Quantcast
[ 3 / biz / cgl / ck / diy / fa / g / ic / jp / lit / sci / tg / vr / vt ] [ index / top / reports / report a bug ] [ 4plebs / archived.moe / rbt ]

Due to resource constraints, /g/ and /tg/ will no longer be archived or available. Other archivers continue to archive these boards.Become a Patron!

/g/ - Technology


View post   

[ Toggle deleted replies ]
File: 312 KB, 1471x932, NVIDIA-Pascal-GPU-Chip-Module.jpg [View same] [iqdb] [saucenao] [google] [report]
49734746 No.49734746 [Reply] [Original] [archived.moe] [rbt]

>"Pascal reportedly has 17 billion transistors and up to 32GB of HBM2 based vRAM"

>"The Pascal GP100 GPU allegedly has 17 Billion transistors and around 32 GB of HBM2 based vRAM. If these reports are correct, then the GP100 >will be an absolute beast in terms of graphics processing power. In comparison, the current Nvidia flagship, the TITAN-X has 8 Billion transistors, so >we are looking at an alleged jump of more than double. There are however, two major problems with the theory. We have about 1 Billion transistors >more unaccounted for, and we are assuming a die size equal to that of the TITAN-X GM200 (601mm^2) which frankly speaking is just not going to >happen on a brand new node. Not to mention that Nvidia will want to keep leg room for future Pascal variants on the same node.
>No, a die size of 601mm^2 is very unplausible. Assuming the report is true, I believe (caution: speculation) we might be looking at a combined >transistor count. In the meantime however, there is finally cause to rejoice. The era of the 28nm GPU is over, and the age of sub 20nm FinFETs has >begun."

Not sure if /g/ had a thread on this yet? Holy fuck though, really exciting times are ahead. Glad that I cancelled the 980 TI that I had ordered the other day. Anyone know when these will start to come out? Maybe Q2 2016? Do you guys think Nvidia will roll out big Pascal first?

Source: http://wccftech.com/tsmc-begins-volume-production-16nm-finfet-nvidia-pascal-gp100-gpu/

>> No.49734764

yeah and amd's card will be a foot long with three watercoolers on it.

>> No.49734802

>>49734746
It's too bad they named it after a gay Frenchman.

It will have a male out, and you'll have to somehow dock it with a male VGA or HDMI cable.

>> No.49734824

>>49734746
that beautiful symmetry

>> No.49734863

>wccftech
So you couldn't find a source any less reputable? Is semiaccurate still around?

>> No.49734875

>>49734746
>32 GB of HBM2 based vRAM
>drops down to 2 FPS when it's above 3.5 GB used
lel

>> No.49734879

>>49734802
At least it's not Volta

>> No.49734886

>>49734746
>32 GB of HBM2 based vRAM
bullshit

>> No.49734888

>>49734863
Source: http://www.taipeitimes.com/News/biz/archives/2015/08/11/2003625049
Source: http://hexus.net/tech/news/industry/85565-tsmc-starts-volume-production-16nm-chips/
Source: http://www.extremetech.com/computing/212221-tsmc-quickly-ramping-up-16nm-volume-production

>> No.49734897

>>49734875
>drops down to 2 FPS when it's above 3.5 GB used
OH MY GOD I AM LAFFING SO HARD THIS GUY IS SO GODDARN FANNY OH GOD I AM DYING HAHAHAHAHA

>> No.49734935

finally, enough VRAM to cope with badly programmed console ports.

>> No.49734957

And it'll only cost $5000
Boy I feel sorry for all you poor people

>> No.49734963

>>49734802
>hating on pascal because he liked dicks
Fuck off /pol/, for real. The entirety of this board wouldn't even come close to matching the achievements of that man. You wouldn't even be able to play your stupid fucking vidya without his advancements in geometry and mathematics, which is probably why nvidia named this line after him.

>> No.49735006

>>49734963
Honestly I think the world would be better off without his degeneracy and achievements both.
They more than cancel out.

>> No.49735008

>>49734957

> $5000 vidya card
> same performance as an optimised Playstation 4

>> No.49735012
File: 427 KB, 1200x1600, 1439711917781.jpg [View same] [iqdb] [saucenao] [google] [report]
49735012

>>49734746
>buying n-wehateopensourcepayforeverfeatureextra-idia

do you hate freedom?

>> No.49735029

>>49734963
I was just trying to lighten /g/'s day, I wasn't being serious.

>> No.49735074

>>49734888
So every other source is making a distinction between 16nmFF and 16nmFF+, the former used for the likes of the low power SoC used in the iPhone, the latter being the type used by GPUs and HPC CPUs, while wccftech is just screaming "ZOMG PASCAL IS GONNA BE AWESOME AND WE THINK IT'S GOING TO SOON. PROBABLY LATER THIS EVENING."

>> No.49735085

>>49735012
The drivers work pretty gud (much batter than AMD) on linux, for the fact that they totally hate open source

>> No.49735249

>>49735085
Yes, nvidia drivers are much batter than anything else.

I also think the other anon is referring to the fact that NVIDIA don't release any details the FOSS community can use, just binary blobs, as opposed to everyone else that at least make an attempt at opening their stuff up.

>> No.49735510

>>49734875
underrated post kek

>> No.49735513

>implying they will double transistor account and give 32gb to average people
>implying it won't be some $9999 elite gaymer model
>implying they will not milk it and release 4 and 8gb models first

>> No.49735575

>>49734802
>gay Frenchman

You sure like being redundant.

>> No.49735635

>>49735008
AMD shill

>> No.49735660

>>49734746
>Glad that I cancelled the 980 TI that I had ordered the other day

You're glad that you didn't order a 600$ card now so you can buy a 10000$ card in a year ?

>implying it won't cost 10k$

>> No.49735663

>>49734746
>Glad that I cancelled the 980 TI

Enjoy waiting a year for a product that probably wont be half as good as people expected it to be.

>> No.49735687

>>49734746
>Q2
AHAHAHAHHA keep wishing

>> No.49735805
File: 648 KB, 1200x744, image.jpg [View same] [iqdb] [saucenao] [google] [report]
49735805

>>49734746
recent image from the pascal test facility

>> No.49735828

>>49735805
I've advised them not to test it with an Amd 9590

>> No.49735860

>>49735687
We can make some bets about what's going to be released first, Pascal or Zen.

>> No.49735866

>>49734746
>17 Billion
Pascal confirmed for new Fermi. All hail the fire gods!

>> No.49735931

oh man, Call Of Duty is going to his 300 FPS with this bitch.

take my money alreadyz

>> No.49735941

>>49734802
Day made haha

>> No.49735968

>>49735805
Damn the render quality is photorealistic. I look forward to this gpu.

>> No.49735977

dollar per gigabyte between HBM and GDDR are the same. Probably higher for HBM due to the limited initial production output. I am unsure how people can think that nVidia will try to sell such an incredibly expensive card to private users.

It's for server farms at best.

Furthermore I don't see how FinFets will shake the GPU industry. The Sandy Bridge -> Ivy Bridge jump to finfets was at best the regular 5% performance boost. Also smaller processes don't necessarely mean an increased transistor count as the heat drasticly increases when you pack the electrical components closer together.

If you compare Sandy Bridge side-by-side with Haswell you'll find that Haswell always has the upper hand performance wise over Sandy bridge even though its transistor count is significantly smaller. So higher transistor count != higher performance

>> No.49736445

>>49734897
nvidiot pls

>> No.49736526

>>49735968
All thanks to Nvidia DebrisWorks™

>> No.49736535

>0.2gb superfast hbm2
>31.8gb ddr1
>it's a feature

>> No.49736808

>>49734746
>op can't into greentext

>> No.49736811

>>49735977
>I am unsure how people can think that nVidia will try to sell such an incredibly expensive card to private users.
>le way it's meant to be le played goy

>> No.49736824

This is a green text failure of biblical proportions.

>> No.49736860

>>49734802
>repressed homosexuality detected

>> No.49736889

>>49734746
So you're going to wait an entire year?

>> No.49736892
File: 1.13 MB, 1200x2000, applethen.png [View same] [iqdb] [saucenao] [google] [report]
49736892

>>49735977
this

3.5gb masked as 4gb standard, 6gb for their $500+ gpu, and 8gb for their $1,500+ gpu

32gb cards will probably be for the professional market, like their quadro's. for they can easily ask $8,000 for it.

>> No.49736980

>>49736892
The Titan X has 12gb and some crazy fucks still buy it.

They will probably make some Titan II or something with 32gb HBM for $1500. They know it sells in the end.

>> No.49737021

I just want to replace my 2gb 7850. Aged pretty well but it is showing its age.

>> No.49737059

>>49736980
Why? they could sell a 12gb HBM Titan for >1500$. Why the fuck should they sell a 32gb card for so little money ?

>> No.49737081
File: 23 KB, 320x320, America_reading_is_for_faggots.jpg [View same] [iqdb] [saucenao] [google] [report]
49737081

>>49735977
If higher transistor count doesn't mean higher performance for GPUs, how is it that they are faster already?

Spoiler: You think you understand modern processors but you don't. Educate yourself.

>> No.49737096

>>49735931
If you already don't hit 300fps on BLOPS II maxed out, your setup needs some rethinking.

>> No.49737119

>>49735968
pascal has explosionworks which realistically creates the explosions feel like you are in the middle of it

>> No.49737126

>>49735977
>The Sandy Bridge -> Ivy Bridge jump to finfets was at best the regular 5% performance boost.

That's because Intel keeps adding more power to the GPU side of their chips.

>> No.49737144

>>49737119
>gpu turns its fans off to simulate explosion tier heat

>> No.49737151

>>49736980
>>49736980
>The Titan X has 12gb and some crazy fucks still buy it.

Retards buy it for gaming.

Programmers and scientists buy it for gpgpu.

>> No.49737173

>>49737119
Don't forget
>Unoptimized on Nvidia GPUs
>Deliberately gimped on AMD GPUs

>> No.49737174

>>49737151
>Programmers and scientists buy it for gpgpu.
No they don't. It has gimped DP performance.

Like you even know what that is

>> No.49737193

>>49737151
Scientists would have enough budget to get the Quadro M6000. Unless they're Russian or something.

>> No.49737211

>>49737144
kek

>> No.49737214
File: 1.92 MB, 160x163, 1427161642469.gif [View same] [iqdb] [saucenao] [google] [report]
49737214

>>49737174
Your mom has plenty of DP performance.

You do realize not everyone needs to work with doubles right? All computer graphics is basically done with floats.

>> No.49737217
File: 30 KB, 650x380, 72530.png [View same] [iqdb] [saucenao] [google] [report]
49737217

>>49737151
>>49737174
>>49737193
>this DP perf

LMAO
M
A
O

>> No.49737247

>>49737214
Yes, all computer graphics are done with floats.

Programmers and scientists use doubles.

>> No.49737353

>>49737247

Hey retard, if programmers don't use floats, who write software for computer graphics?

>> No.49737378

OP, sorry for the retarded green text. I currently game at just 1080p and I need some advice here.

1. Should I just get a 980 or the AMD equivalent now till Pascal is out.

2. Wait till pascal comes out and also get a new monitor.

3. Get a 980Ti and new 1440p monitor now.

One final thing, couldn't they also offer a consumer version of Pascal with 8 or 16 GB?

>> No.49737381
File: 52 KB, 400x400, wccftech.jpg [View same] [iqdb] [saucenao] [google] [report]
49737381

>>49734746

>> No.49737460

>>49734746
Eh, waaay overkill and will most probably absolutely suck at bang for buck. Even the complete and utter shit port of Arkham Knight didn't use more than 6gb vram if I'm not mistaken. I think anything stronger than Titan at this point is just absolutely ridiculous.

>> No.49737484

>>49737378
you don't need 980 for 1080p except if you wanna play AAA games at 144fps

>> No.49737543

>>49737378
Get a 390 for 1440p.
Get a 380 4gb for 1080p.

>> No.49737697

>>49736445
I love buying rebranded cards that were made of rebranded cards.

>> No.49737769

>>49734802
the computer was pretty much invented by a gay english man, the computer is gay and you're going to have to deal with it

>> No.49737793

>>49737081
WE NEED A LONGER BIBELINE :DDD

>> No.49737814
File: 27 KB, 530x284, knnpdqmhtvd6cio810rm[1].jpg [View same] [iqdb] [saucenao] [google] [report]
49737814

>>49737353
>>49737247
>>49737214

>> No.49737821
File: 11 KB, 188x196, 1417166039331.jpg [View same] [iqdb] [saucenao] [google] [report]
49737821

>>49734746
I will kill myself on camera if they deliver more than 12GB version.
I swear to god people would believe anything these days.

>> No.49737849

>>49735249
>binary blobs

>> No.49737902

>>49737821
Are you going to tell me wcctech isn't reputable?

>> No.49737917

>>49737902
curry pls

>> No.49738012
File: 16 KB, 385x383, 1439774388375.jpg [View same] [iqdb] [saucenao] [google] [report]
49738012

>>49737460
> What is 4k
> What is 144hz
> What is VR

>> No.49738077

>>49737460
Just wait for a game like skyrim with a modern api vulkan/dx12 and lots of mods and you'll see how it uses more than that.

>> No.49738107

Has wccoftech ever got anything right?

Please stop being like a battered housewife.

"He'll be good to me now, he was just angry when he pushed me down the stairs every week for 16 years"

>> No.49738144

>>49734963
Jesus christ shut the fuck up /lgbt/.
Seems like you can't see that his disgusting homosexuality stands apart from his achievements.
Not only extremely biased, it's also narrow minded and plain stupid.
Also equalling everyone who dislikes faggots with /pol/ is maximum godwin tier.
Go fuck yourself nigger.

>> No.49738149
File: 39 KB, 453x576, 1439764394707.jpg [View same] [iqdb] [saucenao] [google] [report]
49738149

>>49738107

>> No.49738159

>>49734897
Your ass rage is palpable.

>> No.49738209

amd raping nvidia at dx12

http://www.computerbase.de/2015-08/directx-12-benchmarks-ashes-of-the-singularity-unterschiede-amd-nvidia/2/

>> No.49738229

>>49738209
More a sign of how bad AMD drivers are

Really. Nvidia can't be bothered with DX12 for some reason though (probably because they haven't figured out a way to goywork it yet)

>> No.49738268

>>49737769
Correction, Turing invented the Turing hypothesis which was the idea that any process could be automated using a series of states with a given input.

If anything, he invented computer science.

>> No.49738391

>>49738209
Fuck, Dx12 was supposed to be better for 8 cores like 8350?? but they're performing like shit.... fuck, fuck, fuck...

>> No.49738396
File: 173 KB, 980x552, pascaline.jpg [View same] [iqdb] [saucenao] [google] [report]
49738396

>>49737769
Pascal built the first computer, called the pascaline, which is the first calculator.

>> No.49738450

>>49738391
its performing fine, I don't know what you are on

>> No.49738955
File: 299 KB, 790x808, 1439835207106.png [View same] [iqdb] [saucenao] [google] [report]
49738955

>>49738450
This is no good.

>> No.49738960

>>49734746
>UNPLAUSIBLE
Is implausible too obscure a word? Or are they just using unpossible forms?

>> No.49739184

>>49738396
>computer
It's just a calculator, unless you want to show how to do β-reduction on it.

>> No.49739219

>>49739184
>It's just a calculator
Yeah, still a fucking computer... What did you expect ? Pascal to forge an i7 in the 17th century ?

>> No.49739222

>>49738209
It's nice that by omitting the AMD driver team Radeons can actually perfom on par or better than Nvidia in that game.

>> No.49739495

>>49734886
>up to
You can expect it on HPC derivatives of Pascal, not on gaming hardware.

>> No.49739674

>>49738209
>AMD DX11 drivers are shit
What else is new? I'm glad DX12 games should work well on my 290X CF in DX12, but this doesn't solve the problem that DX11 is shit and most games won't get updated, so until DX11 is completely phased out (it's still being supported alongside DX12, as DX12 is not meant as a replacement, just a low-level alternative) we'll still be suffering from AMD's DX11 crap driver.

>> No.49739696

>>49738955
>DX12 is slower than DX11 if you have a good CPU
Fucking lel

>> No.49739703

>>49734746
pretty sure conservative estimates is like q1 2017 for release....

>> No.49739713
File: 193 KB, 420x560, NVIDIA House Fire X Technology.jpg [View same] [iqdb] [saucenao] [google] [report]
49739713

For 9999$ they should add fire extinguisher for free.

>> No.49739764

>>49739703
You actually think NVIDIA plans to have exactly 0 new graphics cards out for the entirety of 2016?

>> No.49739812

>>49739764
They can rebrand existing ones or ship existing stock with more RAM for more monies. Or higher overclocks.

>> No.49739815

>>49739219
Still, Leibniz was first in making a aritmetic machine. None gave him the credit because of fucking british people ego.

>> No.49740719

>>49734746
stupid currynigger clickbait.

The HBM2 standard supports up to 8GB per module (and Pascal like AMD Fury is expected to do 4 HBM modules), but that doesn't mean that Nvidia will be selling 32GB cards next summer or even that 8GB HBM modules will even exist in commercial quantities.

Also, believing that TSMC will ever deliver on a new fabrication node is like believing in Santa Claus at this point.

>>
Name (leave empty)
Comment (leave empty)
Name
E-mail
Subject
Comment
Password [?]Password used for file deletion.
Captcha
Action