Vulkan Thread
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
NVIDIA BTFO
#AMDefecatedOnNvidia
Is it over for Nvidia? They've been having a good run the past 5 years but their lack of proper async may do some real damage.
You do realize that the next generation of Nvidia cards are going to improve their performance that will be comparable to AMD.
>>55571836
In 2018 yeah, like 2 years from now, that gives AMD plenty of time. Look at the graph, a $250 card is getting 75% performance of a $650 card. They should put that Price/Performance ratio on one of those retarded Nvidia graphs.
Its absolutely ridiculous.
They need to put in the 1070 and 390 on that graph to make the Nvidia fags weep more.
>>55571809
But it's only one game, Nvidia will pay other game developers not to develop Vulkan
Vulkan is the new meme, I hope it goes mainstream and steam makes it a requirement for every game under steam ahaha
>>55571836
By that time AMD will create a newfangled gpu that will solve all of the world's problems while nVidia finally gets the hang of some ancient, deprecated tech called async compute.
Granted D44M isn't the best looking game ever, and its not a huge open world title, but 120fps+ from a $200~ GPU is pretty impressive.
>>55572058
nvidia's last gen had async compute
they already deprecated it with pascal
Driver overhead was holding back AMD all along. finally they came out of that vicious years and leveled the playground.
this is all 4000 series again. AMD is hitting hard.
>>55571809
>Doom
Watch what happens when they turn async compute on for Nvidia
Your gains will be destroyed
>>55572075
AMD hitting hard with what is essentially a GPU designed to be as cheap as humanly possible. Polaris 10 is super fascinating.
>>55571809
>little gain because nvidia already had good opengl drivers
>meanwhile amd's opengl drivers were total ass
Kek
Are you people aware the difference between the reference 980 and the reference 980Ti is at most 15 to 20 FPS?
The 480 is still behind the 980Ti by quite a big margin and this is why a low level API, it should be higher.
You people see big numbers but lack any knowledge to compare.
/g/ gets dumber by the day.
>>55572084
>turn async compute on for Nvidia
no such thing exist.
>>55572099
https://www.youtube.com/watch?v=bL5bew0lwNg
Searched for some quick numbers to compare, in average the 980Ti is fater than the 980 by 15 to 20FPS at DX11.
>>55572110
Nvidia cards can do some async compute, just 32 operations at a time instead of 128 like AMD
Some fags at guru3d wrote an application to test this
>>55572075
by driver overhead do you mean nvidia proprietary deals with developers?
>>55572121
No, he means how AMD's drivers suck ass at OpenGL, which is something that's been common knowledge for over a decade
Also
>opengl
>proprietary deals
>>55571809
Here's a novel idea--nVidia actually has decent driver support.
It's funny how I'll go down the list of grievances with my last AMD card, and everytime, I hear.....
>It's been fixed!
It shouldn't have been broken in the first place, nor do they even mention of HOW LONG it took to be fixed.
I had mouse and flash corruption for almost a year before I gladly paid double for an nVidia.
Source: https://youtu.be/lXijkS_qIzA?t=819
I know you /g/oys think hes a shill but make sure you watch till the end (or start at the time linked in the url), he actually criticizes AMD for having such lackluster opengl performance.
Seems that the [spoiler]true potential[/spoiler] of AMD cards is being unlocked here, AMD was always been giving more for the money because their drivers are such shit. Trying to make up for bad software in overkill hardware.
>>55572093
I really wonder about Vega after these results.
>>55572117
source?
>>55572121
I'm talking about opengl drivers.
>>55572071
and yet pascal is just maxwell 2.1
>>55572159
My bad, it was beyond3d not guru3d
https://forum.beyond3d.com/posts/1868421/
>>55572151
yes we are aware, read the posts.
>>55572075
I'm still hearing about this driver overhead though--literally yesterday people were telling me not to match the new RX 480 with my APU build for that exact reason.
I guess that's why they're pushing DX12 and Mantle (is that still a thing?) and all, but I've long since stopped relying on hollow promises from AMD.
>mfw dx12
>>55572099
Polaris 10 is 36CU, 32ROP. Its not a big 64ROP high pixel throughput design like Hawaii and Fiji.
It is essentially a Tahiti(Radeon 7970) with 4 more CU and newer arch. Its not even fabbed to be as dense as possible, all the structures on the front end of line are spaced well apart from one another to balance defect rate and die area. Its a tiny die, with a layout as simple as possible.
Being near the GTX 980 with that is impressive no matter how you try to contort things.
>>55572159
Nvidia async computer is faster with smaller batches, but slower with larger batches, while AMD is leveled out
https://forum.beyond3d.com/threads/dx12-performance-discussion-and-analysis-thread.57188/page-9#post-1869058
>>55572184
Mantle was donated to Khronos, holder of the opengl and developed the mantle into Vulkan.
if you going to play dx11 games don't buy AMD.
>>55572202
What does that has to do with anything I've said?
The RX480 has more raw performance than the GTX970 and AMD advertise it as being as strong as a GTX980, 2x480 beat a single 1080, a GTX1080 already beats SLI GTX980 in most cases, the performance expected for the RX480 is at least GTX980 performance.
By previous numbers and benchmarks we can know the performance gap between a GTX980 and a GTX980Ti, and the RX480 running on a low level API barely matches it.
>>55571809
>>55572143
>AMD's drivers suck ass at OpenGL, which is something that's been common knowledge for over a decade
nigga you must be trippin, amd ogl drivers follow the ogl spec perfectly, nfagya allows for obscure shit that """developers""" use, cause they "dont know any better"/moneis making shit not work on other vendors
for instance, novidia allows you to omit the version statement at the beginning of a shader, that shit is _required_ as per the spec
go fuck yourself shill!
>>55572331
Look at the pre-Vulkan benchmarks for Doom and stay mad, faggot
>>55572117
Windows literally kills the driver because it thinks it has frozen if you use i ton an nvidia card
>>55571809
AMD won this round
>>55571809
The 480 will be the best bang for bucks for a few years. Should I pull the trigger ? I have a 270x and plans to get a 1440p screen
>>55572331
did you even see the benchmark? it has a massive driver overhead in opengl.
>>55572008
Nah it's the new hotness it will only be more popular Nvidia is in big trouble
>>55572143
>>55572331
also, novidia allows you to mix integer and floats whenever you want without any sort of casting
that shit is not valid code!
>>55572097
It's Vulkan not OpenGL
pay attention
>>55572373
Good thing nobody's talking about OpenGL, come try and keep up kid.
>>55572373
>its easy to win if you cheat
duh...
>>55572414
Read the fucking posts you're replying to and try again faggot
>>55572355
Lol, also try using DisplayPort to hook up your expensive new VIVE to your brand new 1080, lol.
>>55572402
Yeah.
Vulkan only has a huge apparent gain because their OpenGL drivers ran like shit, so Vulkan allowed them to catch up.
It's not that AMD does Vulkan better, it's that they did OpenGL worse
>>55572402
he is right though.
>>55572414
are you an idiot? you don't understand what you read?
>>55572419
DOOM is Vulkan
OpenGL is old and cold nobody cares about it
>>55572058
This. Nvidia is now learning to stack ram as AMD is now learning to stack gpus
>>55572181
Whatever, serves as a nice summary for others.
I knew it was bad but holy shit those drivers were REALLY bad.
>>55572184
Exactly, amd gpus tend to put more work on the cpu, that's why they work very well with powerful cpus, but not so great with low end ones.
On the other hand, one of the features of dx12/vulkan/asynccompute/whatever, is leveraging cpu work on the gpu. If the gpu has the capability to handle this (like this gcn arch), the gains can be huge
>>55572471
Ironic that to get the most out of an AMD gpu you need an intel cpu.
>>55572008
This. Amd has to actually pay devs to use these api's like hitman and recently the new civilization game. Most devs won't give two shits about it. It's a lot of work for very little performance gain (in the case of hitman).
>yfw all future gaymes come with Vulkan and Nvidia's high priced cards gets btfo by much cheaper AMD cards
>>55572509
Actually it's Microsoft pushing DirectX12, AMD just supports it first and Nvidia drags their feet because they're too busy ripping off dumb kids.
Vulkan was a completely open source project that gained traction because it's much faster than OpenGL which is now deprecated.
So it's more "out with the old, in with the new" and Nvidia is just dropping the ball hard.
>>55572331
Even the biggest amd shill on the planet adored tv was pissed at amd for still having shit driver overhead with the release of polaris.
>>55572508
Let's just hope zen delivers, to actually drive down prices and press intel to do some proper r&d. 5% performance increase between generations is laughable
>>55572549
Nah driver overhead is a meme butthurt Nvidia owners spout, AMD drivers are fine.
>>55572508
>Ironic
Why? A lot of AMDs own builds use Intel. AMD's Project Quantum uses an Intel chip. AMD knows their CPUs can not preform like an Intel.
>>55572084
>what happens when they turn async compute on for Nvidia
… in one or two generations of GPUs.
>>55572586
Already possible even on Maxwell,
faggot
>>55572210
>>55572210
Why can't nvidia just use vulcan?
Is vulcan like an alternative to directx?
>>55572597
YEAH, JUST YOU WAIT FOR THE ASYNC DRIVER PATCH AMD IS FINISHED
>>55572586
Maxwell can do it already
they're saving it for the next Titan
>implying Nvidia wasn't holding async back all along just to release it in a new driver and BTFO AMD after everyone got hyped over the 480
>>55572597
>>55572628
In the future we won't need preemption
...wait
>>55572597
>"slower" with larger batches
Slow enough to get the driver killed is not stable in any way or form
>>55571809
>it still performs worse than Nvidia even pre-vulkan
How is that BTFO again?
>>55572649
>implying I don't poop out gold and don't care about the price of add-ins cards that serve a small purpose
>>55572626
Async is a meme. It barely even matters.
>>55572200
delet this
>>55571809
Epic thread family
>>55572443
no
opengl is easy to use and get started with, it's not going anywhere.
async = stuttering
>>55572761
>I'm a retard.
>>55572761
>this whole post
LMAO opengl is dead.
For a $200 card it's crazy what people have been doing to it. [ONLY FOR A SHORT BURST] 2000Mhz on mem and 1700 on the core.
>>55572161
>1070 is maxwell 7,5
>>55572955
Why only short burst?
Instabity or destruction of PCIe?
>>55571809
Vulkan may be an open technology, but it's based in Mantle, which is optimized for the AMD microarchitecture. So is Vulkan.
You can't expect a technology optimized for AMD GPUs to run good on any NVIDIA card.
>>55572772
Source?
>>55573009
>Instabity- Like all cards do when pushed to their limits
>destruction of PCIe- It never killed a board and the PCIe part was fixed even if it pushed the 6 pin out of spec. You're memes are weak around here cowboy.
>>55573013
But it's ok when AMD does it :^)
literally who cares
if you want modern games to be even halfway playable then buy a 1080
budget cards are for budget games
>>55573085
>Proprietary code premade and given to devs
>Devs have no idea what it actually does and it turns out it fucks over everything that isn't current gen NVidia
vs
>Open standard that NVidia has been a part of making and has been in development for quite some time
Oh yeah, totally the same thing. AMD are evil child eaters.
>>55572955
So what? I can OC my 670gtx to really huge numbers in 3dmark as well, but same OC crashes in any game 2min in to the game.
>>55572570
not in dx11 and opengl. that's why AMD created Mantle.
>>55572622
Nvidia already uses it.
Yes it is an alternative to dx12.
>>55572509
Nvidia were pushing Vulkan even harder than AMD until recently.
>>55573013
>>55573085
AMD has been developing ASync for some time now waiting for the industry to play catch up but they haven't thats why Mantel was made. Mantle was also made available to Nvidia but Nvidia said no thanks and pushed a driver for BF4 when shit went south. Thats nvidia's fuck up.
>>55572509
>Amd has to actually pay devs to use these api's
I guess AMD pays the devs of BF1, Quantum Break, Tomb Raider, Hitman, AoTS, Civilization VI, CS Go is getting a Vulkan driver......All paid shills I guess.
>>55573239
Are you aware Nvidia would have had to rewrite Mantle entirely to make it work on their GPUs?
>>55573418
then what did they do with Vulkan? it is Mantle 2.0 basicly.
>>55572955
2000mhz(8Gbps effective) is the default memory clock for 8GB cards, sherlock.
>>55572200
>AMD has higher frame time
No shit sherlock. Lower FPS means more time is spent on drawing each frame, ergo the frame time is higher. One is direct consequence of the other.
>>55573602
He doesn't know that lmfao.
>>55572547
>Actually it's Microsoft pushing DirectX12, AMD just supports it first
Nah, AMD created Mantle, which was then "adopted" into Vulkan and DX12.
>>55573013
>>55573085
>>55573418
>Gimping games with proprietary blackbox-middleware
>It's the same as using an open API which works better on your cards
I wonder what excuse you will have once DX12 hits with similar results. :^)
>>55573495
You are now aware of the fact that it takes no more than three one-hundredths of a second to fill the all the VRAM.
hol up guys question
Since Im buying a i5 6600 processor
Is the intel hd 530 better than my current gt 630?
Gonna buy rx480 / gtx 1060 videocard 2 months later when I get enough cash, but I want a better processor now.
>>55573720
>This image and blog post
Do they think people are stupid or what.
This is virtually saying something like "You can use the Xbox 360 API on the Wii U without having to rewrite any of it"
>>55573648
Don't expect nvidia fanboys to understand the difference between open source greatness and proprietary closed cancer.
>>55573453
Most of the stuff that can be done on Vulkan has already been available for Nvidia using extensions for years.
This includes AZDO and Multi-Threading which is a big part of Vulkan.
>>55573418
no need to
just look how they re arrange the cores on pascal cards...
then look at gcn 1.0
and then laugh hard at nvidia for actually being amd and doing worse than amd
>>55571809
970>480 DX11
970<480 DX12
Not sure why you keep making these threads when the facts remain the same.
>>55573831
ok you got that, but it can be used on Nvidia cards.
>>55573978
Of course, if you want worse performance than DX11 and having to invest thousands of man hours on it just to get it to match DX11 performance.
>>55573960
same reason why gtx1060 threads pop up all the time.
because it's fun seeing fanboys freak the fuck out.
>>55573960
970 gameworks >480
970 neutral games <480
970 dx12 ???? kek
480 dx12 ∞
>Nvidia
Slowly and painfully dragging games to the future with GameWorks
>AMD
Painfully holding back games with consoles and awful ports.
This is the future you choose.
From now on games will run only at acceptable frame rates on a vendor so pick your poison.
>>55573960
they're pretty much the same hence why even the gpu vendors are saying to not upgrade to a 480 if you have a last gen upper midrange card. only retard shills will tell you otherwise.
I am not trying to start flame war but the 480 was a huge letdown, AMD got us all hyped up and then shit the bed with another budget card, we were expecting something great, but we basically got a card that is almost as good a 2 year mid level Nvidia card for $50 less. I guess it's good to have a budget card but I think we all expected more.
>>55574054
it was announced as a budget card. you weren't paying attention.
>>55574135
It was advertised as budget card with top tier graphics, not a budget card with budget graphics, this thing was suppose to compete with the GTX 980 and have to struggle to keep up with the GTX 970, also once you add a cooling system equivalent to the 970 you don't even save any money you just have an expensive budget card.
>>55574204
>and NOT have to struggle to keep up with the GTX 970
>>55574204
what you mean struggle? on most cases that gameworks isnt involved the card is almost near 980...
and is ahead of 980 on dx12 i fail to see how its a let down...
>>55574054
It's the definitive card for max settings 1080p, if you expected more you're stupid. The 3.5GB meme card isn't valid competition since it shits the bed if games like GTA V push it into meme VRAM territory (which it does on max settings 1080p) and can't handle DX12 or Vulkan for shit. 1060 has the latter problem as well so the RX 480 stands unopposed unless you're going for high resolutions or refresh rates most people won't bother with.
>>55574049
>Tessellated oceans and hair is the future guys
>Also those chips that support async compute, parts of the HSA spec and true unified memory? Yeah, they're holding us back
Hang on wait what?
>>55574263
Why is everyone gayming at 1080p when over 1440p, the monitors have come way down in price and I assume all the cards will do it, so why don't people take advantage of the higher resolutions?
>>55574333
you forget tessellated batman capes
but i suspect since amd has its own "conservative raster"function now they wont really use it anymore lol
>>55571809
NV is still faster kek.
AMDesignated drivers are shit. That's all you're seeing.
>>55574436
Why are their so many 1070 boxes when the 1070 is not in stock?
>>55574372
I am a poorfag with a GTX 970 slapped inside of a 10 year old core2quad/xeon machine and I run 1440p just fine, not sure why richfags settle 1080p
>>55572291
>what is multi gpu scaling?
>>55574550
Most people choose higher frame-rates instead of picture quality so they can have an advantage over the next guy that sees the same image a few milliseconds later.
>>55572311
Kek'd
>>55572672
It costs 200 bucks
>>55574701
That's autistic as fuck, especially considering how fucking ugly 144Hz monitors are. When will they manage to make one that doesn't make colors look like total shit? Even 144Hz IPS is a disappointment.
>>55574769
Does it really?
amazon.com/dp/B01HOF7ZF6
>>55574799
Not sure if you are trolling or just stupid.
>>55574499
Depends on which shit country you're in. 1070 is available in the US, 480 is available in Europe.
>>55574701
Why do you think everyone is sticking with 1080p?
>>55574828
Are you really?
>>55574856
This was meant for >>55574778
>>55573239
If you actually watch the video OP is from Async barely has anything to do with these gains.
>>55574860
Try looking up the correct card
>>55574888
every single 480 is instantly sold out in the US.
>>55574856
>>55574871
Why bother with more? 1080p is enough for a 24" monitor at any practical distance and most people don't really need more than that. Just because you can have a higher resolution if you're willing to burn money doesn't mean you should.
>>55572331
https://richg42.blogspot.com/2014/05/the-truth-on-opengl-driver-quality.html
>>55572143
>Also
>>opengl
>>proprietary deals
I was a member of the [redacted] emulator project for about four years. There was a guy from NVIDIA that would suggest a new NVIDIA-specific tweak every week. We ended with a bunch ofif (isNVIDIA()) { /* foo */ } else { /* shit */ }
void foo();
void fooNVIDIA();
class FooNVIDIA: public Foo { /*... */ }
>>55574913
Even a 27" 2k korean monitor makes my acer 1080 ips panel look like shit. Probably the best $200 I've spent on tech.
>>55572117
>software "emulation"
>better than hardware support
kek
>>55574986
are those korean monitors really that good?
can you link some?
>>55574947
I'm guessing that's dolphin.
PCSX2 also has a giant nvidia cocksucker but i don't think he works for them.
>>55572151
that's cute, but he forgot to do any real testing himself or e would have noticed that there's no fucking cpu overhead for amd cards with opengl.
>>55574913
At 1080p I am only using 40% to 50% of a GTX 970 to get to 70fps but at 1440p I can utilize 80% to 90% of the card to get to 70fps, how is this wasting money?
>>55575010
They look great imo. Just search Ebay for dp2710, one I have and currently the cheapest.
>>55575041
You bought a more powerful GPU than you needed. This GPU cost more money than the GPU with the right amount of power for your needs. You could have spent this money elsewhere but instead you spent it on an overkill GPU. That's wasting money. I'm glad we solved this puzzle together.
>>55575095
Keep typing that to yourself in 1080p and maybe someday you will believe it.
>>55574986
>>55575010
>>55575065
My nigga.
27" Yamakasi reporting
The power brick and input scare me (brick gets hot, and the port is fucking weird) but it's been going stronk for almost a year, now. No dead pixels or the like. Only problem is no display port, so I have to use an active DVI adapter if I want to use it with some cards
>>55574799
>amazon
>>55572053
Yes, that would be great. Windows needs to fuck off.
>>55571809
All this proves is that AMD literally cannot into drivers.
Look at the fucking CPU overhead on opengl/DX11.
Now with DX12/Vulkan where the driver implementation is on the developers side, finaly they have their cpu overhead under control
I am just glad that abortion OpenGL got replaced by Vulkan. What a fucking piece of shit of an API.
>>55576932
>kek driver implementation on dev side
that is some next level stupidity
Nobody has yet to answer this question.
Why was both Nvidia and AMD running OpenGL 4.5 in the Doom beta and then AMD's gets deprecated to 4.3 on the games release and is STILL 4.3 even with the Vulkan update?
I don't get the whole 'it's AMD's shitty driver' thing. If it was working fine in the beta why change it???
>>55578682
Both are OpenGL 4.3 actually, using a couple 4.5 features, apparently one was enough to get it marked as "4.5" for Nvidia but "4.3" for AMD even though it wasn't a huge difference
AMD does have shitty OpenGL which is the reason for the worse performance tho
