[Boards: 3 / a / aco / adv / an / asp / b / biz / c / cgl / ck / cm / co / d / diy / e / fa / fit / g / gd / gif / h / hc / his / hm / hr / i / ic / int / jp / k / lgbt / lit / m / mlp / mu / n / news / o / out / p / po / pol / qa / r / r9k / s / s4s / sci / soc / sp / t / tg / toy / trash / trv / tv / u / v / vg / vp / vr / w / wg / wsg / wsr / x / y ] [Home]
4chanarchives logo
Pascal vs. Polaris
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.

You are currently reading a thread in /g/ - Technology

Thread replies: 255
Thread images: 26
File: maxresdefault-2.jpg (115 KB, 1280x720) Image search: [Google]
maxresdefault-2.jpg
115 KB, 1280x720
So, AMD Polaris and NVIDIA Pascal are coming out soon, what's /g/'s thoughts on the upcoming graphics cards?
>>
Pascal will debut in April, and should hit store shelves in mid June, so nVidia will be first to market. Polaris is expected around August, IIRC.

My ancient Radeon 6950 just shit the bed on me, forcing me to upgrade. I was planning to wait for the new cards to drop this summer to upgrade, but this forced my hand. So I purchased an eVGA GTX 950 SSC+ on Amazon for $75 after their $80 instant gift card for signing up with their credit card, so it takes the bitter taste out of my mouth. I'd have hated to spent more money, then have the new cards out with the die shrink shit all over my now older card. So I'll likely use the GTX 950 as a stop-gap until the fall when Zen/Kaby Lake debut to upgrade my entire system with a newer card.
>>
>>53763130
I only like games which dont have bloated graphics so I dont care
>>
>>53763130

>Pascale
16nmFF from TSCM
Refreshed Maxwell architecture (no Async-Compute)
Massive dies packed with transistors
Hot as fuck (housefire possibility)
Less power efficient
low clockspeeds
all the benefits maxwell had on 28nm planar were specific to that node and they lose them by translating the design to finFET
Relying on 512-bit GDDR5 on all cards below titan which is HBM2

>Polaris/vega
14nmFF from Samsung (9% smaller than 16nm TSCM)
Brand new Architecture specifically designed for finFET (optimized layout and performance)
runs cooler
more power efficient
same ACE layout from GCN 1.2
brand new geometry engine
Brand new primitive discard accelerator (this will improve tessellation performance by a fuckload)
HBM2 on top end vega and HBM1 on low end vega

HBM1 is good memory and will run much better on polaris than it did on the FuryX because it wont have the missmatched memory controller that cause leaking and clock instability in the modules

Fiji was designed for 20nm, HBM1 was designed for 20nm when TSCM fucked up their 20nm process AMD was forced back into 28nm so the missmatched size is why HBM is only 500mhz and runs like shit

with HBM1 on the upper range cards against Nvidia's GDDR5 AMD will have a size advantage and a power efficiency advantage
>>
File: 1458411436285.jpg (13 KB, 398x325) Image search: [Google]
1458411436285.jpg
13 KB, 398x325
>pisscal
>>
File: image.jpg (523 KB, 1362x1499) Image search: [Google]
image.jpg
523 KB, 1362x1499
>>53763291
>he signed up for a credit card for $80
Jesus Christ how poor are you Anon. You could have gotten a used card to tide you over for $70.
>>
File: Screenshot_20160330-170722.png (546 KB, 1080x1920) Image search: [Google]
Screenshot_20160330-170722.png
546 KB, 1080x1920
/sqt/ thread is at the bump limit so I'll just ask here.

My 7870 has just passed away and my CPU (Phenom II) has no iGPU, should I buy pic related now or wait for Pascal/Polaris? Is it worth the wait? Will the mid-range GPUs of the new architectures even arrive before 2017 - don't they release the flagships first?
>>
>>53763291
Why not? No annual fee, the interest rate is a little higher than I'd prefer, but it's not unruly. It just seems to be a big waste to me to spend $150+ to rent a videocard for less than six months than plop down $300+ on a new card this summer. I hate ebay/Craigslist, and I don't know of anyone who would buy the card from me, so I'm stuck with it. Besides, need I remind you, Anon, being in debt is extrmely fashionable in America. Look at the federal government.
>>
>>53763130
> be me
> have ~4 year old GTX 670 (28 nm kepler)
> get fucked by JewWorks/driver updates
> wait on Fiji ...
> fuck, no HDMI 2.0/DP 1.3, only 4GB
> wait on Northern Islands -> Polaris
> fuck, no HBM, no ~600mm^2 monster chips
> Vega out who knows when

I'm getting the 480X or whatever the unbinned Polaris 10 gets called and holding on to it at least until Navi or whatever comes out in 2018 or whatever.
>>
>>53763566
I would wait. This is going to be the biggest leap in performance between graphics card generation that we've not seen in many years.
>>
>>53763566
Polaris will be only entry level and mid-range GPUs. Vega will bring new high end GPUs in 2017.
>>
I'm leaning towards Polaris, Pascal seems really weird.
>>
>>53763130

OP Here, personally, I'm going to have a 600-700$ budget to buy the best Pascal or Polaris card, I only intend to get a single card and I have an AX860i for my PSU, so I'll be plenty fine for powering it, I just am going to wait and see what the raw power will be and get the most powerful card in my budget.

>>53763471

I kinda knew about most of this, but low clock speeds for Pascal? Where are you hearing this from? However, I will agree that it seems like AMD is really taking some steps forward that could jolt it in front of NVIDIA... Also yeah, if they get HBM working on their mid-high to high range cards while NVIDIA only has it on the TITAN, NVIDIA could see a big hurt, can't wait to see what happens...

>>53763589

Personally, I have a GTX 760 and the next card I get I'm going to try and make it last until at the EARLIEST, late 2020, I'm just hoping my CPU lasts (FX-8320 @ 4.2GHz)
>>
>>53763622

Originally I thought that too, but I thought AMD made it clear that Polaris 11 WILL be a high end card? I could've sworn they had some kind of high end thing coming out by the end of this summer...
>>
>>53763607
>>53763622
Do you have any idea when the new GPUs are showing up? I can handle waiting 3~ months but if they're not coming until near the end of the year, fuck that shit.
>>
>>53763670

There's rumors that the first polaris GPUs are already shipping and NVIDIA Pascal will very likely be out this summer like Maxwell was in 2014...
>>
>>53763661
Polaris 11 - high-end laptop/low end desktop
Polaris 10 - mid-range (300mm^2) GDDR5 card, expected to be not too far off current Fury/980/FuryX/980Ti levels
Vega 10 - beefy (up to 600mm^2/18B transistors) HBM2-based, expected first half 2017, maybe even first quarter
>>
>>53763704
>Polaris 10 - mid-range (300mm^2) GDDR5 card, expected to be not too far off current Fury/980/FuryX/980Ti levels


Well, this sounds nice. Personally, i'm hoping Polaris 10 to be 390x levels of performance. But rumor is that it's barely above a 380x.
>>
>>53763661

Vega>P10>P11

Regardless of how it performs, technically it cannot be considered high-end in terms of the AMD hierarchy
>>
>>53763704
Hmm, it just seems odd they're not pushing to compete with Pascal sooner...
>>
I intend to buy a pascal titan when it comes out and I will have no regrets.
>>
I can buy a used Asus GeForce GTX 770 2GB DirectCU II OC and a used Palit GeForce GTX 770 2GB for the same price.

Which one is better?
>>
>>53763726
AMD already demoed polaris 10 beating a 390X

and get this it beat the 390X is a 256-bit memory bus and an 800mhz core clock
>>
>>53763807
Source?
>>
>>53763807

Source?
>>
I'll reserve judgement till I know more. Most of what we know now is pure conjecture.
>>
>>53763832
>>53763829
did you not watch capsacin?
>>
>>53763837

Who's that?
>>
>>53763807
No, in the shown situation even the 390 without X gets 60 FPS in an empty room.
>>
>>53763726
That's because the rumors are started by chucklefucks who think you can directly compare entirely different architectures by their numbers alone.

The 5770 (VLIW5) has bigger numbers than the 7770 (GCN), yet guess which one gets stomped?
>>
>>53763471
Pascal it'll be, then.
>>
>>53763392
Underrated post.
>>
>>53763392
Why is it that on /g/, when talking about GPUs, people think about games, when GPUs can be used for many other things?
Is /g/ the new /v/?
>>
>>53763983

OP here, I agree with the statement too, however, it would be nice to run fallout 4 at max settings on 1080p or 1440p and easily get over 60FPS, That's my current stance basically.
>>
>>53763984

The average consumer cares about how GPUs do games other than say, bitcoin mining, or contributing GPU power to a cloud network setup or whatnot.
>>
How much will the new GPUs cost
>>
>>53764024
A billion dollars
>>
>>53764024
Eight Billion Tenths of a Cent
>>
>>53764012
So /g/ is "average consumers" board?
>>
File: shrug.gif (998 KB, 500x264) Image search: [Google]
shrug.gif
998 KB, 500x264
>>53764036
fuck you I was gonna say it first
>>
>>53764024

Realistically, I don't think it'll be much more then the last GPUs, maybe a tiny bump up.
>>
>>53764024
In Brazil the mid-range model will cost 1000$ minimum, because any new technology here is considered alien technology and therefore should cost 5x more than in USA.
>>
>>53763130
>So, AMD Polaris and NVIDIA Pascal are coming out soon, what's /g/'s thoughts on the upcoming graphics cards?
Nvidia still the only company with working 10 bit support and vsync

Nvidia still the only company that will receive my money
>>
>>53764024
Bout tree fiddy.
>>
>>53764046

I was talking about the average consumer on this board, what do you buy GPUs for?
>>
>>53764061

>>53764057
>>
>>53764061
Deep learning, rendering, mining tripcodes.
>>
>>53764056
What do you mean 10-bit support? AMD does VSync? Are you a troll or something?
>>
>>53764079
Then you'll want to look at Quadros, Firepros, and Teslas...
>>
>>53764081
If you don't understand, it isn't relevant to you.
>>
>>53764093
I'm just legitimately curious now.
>>
>>53764092
1. Too poor.
2. Don't need ECC memory or double precision.
>>
>>53764092
orrrrr you could grab a couple 280x's and be done with it...Why the fuck would you buy Nvidia GPUs for number crunching?
>>
>>53764103
Good point, well, maybe the NVIDIA titan then or the 1080 ti (assuming it's named that)? In terms of raw compute, those are your best consumer-grade options...
>>
>>53764061
>what do you buy GPUs for?
High-quality video playback

>>53764092
Yes, good goy - buy the meme cards that are literally overpriced by an order of magnitude or two. Because you “need them” for compute support. Just like you need an i7 to design a website.

>>53764098
30-bit display buffers.

As for vsync, last time I checked AMD's Linux “drivers” literally couldn't vsync at all in asymmetric multi-monitor configurations. (Different clocks, different orientations)
>>
>>53764107
Umm, Firepros are AMD...
>>
>>53764116
Probably. I'll maybe also get some mid-range AMD for integer computation.
>>
>>53764134
It all depends on your specific needs I suppose.
>>
>>53764107
Because a lot of software uses CUDA.
>>
>>53764145
This is also true, forgot about that...
>>
>>53764124
Hmm, interesting, I do like hearing about all the points brought up here and in the thread, there are things I haven't quite thought about before...
>>
>>53763471
I would buy AMD even if Nvidia were faster.
>>
>>53764180
Then you might also be interested in AMD drivers' gimped Linux/OpenGL performance (compared to Nvidia).

Also, do AMD's drivers even support KMS yet? What about VDPAU? Vulkan? Anything?
>>
>>53764207
OP here, I'm open to either side, leaving yourself closed can mean you don't quite get the best experience you can have overall...

>>53764216
Now I thought I heard that AMD drivers do fully support vulkan or will very soon. What's KMS and VDPAU?
>>
>>53764233
I just don't want to experience giving money to Nvidia ever again.
>>
File: bench.png (7 KB, 600x1111) Image search: [Google]
bench.png
7 KB, 600x1111
>>53764233
>What's KMS and VDPAU?
What's wikipedia?
https://en.wikipedia.org/wiki/Kernel_mode-setting
https://en.wikipedia.org/wiki/VDPAU
etc.

>Now I thought I heard that AMD drivers do fully support vulkan or will very soon.
Looked it up, apparently they actually support vulkan. And wow holy shit does this reveal just how badly their cards were bottlenecked by their crappy drivers.

It's little surprise to me that AMD was pushing Mantle/Vulkan etc. to begin with, it really helps the companies that can't afford to write good drivers.
>>
don't feel liek starting a whole new thread for this one question so I'll just interject for a moment...

how shitty is Intel Celeron 900 CPU?
>>
>>53763130
A lot of the rumours doesn't add up. Polaris 10 has less SPs than a R9 390 while Vega 10 is said to have the same amount of SPs as Fiji. The transistor count is expected to double this generation. Where are all these transistors being put into?
>>
>>53763130

I don't care. I bought a R9 390 for $275 for Xmas (if you include rebate) and that will have to do for the next several years. With the big exception of Fallout 4 and GTAV, everything else works great (May not be GPU's fault. May be i5 6600 or DX12 fucking shit up).

I figure I'll upgrade right around when PS5 launches. I've upgraded several times that past 6yrs and it's just too much a waste of money just for the minor details to look better for a game I burn through in a few weeks.
>>
>>53765202
i have a fucking 380x paired to a fx -8350

and gtav gives me 50-60 fps on 1080p maxed out
>>
>>53763566
I'd say buy a bottom-barrel gpu and wait with no vidya for 3-4 months, then choose between Polaris and Pascal.

But thats me. And I'm not you.
>>
>>53764281
>OpenGL
>>
>>53763984
Do you seriously believe that AMD and Nvidia market their cards to the mainstream for their utility performance?

Theres a reason that there is a dedicated workstation market, which basically no-one knows shit about. Graphics cards are for vidya. No-one knows what a workstation is good for/which ones are good. Cause no-one cares about that. Hell, I'd even say that the reason cuda is more used than OpenCL is because of the trickle effect of having a strong media and market presence. The professionals saw the forefront of the mainstream market, and as blindly as the consumer demanded more Nvidia.

In summary. No. /g/ is not the new /v/.

/g/ is filled with normies, like any other slightly popular board on 4chan.
>>
>>53764103
Buy GCN 1.0 cards.
>>
>>53763392
Translation: "I am poor"
>>
If only

If only AMD GPUs were well supported on linux

No business maxing out GTA V and having a choppy gnome
>>
>>53764308

Very
>>
>>53765252
>/g/ - Technology
Oh sorry, where you looking for /v/ - Windows?
>>
Last chance for AMD.

If they fail again, their market share would drop below 10%.
>>
>>53765446
I tried Ubuntu with 980Ti and I couldn't get it working. Every driver fucking caused horrible issues.

Week or so ago.
>>
File: well shilled.png (186 KB, 602x334) Image search: [Google]
well shilled.png
186 KB, 602x334
>falling for the AMD meme
>>
>>53764281

>Looked it up, apparently they actually support vulkan. And wow holy shit does this reveal just how badly their cards were bottlenecked by their crappy drivers.

it's obvious that they still are, the 290 should at least be performing marginally better than the 960.
>>
>>53763471
Youre a smart guy.
>>
>>53763566
Nigga my 7870 also died not long ago after 3 years of use. I am waiting for Polaris 10 desperately since it will be much better deal that 380.
>>
>>53763661
>>53763622
Polaris 11 = 950/r7 370 tier
Polaris 10 = 970/390 tier
>>
>>53764061
Mostly high quality anime video playback. A bit of vidya games.
>>
i dont care, I only ever play quake, csgo, doom and emulators
>>
>>53766044
Polaris 10 is between 380x and 390 speedwise. So it will replace the 380.

http://ranker.sisoftware.net/show_run.php?q=c2ffcdfbddbcdde0d2e3d4e7d0e8cebc81b197f297aa9abccff2c2&l=en
>>
File: 1442193039299.jpg (68 KB, 640x480) Image search: [Google]
1442193039299.jpg
68 KB, 640x480
>>53766255

>So it will replace the 380.

>amd will rebrand the rest of their 28nm lineup with a single midrange 14nm GPU coming out alongside it

could these guys be any stupider? why the fuck did they hire so many indians?
>>
>>53766282
Diversity
>>
>>53766282
Polaris 11 = Entry-level
Polaris 10 = Mid-Range
Vega 11 = High-end
Vega 10 = Enthusiast

There is no need for more than 1 card per price range.
>>
File: tips fedora.jpg (14 KB, 228x243) Image search: [Google]
tips fedora.jpg
14 KB, 228x243
>>53764024
a trillion dollars
>>
>>53766341
>There is no need for more than 1 card per price range.
I agree desu.
>>
if nvidia continues to pay developers to use shitworks and pay publishers for bundle rights, does it matter how good AMD's card will be? 390 is a lot better than 970, literally 20-30 more fps in newer titles like farcry primal and the division, yet 970 outsell 390 by probably several hundred percent.

the 3.5 meme is now well known, yet it doesnt hurt nvidia at all. I have no faith the next gen will be any better
>>
>>53763130
DP 1.3 is the only thing I cared about. What's the deal with this? Is it so hard to implement? Is it licensed and expensive?
>>
>>53766561
>390 is a lot better than 970, literally 20-30 more fps in newer titles like farcry primal

no AMD GPU is superior to NVIDIA's offerings in real world performance currently, http://forums.guru3d.com/showthread.php?t=398858
>>
>>53766561
>90 is a lot better than 970, literally 20-30 more fps in newer titles like farcry primal and the division


hahaha holy shit this what amdrones actually tell themselves to feel better

roflmao dude, roflmao
>>
>>53766561
Those games are only getting ahead because they are console ports which are designed for AMD architecture, developers aren't going to bother to rewrite everything.

The Division don't even has texture quality settings, it just streams everything, which doesn't works as good on PC as it does on consoles, because PC doesn't have shared memory like the PS4, the new Wolfenstein does the same shit, but The Division is worse because it will continue to keep loading stuff into the vram until totally full and then use the RAM.
>>
>>53766682
I played division for hours on 970, and it runs like a charm
I have 16gb ddr4, checking with afterburner I never see the game using more than 3.5 gb

there are games where the 390 is ahead, others where the 970 is, advocating one over the other is dumb and this anon >>53766561 is just baiting softly

and I dont want to be an ass, but every single benchmark comparing the 390 and the 970 have a 970 at stock clock + nvidia dynamic boost, at something like 1170 mhz.
My 970 boosts to 1350mhz by default, and I can OC it to 1500 stable.
The gains vary a lot depending on the games, but its there.
Also power consumption.
>inb4 doesnt matter
The 390 literally draws TWICE the power a 970 does for the very same performance.
Let that sink in, AMD.
>>
I guess it is time to upgrade then.

[spoiler]950 or 960?[/spoiler]
>>
>>53766727
960 is you have a budget CPU.
>>
do you think next generation will be able to pull 1440p 144fps? can 980ti do it now?
>>
AMD really must do something about their frametimes.

Average fps doesn't tell the whole picture.

Microstuttering from framedrops weigh heavy on your experience.

They should have dropped the Fiji cards.
Doubt those make any profit anyways.
>>
>>53767343
>1440p 144fps
>can 980ti do it now?
for most games, no. 1440p@144fps is equivalent fillrate to UHD4k@60fps plus a higher geometry load ignoring any added host-side burdens.

>do you think next generation will be able to
Vega and big Pascal (both sometime next year) should. First-gen Polaris and Pascal probably not except for older games, sorry.
>>
File: 1458095727937.png (457 KB, 604x601) Image search: [Google]
1458095727937.png
457 KB, 604x601
>>53763471
>Pascal
How its meant to be played
>Polaris
Not how its meant to be played

Nvidia wins.
>>
>>53767407
>1440p@144fps is equivalent fillrate to UHD4k@60fp
oh, two more generations then to bruteforce it all
>>
>>53767422
quite funny considering how the DX12/Vulkan driver model is completely fucking the JewWorks/TWIMTBP sponsored optimization paradigm.
>>
>>53765202
>May not be GPU's fault. May be i5 6600
What the fuck
>>
>>53766255
Polaris 11 is between the Fury X and 390X. Also polaris 11 = entry level. Polaris 10 is supposed to be better.
>>
Polaris
nVidias track record for giving what you pay for isn't exactly good
>>
>>53767422
Pornstars with natural boobs this size?
>>
>>53763738
why does this seem odd? the do the sMe thing literally every single time.
>>
File: 1446266362165.png (302 KB, 920x560) Image search: [Google]
1446266362165.png
302 KB, 920x560
1080ti when?
>>
>>53766255
Polaris 11
Greenland
Ellesmere
Baffin


Polaris 10
Weston
Banks
>>
>>53763566
I'm fairly certain I've seen the 4gb version go for $200 a month back.... tho for the US only.
>>
>TFW I just learned how damn loud a Fury X can get when the fans decide to tick onto 100% and stay there
Was surprised at how damn powerful it was desu.
Don't know why it was stuck at 100% @ 23c though.
>>
>>53767742
Many.
>>
>>53763692
>He fell for the waiting game meme
never again /g/
>>
AMD is going to fuck over anybody who buys anything new right now. Everything I've read and heard supports my expectation that Polaris and Vega are game changers. Not only performance wise but also they'll be aggressively priced.

Nvidia basically is going to feel like amd did when facing Sandy bridge back in 2011...
>>
>>53768696
With the only difference that they would still sell a shitload of gpus to dumb gAymurZzzZz.
In order to gain something like 50% marketshare AMD should wreck all nvidia's gpus at every pricerange for at least 3-4 years in a row, like Intel is doing now against AMD cpus
>>
>>53766341
so i guess in your world, every chip comes off the wafer perfectly working so you don't need to test anything, right?

fucking retard, this is literally what binning was made for.
>>
>>53769175
Of course. AMD is perfect.

If some people cannot afford it they can just disable a few shaders and remove an 'x'
>>
>>53766719
>The 390 literally draws TWICE the power a 970 does for the very same performance.
Source. Literally everything I'm seeing is the 970 drawing ~30W less under gaming loads, and ~50W less under furmark/stress tests.
>>
>>53763291
Enjoy your $100 amazon prime membership that comes with the card like I fell for the gift card too
>>
I have high hopes for nvidia. I'm never fucking buying and after I bought it once and the driver support was horrendous. I couldn't even play emulator games at 60 fps.
>>
>>53767742
>>53768039

Only BustyBuffy, but she is brain dead
>>
i had fun with my A10 Trinity, but i exhausted the majority of my backlog, and have a few steam sales worth of games waiting for me

gonna wait until both are released then get something in 250€ range
>>
>>53769288
There is no source, he's a fucking mentally cunted Nvidiot who has been brainwashed and probably also owns a 970 and thus wants to feel better about getting bent over and raped.
>>
>>53769669
>the average arguments of an amd user

Its hilarious how you guys never use actual figures to support your claims and live in this delusion when nvidia cards have been proven to beat amd time and time again. Keep buying $300 house fires goy.
>>
File: get.jpg (31 KB, 580x228) Image search: [Google]
get.jpg
31 KB, 580x228
Slightly better card from a year and a half ago poor fags can afford(most steam user have sub $200 gpus) but have a bunch of extra crap that just negate the performance gains when turned on.
>>
I don't give a fuck, I will just buy a R9 390 because the prices will drop down hopefully
>>
>>53769484
0.10 USD has been deposited into your account.

Keep it up, Pajeet.
>>
>>53770117
>keep it up pajeet

And is the company that's run by an Indian with an Indian workforce. At least understand your memes before using them.
>>
>>53769904
> nvidia cards have been proven to beat amd time and time again. Keep buying $300 house fires goy.
Kepler
>>
File: 1426754955116.png (1 MB, 810x7800) Image search: [Google]
1426754955116.png
1 MB, 810x7800
>>53770913

This picture is ancient, but remember how the titan was the undisputed fastest card in existence? Now look at how AMD's response - the 290x - does.
>>
>>53767988
I saw this too. I was tempted
>>
>>53770956
>the power of 2 gpus is stronger than one

No shit. It also doesn't help that games nowadays are badly optimized and really bandwidth starved.
>>
>>53771031
>badly optimized
you mean goyimworks?
>>
>>53771062
>New graphical features in early hardware are somehow really intensive.


I bet you don't even remember but when SSAO was first introduced it destroyed any GPU and that was for years until new generation GPUs were released.


It will be the same with gameworks which to this day are the most impressive set of graphical features capable of real time rendering.

Just because you GPU can't run them and you want to slowdown development doesn't means everyone thinks the same.
>>
>>53771031

Apparently you can't read. I said titan and 290x, not titan x and 295x2.
>>
File: 1457066706587.png (142 KB, 278x245) Image search: [Google]
1457066706587.png
142 KB, 278x245
>>53771195
>It will be the same with gameworks which to this day are the most impressive set of graphical features capable of real time rendering.
>>
>>53771195
>SSAO was first introduced
hm, i remember seeing it first in amnesia of all things, 2010 i think
i had 240 in 2010 due to circumstances and it worked fine on low

>It will be the same with gameworks which to this day are the most impressive set of graphical features capable of real time rendering

oh, okay then, ignorance is a bliss
>>
>>53771764

Don't worry, he probably thinks Nvidia invented tessellation.
>>
>>53763471
>Refreshed Maxwell architecture
It's derived from Maxwell, but it's by no means a refresh like the AMD 300 series.
>no Async-Compute
Not an issue, since it doesn't even benefit AMD cards by much anyway. And it's hard to develop so most game developers will rightly ignore Async for the easier to work with Gameworks.
>Hot as fuck
Source
>Less power efficient
Again source
>low clockspeeds
Again source
>all the benefits maxwell had on 28nm planar were specific to that node and they lose them by translating the design to finFET
Why are you lying through your teeth? FinFET changes nothing other than reducing voltage leaks for more consistently performing dies.
>Relying on 512-bit GDDR5 on all cards below titan which is HBM2
AMD's Polaris is relying on 128-bit and 256-bit GDDR5, and not even GDDR5X like Pascal.
>>
>>53767707
>Polaris 11 is between the Fury X and 390X
Keep believing that, AMDrone.
It only has a 128-bit memory buswidth and less transistors than a 380
>>
>>53771764
Are you idiotic or retarded anon?

What GameWorks does it is indeed super impressive because it can be done in real time with consumer hardware, yeah there are better effects out there but none of them are done in real time and require impressive hardware.
>>
File: 1458891768130.png (650 KB, 512x800) Image search: [Google]
1458891768130.png
650 KB, 512x800
it's actually depressing to see that people didn't learn a single thing from maxwell about how shitty of a company nvidia is.
if you continue to support them you're literally a fucking retard.
>>
>>53771842
STALKER COP Came in 2009 and back then SSAO was already pretty common, AMD's HDAO and Nvidia's HBAO were already common too.
In the early days SSAO caused almost a performance hit of 40% because it require compute and bandwidth which would cause older generations card to bottleneck.
>>
File: 1454395771212.jpg (26 KB, 478x483) Image search: [Google]
1454395771212.jpg
26 KB, 478x483
>>53763130
All this butthurt and shitposting over which overpriced casemonster is going to give you the highest res textures in the next overblown overhyped AAAA+ $60 (With $30+ DLC) shitfest.
>>
>>53772615

There is nothing inherently impressive about gamesworks - all of its effects can be implemented in other ways (some of which are more suited to modern architectures).

AMD's tressfx fundamentally does the same thing as hairworks but tressfx is more compute based which is the direction game engines have been moving in for a while. Raw compute is pretty relevant these days and its just something Nvidia tends to suck at from kepler onwards. Sure there are plenty of other factors in performance but the writing is on the wall.

Fun fact: on paper a 390x is as fast as a 980ti for single precision compute (and a fury x is the fastest single gpu ever built). Naturally paper specs do not factor in software bottlenecks and a host of other things.
>>
>>53772829
>and a fury x is the fastest single card

My God, the shilling is in at full force today.
>>
>>53773139
>Its hilarious how you guys never use actual figures to support your claims and live in this delusion when nvidia cards have been proven to beat amd time and time again. Keep buying $300 house fires goy

Yep
>>
>>53772626

>releasing a new architecture instead of rebrands is a bad thing

if >3.5 is your reason for disliking maxwell, you'd be retarded. nvidia's cut down cards have always had cut down memory buses. see the 8800gt, 560ti and 660ti for other examples of the sae shit. people still loved those cards for the price/perf they provided.
>>
>>53772664
holy shit, someone who actually gets it.
>>
>>53773199
I don't even play games or buy anything. I just Luke watching you guys get mad over menial bullshit like which card can offer you 5 frames more
>>
File: REST IN PEACE AMD.png (33 KB, 600x1111) Image search: [Google]
REST IN PEACE AMD.png
33 KB, 600x1111
>>53772829
>Fun fact: on paper a 390x is as fast as a 980ti for single precision compute

kek
>>
>>53773139
>>53773157
>>53773308


http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-GeForce-GTX-980-Ti-6GB-Review-Matching-TITAN-X-650

>980ti
>5.6 Tflops
>290x
>5.63 Tflops

http://techreport.com/review/28513/amd-radeon-r9-fury-x-graphics-card-reviewed/4

>fury x
>8.6 Tflops

Stay retarded /g/.
>>
retarded fucksticks all of you

Polaris10 is the midrange GPU
it has the following:
>256-bit memory bus GDDR5
(new modules are 15% faster than DR5 modules on R9-300 series and GTX900 series)
>2,300 shaders
maximum core clock still unknown

this is actually an amazing card because my little /g/tards it fucking beat a 2,800 shader 390X running Hitman on Max settings at 1440p

and it beat the 390X with only a 800mhz core clock (engineering sample with alpha drivers)

thats right kiddies. a 800mhz, 256bit, 2,300 core Polaris10 card beat a 1050mhz, 512-bit 2,800 core R9-390X at 1440p on one of the newest games on the market.

Protip: the stock clock will be higher than 800mhz and the overclock will be WAY higher than R9-300 cards.......

AMD used to have weak cores and relied on massive memory bus and memory ammounts to stay on par with nvidia. Now it seems they have reversed this completely and they have build insanely roided out cores with less memeory bandwidth. Im guessing they will rely on the memory layouts to keep each card in its performance category the only explanation for why they would design a GPU like this is to push 144hz at 1440p thats literally what the polaris10 was designed for and the 256-bit bus chokes it at 4K which is probably intentional since they want to sell vega cards for 4K and dont want a single midrange polaris card be able to push 60fps on 4K on high settings.
>>
>>53773341
>bigger numbers guies1111!!!

Just admit you have no idea what the fuck you're posting about.
>>
>>53773341

it's not retarded to laugh at your claim at all, tflops are not an objective measurement for comparing two completely different GPU architectures. they're just a measurement of how many instructions the GPU can process, it doesn't take the actual time that will take into account or whether you can even design a workload to 100% utilize the GPU (which is a major flaw of GCN currently)
>>
>>53773398

you sure to seem to know a lot about a card that doesn't even exist on shelves yet.
>>
>>53773505
They demoed and showcased it. Most of what he said is true.
>>
>>53773553

>believing amd marketing statements

were you not around for phenom 1 and bulldozer? kek
>>
>>53773589
At least it's something. So far nvidia showed nothing from Pascal and it may not be that great.
>>
>>53773398
>This one car's engine ran at 4000rpm
>That other car drove 100mph
>Therefore the car in that garage over there is faster than some completely different car
Nobody knows what clocks the card at GDC was running at so your entire post is moot.
>>
Also isn't Pascal going to last after 2018? Or is nvidia going to build another card to compete with Vega?
>>
>>53773589
>>53773636

In order to get the performance they have in that hitman setup with their layout they had to dramatically redesign their architecture.

by doing what they are doing they are actually improving performance and reducing costs because a card thats kept in check with a small memory bus as opposed to a gimped bus (le 970 3.5 meme) will perform much more smoothly and consistently than the reverse.

Polaris 11 is entry level budget card we dont know its actually performance because they did a capped 60fps demo to only show its power efficiency not its actual performance. based on the 128bit bus (15% faster than the 128 memory on gtx950/960 do not forget this is important) its designed specifically for 1080p and if the architecture runs the same as polaris 10 but smaller then its going to curb stomp the 960 and the 380 and its going to run hot on the heels of a gtx 970 at 1080p with the 380X and 970 pulling away only at 1440p because once again its getting choked by its memory bandwidth.

they are tailor made cards for specific use cases that will stomp the previous generations and be cheaper but are designed in such a way that they dont fuck with the sales of their higher tier cards
>>
>>53773766
More powerful pascal chips will come out to compete with vegas. It's not possible for either nvidia or amd to produce large dies on 14/16nm yet so they're both treading safe ground a producing upper midrange cards this year, then early next year it's the HBM2 £550 monster cards.
>>
>>53773938
Then what's the point of even talking about it when both cards have nothing but speculation and both higher end cards will release next year? We barely know anything about either to begin with, other than and playing games at 60 fps at 1440p (which isn't that impressive to begin with, current cards already dish out 60+ frames in most games).
>>
>>53773810
>muh buzzwords
Shill for the high score
Remember muh 3.5 Gigabytes
>>
I sure hope amd polaris is gonna allow me to get 60 fps in diablo 2 xD. You see, I run a 20 monitor setup because I need to increase the effeciency. I bot 20 characters at once but recently with my gtx 970 im starting to lag even with 1024/1074 screens. Hope polaris make my pc runs good nough to allow me to get my characters to lvl 40 fast enough.
>>
>>53775256
Are you retarded?
>>
>>53763130
New node is gonna b gud, unless you're a hater for some miniscule reason. Better price/performance than 300 or 900 series, better everything. Will it be good enough for /g/? Probably not.

>>53763471
Undoubted both companies will come out with good cards, although AMD's have the advantage of being in the console family.

>>53763661
Polaris 11 is 470
Polaris 10 is 480
Vega in 2017 is 490 and Enthusiast

>>53763807
Polaris 10 won't beat 390x. Polaris 11 might, but doubtful.

>>53764024
I'm expecting Polaris 10 (480) at 250USD
Polaris 11 at 200 or less
>>
>>53778446
>Polaris 10 won't beat 390x. Polaris 11 might, but doubtful.
Fuck, I meant
> Polaris 11 won't beat 390x. Polaris 10 might, but doubtful.
>>
>>53763582

>hate ebay/craigslist

there's your problem, you pussy ass bitch.

>$450 980ti on my craigslist
>>
I can't let down my homeboy Huang

Definitely going for Nvidia this time

Every good advice from AMD was just not worth it and their cards just dissapoint desu
>>
>>53763527
>brand new card for $75
>or
>used card for $70
Are you fuckin terminally retarded?
>>
Damn it, never mind the 380x anymore. Waiting for your arrival, Polaris 10.
>>
>>53779576
I'm sure it will cum all over you.
>>
>>53763527
>Are you stupid or just terminally naive

The irony is too strong.
>>
>>53773553
They demoed Hitman running at 1080p and achieving FPS higher than what a GTX 980 running Hitman at max could. Ok that's impressive, but they didn't specify if it was running at medium settings or max.
>>
>>53763130
Nothing below 5000 shaders and 8 GB HBM2. I currently run a 780Ti and will keep it until the 1080Ti.

Pascal cards will be shitty GDDR5 cards, so lets see if AMD can deliver anything.
>>53763589
The drivers for Kepler seem shitty, but AMD is also shitty. Nothing but rebrands and a crappy 4GB 4K card.
>>53767883
Not before 2017. I hope they release it weeks before ME Andromeda releases. Otherwise I will need to survive some weeks on shitty mdeium 4K quality.
>>
>>53765223

that include shadows,grass and the advanced graphics settings?
>>
here in canada a GTX 980 costs about $660.
would the GTX 1080 be rougly the same? or atleast under 750?
>>
>>53781488
Well the 1080 is faster so expect around 1000
>>
>>53781831
yeah but aren't the new gen of cards usually supposed to be around the same price as the previous gen?
>>
I want Vega but my 550 ti died on me last week.

Should I buy a 390x after the price cut and wait for Vega?
>>
>>53765318
cuda is used over opencl because it was first, and its easier to implement.

as for workstation cards, every workstation card in a system that they designate is guaranteed to not crash.

in almost EVERY situation, ram is the bottleneck, and amd has more ram, however if they don't have opencl or well implemented opencl, you go nvidia because it definitely has cuda.
>>
>>53764248
>>53764233
as long as amd is ball park as good, im going amd, and so far, amd has been that every single generation.
>>
>>53764054
brazil is a shithole
>>
>>53772388
Polaris is the low end... how many times do we have to go through this, vega is what amd is calling their good shit now, and its lining up to launch around the time nvidia puts their shit out too.
>>
>>53763589
dp to hdmi 2 converter... they are 30$ but they work... that said, why are you using a 4k tv for a monitor?
>>
>>53782590
>Radeon HD 4890
$250
>Radeon HD 5870
$380
>Radeon HD 6970
$370
>Radeon HD 7970
$550 (nominally; $750 everywhere for the first few months)
>>
>>53763607
>This is going to be the biggest leap in performance between graphics card generation that we've not seen in many years.

Where have you read that? I thought it was just another increase like we've seen from the r9 200 series to the r9 300 series. What makes you think the next wave is going to be such a giant jump? Do you have a source?
>>
man I'm fucked

I have an lga 1151 socket - the next CPUs from Intel aren't even going to be 1151
>>
>>53783229
>as for workstation cards, every workstation card in a system that they designate is guaranteed to not crash.
Wat. Workstation cards have unlocked OpenGL and OpenCL performance, while consumer market cards have throttled performance in professional applications. That's the only difference.
>>
>>53783323
Unfortunately true. Only the bigchips are more expensive, a 980 will cost the same as a 980Ti on release. Thats why I only buy bigchip cards.
>>
>>53783324
>Where have you read that? I thought it was just another increase like we've seen from the r9 200 series to the r9 300 series.
You realize that the 300 series is literally just a 200 series rebrand with a memory OC, right?
>>
>>53781325
the advanced settings break the engine, the game literally can not support them but because the game out right doesn't crash, they put the settings in a separate tab.

you know how fallout 4 looks like crap for how it runs? the base of the engine is heavily modded gamebryo, they pushed the engine FAR beyond the point of what it should do and it has a horrendous hardware penalty for it, thats what the advanced settings are, pushing the engine FAR beyond what it was ever made to do.
>>
>>53783361
I wondered why my 6950, even OC'd, couldn't keep the FPS above 45 on a 1080p screen even on medium settings.

It's a joke.
>>
>>53783358
Okay, rephrase: what makes you think the performance jump will be better than what we've seen in the past 4 years?
>>
>>53783345
its not throttled performance, its just pure no optimizations what so ever, because those optimizations are what cost money. if i remember right to get certified, it costs something along the lines of 20-30 grand a program, that cost is put 100% onto pro studios that need them.

also, amd puts out a what, 16/24/32gb ram card, i forget its been a long time sense i looked into it, when you need more ram, theres literally no choice outside of a pro card.
>>
>>53783386
We missed 2 die shrinks and now we're getting three at once. Remember the difference between 40nm (GTX 580, HD 6970) and 28 nm (TITAN, HD 7970)? Triple that.
>>
>>53763874
was it? remember, the 5000 line the 5870 was the highest end single gpu card, then then with the 6000 they changed it to 6970 was the highest single gpu card.

i mean sure the difference between 5770 and 7770 would be there, but was it that big a difference? if you want to compare like to like, it would be 7870 vs 5770
>>
>>53765394
fuck off
>>
>>53783449
>We missed 2 die shrinks and now we're getting three at once.

No, we're getting ONE die shrink. Cite yourself.
>>
>>53772444
to be fair, it also has a 14nm smaller node that can put out 40% more performance if you forgo power savings.

that said, i doubt it will be that high in the hericy pole... now Polaris 10, that one i have no idea about, the demo was maxed out hitman in an area that even high end cards drop below 60fps, but this thing didnt go below 60... thats the interesting one to me.

vega and that are nice, but i dont want to spend 600$ on a gpu that im going to need to replace because it still can get a solid 60 at 4k without reduced detail... the next next gen is the one im willing to fork out 600 for.
>>
>>53765241
or get the extreme high end and sell it right when the new shit comes out for a minimal loss.
>>
>>53763566
Don't buy rebrands.

Get a used 7970 or something. Or another 7870.
>>
>>53783449
>>53783569
It's a two node jump. 28->20 is one, 20->14 makes two.
>>
>>53783613
My 6950 annihilates the 7970.
>>
>>53771195
i would rather effects be baked and faked then done real time, take the farcry 4 holy area with the bells, the inside, nothing can be moved, but it requires ssao of some kind because thats how all the shadows are handled, the game forced me to turn ssao on, something i will gladly give up because in most games, it adds very little for a significant performance hit.
>>
File: herp derp.png (8 KB, 1235x717) Image search: [Google]
herp derp.png
8 KB, 1235x717
>>
Polaris 10 = R9 390 replacement
Polaris 11 = R9 370 replacement
Vega = Fury replacement / next gen Titan competitor
>>
>>53785112
But 390x already rivals Furyshit in every new game.
>>
>>53785112
Polaris 10 = R9 380/X replacement

Fixed that
>>
>>53783739
I bet you would also love fake reflections.
>>
>>53785084

>Green column thinner and worst drawn

Take that, Nvidiots!
>>
>>53783340
Source?
>>
File: 1459181504053.jpg (20 KB, 552x460) Image search: [Google]
1459181504053.jpg
20 KB, 552x460
>>
File: c3%20h%201920[1].jpg (156 KB, 621x980) Image search: [Google]
c3%20h%201920[1].jpg
156 KB, 621x980
>>53783626
How about you check some reviews first before spouting bullshit, anon? 7970 sometimes almost twice as fast as your card.
>>
>>53763130
I'll check the prices before buying.
>>
>>53785149
That would be Polaris 11.
>>
>>53767644

Because GTA5 maxes out my CPU. Almost as soon as I start it up, it becomes a pixilated mess. Some rumors on the net claim skylakes may not work on some games. No idea if that's true and have NO fucking clue why GTA, a game released 2yrs ago, won't run on current gen PC hardware. Already did all the usual stuff...twice. So giving up and just going to buy it on PS4. I don't have the patience I once had for Windows bullshit.

As long as this R9 390 runs most games, for $275 for an 8GB on-board memory card, it was still worth it.

If anything, this PC build has taught me getting the newest hardware doesn't equal the best value. Hell, GTAV and Fallout 4 ran at "very high" settings on a 8yr old Mac Pro w/680 GTX Mac Ed perfectly. WTF?!

So fuck getting a Polaris and Pascal anytime soon.
>>
>>53785919
you're doing something retarded
>>
>>53785887
Polaris 11 is the slower card.
>>
>>53785940
mostly it's for futureproofing, its better to get an overpowered pc to last you for decades than an underpowered one. you're right tho, games from this gen are so badly optimised that even current hardware cant handle it.
>>
>>53785940
Skylake had trouble with some prime numbers, update your bios to fix it.
>>
>>53785998
But then OCing doesn't work.
>>
>>53785986

Yeah, I get that. I guess the macfag in me just wants shit to werk without it being a fucking project. I've got too much other things to be working on to research and patch games. I kid you not, the compass in Fallout 4 doesn't work on 390s. Someone even made a downloadable fix for it. In the end it didn't matter as the game just disappears after a few minutes of playing (No error message or anything. It just cuts out).

>>53785998

Waiting for motherboard bios to get out of beta (have the newest below that). That prime95 doesn't sound like it would cause a game to max a cpu. My best guess is it's a direct x 12 issue. It's no secret Rockstar doesn't give a shit about PC gaming. Maybe it was just me, but getting that rockstar social club working took way more effort than it should've.
>>
>>53783626
In which dimension? Maybe a 6990, but not a 6950. Its below the top tier 6970 and this is an old flagship that was released before the 7XXX.
>>53785519
Maybe he is missing something. I recommed him to try the 6950>6970 shader unlock.
>>
>>53785953
Exactly.
>>
>>53783374
That would be your cpu/ram bottlenecking if fo4. It's weird like that. I get 45-60 fps at 1080p on my 2gb 7850 with medium high settings. I would get down to 25 pretty regularly before updating my processor.
>>
File: image.png (71 KB, 251x251) Image search: [Google]
image.png
71 KB, 251x251
>>53785998
>have to update bios for xmp profile
>do it
>next day new bios for prime issue leaves beta and goes full release
>>
>>53763471
Pascal gets my vote.

Tsmc 16nm > samsung 14nm. Just look at batterygate.
>>
File: 25049629233_21154110b5_o.jpg (64 KB, 800x600) Image search: [Google]
25049629233_21154110b5_o.jpg
64 KB, 800x600
>>53787768
Live on the bleeding edge, you get cut. You're basically an unpaid beta tester. Get used to it, faggo.
>>
NVIDIA CRAP GIMPIDIA NOT WORTH IT. WHEN PASCAL COME THEY WILL GIMP MAXWELL SAME AS KEPLER WAS GIMPED DELIBERATELY. DONT LIE YOURSELF NVIDIA IS SCAM, FRAUDERS AND ASSHOLES. WITHOUT COMPLAINS THEY WOULD NEVER RELEASE DRIVER FIX, 10 FPS MORE W3 AND TAKE IMPROVEMENTS AGAIN IN NEW DRIVER! THOSE NVIDIA GARBAGE COMPANY SHOULD BE BOYCOTTED! AMD 4 EVER AMD IS FUSION AND FUTURE NEVER EVER NVIDIA GPU AGAIN
>>
>>53772388
>no Async-Compute
>>Not an issue
Nvidiots.
>>
File: why.png (113 KB, 229x238) Image search: [Google]
why.png
113 KB, 229x238
>>53789810
you ok there buddy?
>>
>>53789666

No
>>
>>53789666
youve got a point but we still dont know how those problems will translate to a full scale gpu. samsungs fabrication process is shitty compared to tsmc though
>>
>>53764024
1 magma
>>
>>53789846
Found this on a youtube video.

The AMD fags are really suffering.
>>
>>53764024
GTX 1070 - $350
GTX 1080 - $450-500
GTX 1080ti - $650
>>
>>53790372
Nice one.
>>
>>53764024
We know that Polaris 10 will cost less than $300 and that Polaris 11 will be even cheaper.

That's it. The rest is pure guesswork.
>>
>>53790372
1080ti won't be any less than $1000. Judging by the price inflation in gpus and cpus recently, it may even cost more. Amd cards are probably going to be around the price ranges you mentioned.
>>
>>53789843
>le async compute meme
>le +5 vram meme

Love how and only keep bringing these up.
>>
>>53792023
Yeah because a 50-100% speed improvement for free is totally not a big deal.
Fucking nvidiot.
>>
>>53792601
>for free
uuuuh....
>>
>>53772388

>It's derived from Maxwell, but it's by no means a refresh like the AMD 300 series.

polaris will be a completly new thing, pascal isnt.

>Not an issue, since it doesn't even benefit AMD cards by much anyway. And it's hard to develop so most game developers will rightly ignore Async for the easier to work with Gameworks.

Being this retarded to actually think async compute isnt a major factor just because a guy new to dx12 found it difficult, you do know that devs had issues with dx11 as well right? they all do when its new.

Also gameworks is shit. stay mad, stay bad, and stay sad nvidia slave.
>>
>>53786135
>It's no secret Rockstar doesn't give a shit about PC gaming

That was apparent with that POS port of GTA IV that ran like dogshit
>>
>>53785168
depends on how well faked it is. i have seen ones that the only way i knew they weren't real is they didn't move, and then i have seen implementations that are so fucking bad i would rather them not be there at all.
>>
>>53791975
price inflation is largely due to size of the dies.
>>
>>53763130
Polaris looks like it'll outshine Pascal, but I'm content with keeping my XFX Fury for the next few years. Anyone thinking of upgrading, I would hold off until late summer when Polaris hits shelves, if you can wait.
>>
>>53792942
>polaris will be a completly new thing, pascal isnt.

lol? where are you even getting this retarded meme from? AMD already came out and said 'polaris' was derived from GCN. the only meaningful difference between polaris and fiji is going to be an increased ROP ratio and HEVC decoding/encoding supported added to the VCE.
Thread replies: 255
Thread images: 26

banner
banner
[Boards: 3 / a / aco / adv / an / asp / b / biz / c / cgl / ck / cm / co / d / diy / e / fa / fit / g / gd / gif / h / hc / his / hm / hr / i / ic / int / jp / k / lgbt / lit / m / mlp / mu / n / news / o / out / p / po / pol / qa / r / r9k / s / s4s / sci / soc / sp / t / tg / toy / trash / trv / tv / u / v / vg / vp / vr / w / wg / wsg / wsr / x / y] [Home]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.
If a post contains personal/copyrighted/illegal content you can contact me at [email protected] with that post and thread number and it will be removed as soon as possible.
DMCA Content Takedown via dmca.com
All images are hosted on imgur.com, send takedown notices to them.
This is a 4chan archive - all of the content originated from them. If you need IP information for a Poster - you need to contact them. This website shows only archived content.