11.45 TFLOPS of single-precision compute performance on one board, and twice the geometry throughput of the Radeon™ Pro WX 7100.2 The Radeon Pro Duo enables professionals to work up to four 4K monitors at 60Hz, drive the latest 8K single monitor display at 30Hz using a single cable, or drive an 8K display at 60Hz using a dual cable solution.

>https://videocardz.com/press-release/amd-announces-new-radeon-pro-duo-with-two-polaris-10-gpus

>>60051298
This shit is slower than 2016 pro duo fury

>>60051516
But has more memory and uses half the power.

I'm sick of all this rumors when can I expect Vega

>>60051516
>>60051298
>>60051584
also slower than 295x2

>>60051990
Q2

File: 1491502354030.jpg (73KB, 633x758px) Image search: [Google]

73KB, 633x758px

NEW PLAN WAIT FOR VEGA 2

Testing the waters

>>60045949
Lol not even

>>60052414
that's mainly for workstation like quadro

>>60053795
It would be good to specify it's totally capable of still pulling that same 64 wide load.

>>60053795
Trucklet!

>>60054605
This

>>60045972
#rekt

>>60054529
Obviously, wouldn't be much of a parallel SIMD processor if it couldn't do that.

>>60050823
Wut you going to do with it?

>>60045921
What ever happened to HBM?

Like a year or three ago people were touting HBM as the GDDR5 killer. Memory bandwidth was supposed to be something ridiculous like 4TB/s. I pretty much have never seen it talked about since though granted I haven't really been following the GPU market as of late. Last I remember the bandwidth was roughly double GDDR5 for the same bus size which is good but fell way below the claims. I think 600GB/s was typical for the higher tier flagship cards like the R9 380X and 390X.

>>60050608
this. i just finished watching an rx 580 review and it loses to their aftermarket 1060 in most tests and matches it at best. this is a joke.

>>60055093
HBM2 is 1TB/s and even that's enough for the $7000 P100

Too much bandwidth is just a waste of power if the shaders can't use it

>>60055175
Wow, I have been out of the loop for awhile. Didn't even know they came out with HBM2.

File: rx580-oc3d.jpg (2MB, 752x6264px) Image search: [Google]

2MB, 752x6264px

>>60055132
>1080p babby talking
Get a real resolution, cuck. I have never once sullied my eyes on 1080p. I went from 1600x1200 to 1920x1200.

Nvidia makes cards for plebeian resolutions for you to enjoy, good for you. I require better.

>>60055292
is this bait?

>>60054943
The point is this card is not real/paper launch. Nobody use it because we know nothing about it...

>>60045937
Raja holding a Vega chip
http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser/4

File: 53583_03_hbm3-released-2020-offers-more-bandwidth-less-power_full.jpg (48KB, 700x402px) Image search: [Google]

48KB, 700x402px

>>60047151
>HBM3 also coming next year.

I have some bad news for you, famalam

>>60051298
>>60052973
http://i.4cdn.org/g/1493065315614.png
>tl;dr Vega showed pulling 8k earlier today at NAB show. Still no comment on release.

Source is PCworld

>>60055871
Before HBM3 drops AMD will launch a VEGA20 GPU with 4 HBM2 Stacks which would result in 1024GB/s

File: 50916_02_sk-hynix-hits-mass-production-4gb-hbm2-q3-8gb-q4-2016.png (149KB, 620x464px) Image search: [Google]

149KB, 620x464px

>>60056244
Capacity is more important than bandwidth

>>60048768
>early fucking January
You do realize the WORLD of difference 4-5 months can make right?

File: off yourself.png (172KB, 353x311px) Image search: [Google]

172KB, 353x311px

>>60049062
>Navi doesn't even have HBM3
That's not the point I was making. The point is that Navi will probably have more stacks over Vega to reach 1TB bandwidth.

Shills always try putting words in other peoples mouths don't they?

>>60045921
>Novidya rushing out Volta to try and compete with Vega
>It will be unfinished and held together by woodscrews again
wew

>>60056324
That's the exact opposite approach AMD's taking with Vega.

They're arguing that if you have robust enough background memory transfers, moderate size "high bandwidth caches" can be used transparently for terabyte of assets.

>>60056588
Where is vega?

>>60056735
Q2

>>60056709
Does it matter? How often has a game even used more than 5gb of VRAM, lot alone close to 8. Besides, if 8gb is going to be a bottleneck soon, then I'd hate to have purchased a 1080 or below, which brings me to my next point. So many people have less than 8gb, that you probably won't see games require 8gb for a long time.

>>60056795
So nvidia is ahead

>>60056836
Yeah, and Intel was "ahead" for all of 3 months until Ryzen hit the scene.

Enjoy it while it lasts.

>>60056836
Ahead in what? Volta is late Q1 2018 at best.
That's a far cry from Q2 2017.
Have I to remind you that 1080ti and Titan Xp launched at the end of Q1 2017? Basically a little more than a month ago?

>>60056735
this quarter so june 31 at worst

>>60056925
Any sensible company would launch at computex, even if a paper launch just to have huge coverage.

>>60056881
980ti oc, 1070, 1080, Titan X Pascal, 1080ti, Titan Xp, Quadro GP100, Quadro P6000

>>60056984
If we're going like that then half that line is obsoleted by the Mi25, released in December.

>>60046177
XDI CALLED HIM PAKJEET LOLOL

I'm thinking of splurging on a 4k screen and a 1080Ti card this year, but I'm not on a hurry. If i wait until the summer and Vega, is there a chance of finally having a fucking price war worthy of the name? Or i might as well go out shopping now?

File: 1491417983950.jpg (253KB, 876x1000px) Image search: [Google]

253KB, 876x1000px

>>60047275
Hawaii is the gift that keeps on giving.
That RX580 tho. Can it clock to 1500MHz? That would put it very close to the Fury maybe even beat it.

>>60057034
if you want it now, go for it, otherwise wait.

I would really recommend waiting until they produce a 120+hz 4k monitor with HDR10 with either freesync or gsync respectively

>>60049868
nice meme idiot

hbm/hbm2 aren't cost-effective for consumer graphics cards. it's why nvidia has been raping amd's shit because amd fell for the hbm meme. as much as nvidia is "fucking" you, it's a free market and amd is free to compete but in the end nvidia has the better offer for the consumer.

http://wccftech.com/amd-confirms-rx-vega-launching-quarter-showcases-running-8k-video-editing/

>>60057167
So use HBM for $600+ SKUs and use GDDR for poorfag shit, pretty simple.

>>60057150
Does xSync + 120hz still make a huge difference if you can't go past 60-70 FPS? I'm not gonna do SLI so that would be my upper limit most of the time I think.

Also thanks for the HDR10 tip, I'm very out of the loop and didn't know about it.

We must be just over a decade away for single-chip petaflop perf, right?

>>60057230
Are you for fucking real? We're barely at 13TFLOP now.
Even a 50% increase per year, which is not gonna fucking happen with how rare shrinks are, will get nowhere near a PFLOP in a decade.

>>60048980
yea, but just imagine in 2050.............. just wait......

>>60057271
13*1.5^10 = 750 TFLOPS

>>60045921
I'mma eat the fuck outta that Snickers.

>>60048849
>>60057328
fag, if he already has a 980 ti he's probably not very desperate to upgrade

>>60046015
Probably one generation behind.

>>60046021
AMD wouldn't have launched a series of rebranded RX 480s now if VEGA was close to being production-ready.

>>60046066
>yeah dude, I am totally waiting for vega
This is pointless. There will always be new technology around the corner.

>>60046072
>AMD always was faster at compute.
No. Both GTX 1060 and GTX 1060 can compute. This wasn't always the case but it is the case now.

R9 290, for example, can compute zcash at 285 sol/s. GTK 1060 manages 305 sol/s and this is comparable to the RX 480. GTX 960 can barely do 150.

>>60057271
Actually, we'd only need nine years with a 50% increase every year starting from 13TFLOPS in order to reach 1PFLOP

>>60057378
Which is still a good 40% off a TFLOP at a best case improvement (50%)

>>60057378
Your math is autistic.

>>60057419
Ignore that dumb faggot.
>http://www.math.com/students/calculators/source/compound.htm
Fields:
10
50
13
0
We'd have consumer cards with nearly 1.75PFLOPS in 10 years

>>60057419
which is probably why anon said just over a decade, not exactly a decade

>>60057427
your retarded

>>60057462
>compounds it monthly
the premise is a 50% increase per year not per month fucking underage burger kid

>>60045921
HOW MUCH ETHEREUM PER MONTH WILL YOU BE ABLE TO MINE WITH THIS THING?

File: 956b780df5e42b595bd52db28b484be2_image-meme-does-not-compute_450-334.jpg (81KB, 450x334px) Image search: [Google]

81KB, 450x334px

>>60046150
>AYYMD IS FINISHED & BANKRUPT
You're not that far off there, their profit margin was -5% the last quarter.

>>60046239
>4k introduced?
I would personally like something that could handle 3x1440p. I doubt this card will be great for 4k or 3x1440p. I would love to be proven wrong.

>>60046291
This does not compute.

>>60057124
Could've just gotten a fury then because the card is going to pull over 200w.

>>60051298
This shit can't even mine cryptocurrency properly.

>>60057483
>>60057465
Are you two just mentally incapable of doing math or are you that fucking thick-skulled?

>>60057465
>your retarded
>your
>I WAS ONLY PRETENDING
I guess math isn't the only thing YOU'RE incapable of performing.

>>60057623
>he has to resort to using an online babby calculator
>which does monthly compound interest when we're talking about yearly
>implying it's not obviously 13*1.5^10 = 750 TFLOPS
kys

>>60057654
>he still can't show his work
Fucking brainlet

>>60057673
that's literally the work you fucking retard

i got the max grades in all of math in a country which wasn't grade-inflated when i went to school

>>60057406
>Both GTX 1060 and GTX 1060 can compute
impressive!

https://www.youtube.com/watch?v=Wllc5gSc-N8

>>60057703
>he couldn't figure it out
Its okay autismo, you can always come back and try again later

>>60051298
amd's progress has been at a standstill since pooduri took over radeon

>>60056324
AMDs GPU architecture profits more from higher bandwidth than higher capacity

>>60045921
>waiting for volta because pascal is trash

File: 1462152951927.jpg (43KB, 600x600px) Image search: [Google]

43KB, 600x600px

>>60057985
>waiting for volta because pascal is trash

>>60058038
Pascal is simply overclocked Maxwell.

>>60058050
it's also much more power efficient so it's not just an OC

File: Superposition_Benchmark_v1.0_2611_1492779061.png (603KB, 1024x640px) Image search: [Google]

603KB, 1024x640px

>>60047074
>>60047096
>>60050582
>>60050726
What is this meme?

>>60058360
That's what happens with a dieshrink, clocks go up, power goes down.

>>60047368
I have an r9 fury on a 650w seasonic. Works fine. Stop being a fanboy faggot.

>>60058050
>Pascal is simply overclocked Maxwell

Citation needed.

>>60058050
>980ti
>2816 cores @384bit
>1070
>1920 cores @256bit
>1070 is faster
>Pascal is simply overclocked Maxwell

>>60059998
Don't argue with clueless amdrones.

>>60057753
Hoooooly shit you're a retard.

Is Pascal to Volta going to be as big of a jump as Maxwell to Pascal was? Because an 1170 that performs as well as a 1080ti would get me to upgrade. Hell, an 1160 thats better than a 1070 would be tempting.

>>60058413
I have been using a GTX 480 with a Phenom II on a 550W PSU.
No, my house didn't burn down.
What manufactuers put in specs as required PSU wattage is always hugely overblown. As long as you don't use a chinkshit PSU that doesn't live up to its specs table and your CPU isn't a fucking FX 9590 you can usually go ~100W below what NVIDIA/AMD puts as recommended wattage, you just may not be able to overclock without triggering overcurrent protection during spikes of intensive load on both CPU and GPU at the same time.

>>60059998
>Pascal is simply overclocked Maxwell
But it is though. 5.6 vs 5.7 TFLOPs. The 1070 is just running at a faster clockspeed and a narrower memory bandwidth and those clockspeeds achievable thanks to the die shrink. It's the same architecture, this has already been demonstrated by artificially downclocking Pascal to Maxwell levels and comparing performance. Clock for clock they perform almost identically.

>>60046124
How dare you, Poojeet? How dare you call a member of the Nvidia/Intel master race an Indian? I mean, what is this? It has to be a joke because I've seen this same exact post hundreds of times this year. It's always "Ramesh" so I assume it's you who keeps spamming it. An Indian, who buys Indian "GPUs" and "CPUs" accuses others of being Indian? Who do you think you are?
AMD is literally an Indian company with an Indian CEO. They hire exclusively Indian shills and you're one of them.
How fucking dare you? You're filth. You're poor and you're disgusting. Your skin is brown and you're not even human. Don't ever call an Intel/Nvidia God Ramesh EVER again, you greasy little shitskin. This is a warning, Pajeet. Move your greasy little shitstained fingers away from the screen of your filthy little android phone and stop posting, Pajeet.

>>60056810
Skyrim, Fallout 4, maybe Mankind Divided, and all the major recent open world games need at least 5GB VRAM to be playable

>>60060767
>implying the 1070 have 2816 cores and a 384 memory bus

>>60061246
It clocks higher.

>>60061250
Why Polaris is not fast as Fiji?

>>60061269
Because it's half-Fiji. With much slower memory.

>>60061187
So basically 5% of the PC gaming population can achieve "playable" settings on these games? Sure sounds like a winning strategy for those game devs. Please ponder how retarded your statement is for a moment.

>>60061246
What the fuck are you not understanding here? The bandwidth and CUDA core count of the 1070 are gimped so it can hit a performance target that Nvidia thinks is reasonable for a $400 card. It's Maxwell hopped up on speed and then squeezed through a strainer to approximate 980ti like levels.

If its core count and memory bandwidth weren't nerfed then it'd be a fucking 1080 seeing as how they're the same GPU.

>>60061345
>The bandwidth and CUDA core count of the 1070 are gimped so it can hit a performance target that Nvidia thinks is reasonable for a $400 card

nah

>>60060797
Calm down, Ramesh.

>>60061345
So...Pascal is VERY fast

File: flat,800x800,070,f.jpg (91KB, 674x800px) Image search: [Google]

91KB, 674x800px

>>60061417
Just slap an Nvidia logo on this.

>>60061394
>nah
What did he mean by this?

>>60057033
>>60046177
we are all pajeets here.

>>60061767
Yes. Long live the poo.

>>60060080
Pascal to Volta will be an even bigger jump than last time and at a reduced cost too.

>>60061880
Oh, is Volta another two node jump from planar to finfet?

>>60061908
Nope.

>>60061880
You dumb nigger. Both AMD and Nvidia got huge gains because they skipped 20nm. Both Maxwell and Hawaii were supposed to be even better.

>>60061923
That's the joke

>>60057167
>amd fell for the hbm meme
>amd created the hbm meme
fixed

File: zen.webm (1MB, 720x1280px) Image search: [Google]

1MB, 720x1280px

Fuckijg ayyyymdpoojeets I swear
All must be forced to mandatory use the loo or get shot on sight

>>60061187
This is only true at 4k resolution.

File: 1379179419029.jpg (120KB, 976x740px) Image search: [Google]

120KB, 976x740px

So finally GPU's are advancing faster than CPU's. Is it just luck or is it because of the 4K meme push?

>>60063784
>so finally
>finally

Underage get out

>>60063828
I have been waiting for a gpu powerful enough to run Fallout 3 with mods for over ten years.

>>60063861
You're better off playing an actual good game like Fallout 2 or Baldur's Gate 2

>>60063898
A printer could run those games. I like 3 better than fallout 2. Never played baldurs gate.

>>60063784
they're not advancing, desktop graphics cards have been hindered on purpose to promote planned obsoleteness

The desktop gaming card is just having more cores added to the die, while at the same time they're shrinking to die, thanks to smaller dies they can create lower wattage transistors, and fit more in the same space thanks to the advent of 3d transitor technology.

the problem though is that high end cards meant for super computers are faster and better than even your titan x cards, because they have more ram and cores.

gaming cards have been gimped on purpose as to not compete with grid array cards.

grid array cards drivers are gimped to not perform well with games.

again drivers are the important factor in performance with games, nvidia being optimized for most games even though their cards have slower compute is a clear example of why windows cucks and nvidia shills can beat amds rebranded cards.

atis architecture is better, that's why all bitcoin mining rigs used amd, faster computation.

>>60063916
ewwwwwwwwwwwwww LEAVE

File: Raja.jpg (97KB, 694x454px) Image search: [Google]

97KB, 694x454px

>>60055862
>Raja holding a Vega chip

With a die this size and HBM2 Nvidia should be fucking panicking.

This bitch deliver 60+ FPS on all Ultras in 4k where Nvidia cant even set foot.

And hardly will with GDDR6 unless they go full HBM