> you fell for the "gcn will dominate maxwell in dx12" meme
>>52056845
>single async compute usage
Nvidia card only support 1 async compute. AMD Fury has 32+. 1/32 units are being used. There is no dynamic usage. You have to remember, its a Unreal engine, which is nvidia partnered company and will only optimize for nvidia cards.
both cards are a massive waste of money.
show me 390 vs 970
>>52057219
Not 390, but a 290x which is roughly 3% weaker than 390, so use this.
Also old 7970 (280x) beats 960 by more than 30%.
>>52057306
>>52057219
390 beats 980 as well.
>>52057306
>680
Why do nvidia cards age so poorly?
the facts are that yes even in dx12 the 980TI compares to a FuryX, but thats because the card is pretty amazing, still not worth the 650$
tho, but in the mid-range the 390 is clearly outpacing the 970 in every benchmarks
the 970 is the main card nvidia is selling so yea.
>>52056845
>gcn
>dominating anything other than your power and heating bill
>>52056845
>beta
Not a valid benchmark, move along. Besides, if devs fail/refuse to use asynch compute that's their fault.
Also, optimise your pngs.
>>52057371
Nvidia basically stop working on performance improvements the moment they release something new. AMD on the other land literally only just stopped working on performance improvements for the HD 5000 series.
>>52057337
gifs can be optimised too
>>52057495
>if devs fail/refuse to use asynch compute that's their fault.
Why do you say its a dev's fault? UE4 is a primary nvidia game engine. Its a design choice to limit async compute to fit the nvidia paradigm. This will help them sell more 980 ti.
>>52056900
If this generation is going to be anything like the last one, unreal is going to be an important engine to be optimized for.
>>52057532
Game developers are not graphics card retailers. Some are, however, corrupt and are bribed to produce games that are designed to run poorly on certain hardware (usually AMD cards). This is unethical, such games should be boycotted and certainly shouldn't be used as benchmarks.
>>52057557
I wouldn't necessarily say they are corrupt. They don't actively try to harm AMD cards, but rather optimize nvidia cards.
Its same with Intel C++ compiler, where they don't actively harm AMD cpu but rather offer optimized path for Intel CPU.
>>52057557
You can fight the cuckwars yourself while we enjoy video games and get the maximum fps
>>52056900
Ashes of the singularity is vendor neutral, arguably leaning towards AMD since they're a former Mantle partner, uses async compute on AMD systems, and the 980ti is still faster than the Fury X.
Async is so overhyped, jesus.
>>52056845
Reminder that's only on stock settings. Maxwell gets insane overclocking scaling.
>unreal engine 4
>not being cucked to the max
pick one
Will Pascal be any good, or just another minimal improvement? I know it's a stupid question to ask since you probably don't technically know, but will it be the big leap everyone is hoping for, or just another lil' bump?
>>52057578
No, it's not like ICC. Gameworks adds additional, unnecessary GPU-intensive operations (generally insane tesselation) that nvidia cards can perform faster than AMD cards. If this was not done, nvidia cards would get better performance as well. Similarly there's no benefit to nvidia in not using asynch compute, in fact with decent drivers nvidia would probably be able to get more performance out of their cards if they used it compared to their own cards if they didn't.
>>52057581
Enjoy your gimped performance just for the sake of trying to gimp the competition slightly harder.
>>52057632
Likely the biggest jump we'll see in years. It's going to excel at everything.
>>52057632
We don't know much about Pascals architecture other than it's bringing back FP64 and will introduce native FP16 to GPUs, neither of which is relevant to gaming in any way.
Wait and see senpai.
>>52057619
Doesn't like 90% of pc games use unreal
>>52057632
>>52057654
It'll be a die shrink, which will probably mean much wider and thus faster chips, but 16FF is fucking expensive so don't expect any increase in perf/$. Also there's HBM2 but really all that does is enable the scaling from moar cores without horrific memory bottlenecking at this point.
>>52057654
Whats FP ?
>>52056845
No, I didnt's buy nodrivers.
>>52057688
More like 5% if you're into new games
>>52057695
Yeah the die shrink and HBM alone will give a massive boost, I just meant we don't know much about the architecture itself.
>>52057696
The size of the numbers the GPU can process. Games use FP32 (32-bit) for everything, scientific computing benefits from FP64 (aka double precision or 64-bit), and some new emerging fields like artificial intelligence can leverage FP16 (half precision).
>>52057696
Floating-point. It's the type of data GPUs operate on - FP16, FP32 and FP64 refer to the number of bits of precision used. FP16 is shit but nvidia are pushing it because their designs can do 2 FP16 ops in place of one FP32 op, allowing marketing to claim bigger numbers. FP64 is only really used for serious compute stuff, I guess maybe some physics in games.
>>52057710
Isn't that because UE3 is ancient and UE4 hasn't been stable for long enough for a big game to be developed on it yet?
>>52057557
Bribed? lel
Fuckibg amdfags are angry that some companies worj together and call it corrupt now?
none of this matters and you're all faggots
everyone but poor slavs and HUEHUEHUEs will be playing DX12 games on arctic islands or pascal
>>52057782
Actual programmers hate gameworks. This has nothing to do with people working to *improve* performance anywhere.
>>52057744
>>52057654
FP16/32/64 is the number of decimal place right?
Lower the decimal place for float calculation cuts down the processing power and makes it easy. Those are useful for cheap approximate calculations. You don't need super accurate calculations for most game stuff. However with nvidia pushing FP16, they can get a bigger share of the benefit than AMD. I know that nvidia cards can do faster single calculations, which helps this. AMD does better multi calculation but weaker slightly single.
Its a design choice difference.
AAAAAYO YO SO YO-
*installs one year old driver*
SO YO ASS TELLING ME NVIDI-
*mouse cursor corrupts*
AITE AITE SO YO MOTAFUCL-
*buys 1700w psu*
YO PHAT ASS BETTEE NOT BE TEL-
*cranks open window*
THAT NVIDIA WIT-
*orders fire extinguisher with prime shipping*
AIGHT AIGHT SO THAT DRIVE-
*turns gpu fan to 100%"
DEZ MOFUCK NVID-
*waits for ccc to install*
U BE TEELING ME DA GTX
*turns down tessellation*
YO AIGHT AIGHT THAT GEFORCE
*shitposts about gameworks*
HOL' UP
*buys aftermarket cooler*
YOU FINNA SAY
*manually disables visual effects*
YO BITCH ASS TRYNA
*buys another case fan*
NVIDIA GONNA
*buys a UPS*
FINNA ADD
*crosses fingers that the game doesn't crash*
SOME BITCH ASS
*googles best driver version*
GRAPHICAL FEATURES???????
*shitposts about hairworks*
>>52057826
It's not as simple as number of decimal places. For one, a single bit is much less information than a single decimal character. See: https://en.wikipedia.org/wiki/Half-precision_floating-point_format
I don't think FP16 is likely to see much/any use in games. The only use is 'deep learning', a minor but cool-sounding field of AI research that nvidia are hyping to make their ideas look relevant.
>>52057942
>I don't think FP16 is likely to see much/any use in games.
Surely FP16 is enough to sufficient for working with colors and normal vectors senpai?
>>52057851
>>>/b/
>>52058003
The cheapest displays use 18-bit colour, and 'true colour' is 24-bit. To avoid rounding errors it's sensible to work above that precision.
>>52057495
Fuck off nigger a 3KB difference is nothing
>>52058059
No it's not. 0KB would be nothing. Are you retarded?
>>52056845
>1080p test on 4k cards
what kind of site would do this? it serves no realistic purpose, only to skew results.
>>52056845
>Both past 70fps
>implying there's any discernible difference
Fanboys everyone.
>>52058073
Since I don't live in a third world country the amount of time for me to download 3KB is too small for me to notice
>>52058055
24-bit in the context of monitors is bits per pixel, so it's 8 bits per channel.
FP16/32/64 refer to the size of each channel, so FP16 RGB is 48 bits per pixel.
>>52058098
Your logic is literally "nothing to hide, nothing to fear"
>>52058104
I don't know what's going on but it looks like you just rekt an amdpoorfag on Christmas and I like it
>>52058081
More to the point it's a short demo from a game a long way from release. No sane driver team will waste time optimising for it yet. The original builds that weren't optimised for anyone favoured AMD.
>>52058104
You realise all processing of colour in that form is done by ROPs that are already capable of and optimised for an appropriate level of precision, and has nothing to do with FP16 in stream processors?
>>52057578
Except when they actively harmed non intel CPUs
>>52058219
But what about all the intermediate variables used in the pixel shaders before the final color gets pushed to the ROPs senpai?
>>52058343
Pixel shaders already use appropriately low precision and have since forever. Do you think hardware designers are idiots?
>>52057306
>Also old 7970 (280x) beats 960 by more than 30%.
bullshit that sounds too good to be true
i imagine a used 7970 or 280x are way cheaper than 960
>>52059751
Got mine for ~$110
>>52056845
Show the 4K results.
>>52057219
>both cards are a massive waste of money
this
anyone buying anything higher than a 390 right now with hbm2 around the corner is a retard. if you have to have a card now get a 390 at most or wait and see who wins the hbm2 war
>>52057605
>this pleb doesnt know NShilldia paid $50 mil to Ashes of Singularity devs to shill for them and fuck up async
>remember all those links with real benchmarks where nvidia got rekt by 50% perf? Yeah, those, they are all gone because Nvidia paid for their deletion
>>52059872
They are barely getting above 60FPS at 1920x1080 and you want 4K?
>>52059936
This, nvidia is the illumanatty
>>52056845
>I fell for the SSD meme too
I literally bought a shit ton of empty space...
>>52057790
I'm Brazilian and i had to sell my kidney to buy the GTX 970. I don't have any other body part to sell, life is not fair.
>>52057306
Nvidia's async requires driver commands and Nvidia hasn't released any gameready drivers for Fable Legends.
muh gaming!
>>52059936
>>this pleb doesnt know NShilldia paid $50 mil to Ashes of Singularity devs to shill for them and fuck up async
oxide is sponsored by amd, so the very idea that this was even considered is very dubious.
>>52057371
>why is this video card from early 2012 so far down the list?
indeed.
>>52060207
The game engine that does the graphics and the rendering work for Fable is Unreal Engine4. UE4 and EPIC have a very close partnership with Nvidia so they are already Nvidia driver optimized. UE4 specifically avoided async compute as long as they could and didn't support it.
