When's the magical driver update that fixes everything supposed

Maybe six months, maybe a year, maybe when Navi launches. Maybe when I fucking feel like posting the drivers on github just to fuck with people.

File: oiK8Bb9.jpg (906KB, 793x877px) Image search: [Google]

906KB, 793x877px

>>62471652
Whenever 17.320 hits the public drivers.

>>62471652
RTG is one meme shy of being erased from the timline... They better deliver on what they promised and I'm not talking about AMD proper releasing a proper APU. I'm speaking about their full fledged GPUs and their inability time and time again to publish functional drivers and dev tools yet promising the world above their competition.

>>62471652
I'm still waiting for nvidia async drivers -=\

>>62471821
There is no such thing, but I understand what you're meme is
Still irrelevant bud

>>62471821
I want to see a pygmy fuck a hippo too, but some things just aren't possible.

>>62471821
Nvidia use the brute force

>>62471821
Nvidia had been using Async since the GTX 480 dude.
480 and 580 had partial hardware schedulers
680-1080 have software based Async.

they're waiting for 1070ti

File: async vs pre-emption.webm (2MB, 960x540px) Image search: [Google]

2MB, 960x540px

>>62472393
>680-1080 have software based Async.
wrong

>>62472393
>Nvidia had been using Async since the GTX 480 dude.
No they haven't. It's not even a level of control exposed for DX11.

>>62471821
They don't need it. The reason why AMD sees "large" improvements with async compute (when done right and for a sufficient workload anyway) is because AMD is bad at distributing work to the shaders. In AMD's case the reduced latency between tasks means shaders get more work. Let's say shaders are working on some shit, and with async compute they get issued more work quickly because the memory ops and graphics ops were scheduled and worked on quicker. More work is issued to shaders while they're still working, in theory it sounds trivial at best, bad at worst because you assume shaders are given work in such a way that you get 100% resource efficiency. But in AMD's case the shaders are normally underfed, which means getting more work to do is a good thing. Of course, things don't always work out perfectly and sometimes async compute doesn't do much, and sometimes it makes a big difference. Of course, the biggest difference can be made with fixed hardware, and ultra-low level control, and predictability (console games, like The Tomorrow Children).

AMD's shaders right now are like the genius expert who's getting paid 6-figures to sit on his ass all day because everyone else is too slow at their own work and bad at giving him something to do.

>>62471821
they added dx12 support to kepler, two years later

File: its_got_a_soul.jpg (193KB, 1020x1020px) Image search: [Google]

193KB, 1020x1020px

>>62471652
9/23/17

>>62471725
lol, this fucking shill again
same picture name in every single thread

not going to dismiss this dispite these shill insisting we give up on vega.

I know its shit and always going to be based on power draw a lone.

I'm ignoring shills that talked about 1080 ti or vega together at anytime.
because only a idoit would think 8GB GPU can compete with 11GB gpu.

So what I think is that Vega 56 is hugely successful regardess what /g/ says.

but Vega 64 is a flop the power draw and the lack of refinement and Improvement isn't there we don't care if it doesn't beat the 1080ti but its not close and at the target where the tech communtiy expected.

But because this is a New Uarch i'll give it untill end of December to fix it.

Its good though that at least vega 56 did compete because now we are forcing nvidia to release the 1070 ti.

I hope Navi recovers what vega 64 lost.

t.GTX1070 owner.

>>62471821
You're a fucking retard. How many times does the same thing need to be explained? AMD sucks at DX11. Nvidia not gaining a performance percentage in DX12 doesn't mean NVIDIA performance in DX12 is bad.

File: warp_pascal_volta.png (78KB, 1639x682px) Image search: [Google]

78KB, 1639x682px

https://devblogs.nvidia.com/parallelforall/inside-volta/

>Volta GV100 is the first GPU to support independent thread scheduling, which enables finer-grain synchronization and cooperation between parallel threads in a program.

THANK YOU BASED NVIDIA

>>62474596
can anyone explain this to me like I'm 12

>>62474630
No deadlocks.

File: 1422933963072.jpg (63KB, 874x451px) Image search: [Google]

63KB, 874x451px

>>62474665
>implying

File: 1489123385180.png (80KB, 600x413px) Image search: [Google]

80KB, 600x413px

>>62474596
What is this image trying to convey

>>62471652
Drivers which fix everything will arrive with NAVI

>>62474683
No deadlocks.
Threads in one warp can have independent scheduling.
>>62474032
>posting the official whitepaper is shilling
?

>>62474766
>Threads in one warp can have independent scheduling.
isn't it whole point of GCN?

>>62474766
does AMD really not have something comparable?

>>62474785
No, GCN allows interleaving of graphical and compute slots in one wavefront without GPU shitting itself causing fuckton of cache flushes.
>>62474790
Not right now.
But for them the change is trivial.

>>62474816
so, is it same deal as with AMD side? have to specifically optimize for such scheduler?

>>62474827
Yes, it's CUDA9 thing and it's irrelevant for gayms.

>>62471652
It's what, October for AiB cards?
That's when you'll have your drivers.

>>62474936
I want it to be good so bad, and I don't even own a GPU made in the last couple of years. Nvidia needs to be taken down a notch.

>>62474949
>Nvidia needs to be taken down a notch.
RV770 failed to do so.
Evergreen family failed to do so.
Tahiti failed to do so.
Hawaii failed to do so.
It's a mindshare-driven market and Radeon lost it during R600 times.

>>62474962
I suppose I know Nvidia will never be taken down a notch, but people in the know knew AMD has been the better choice at times. I just want Nvidia to at least stop their milking and jewing to an extent.

>>62475227
>I suppose I know Nvidia will never be taken down a notch
No, it will, but not in the consumer market.

>>62471652

I'm an AMD fanboy, but I don't get Vega nor the "poor volta" marketing shit.

I mean, wtf, nvidia just responded with numbers, one year in advance!

I don't think there will be any "magical driver" for any not yet active features nobody cares about.

and what's worse is that this shit draws a fuckload of watts, seriously this is a fucking joke. It's like the bulldozer of gpus.

and I've always bought AMD and ATI, this is just disappointing.

and not to mention the mentally insane pricing policy, oh boy...

>>62475291
>features nobody cares about.
>no one cares about generalization of non-compute shader
t. never read documentation for any relevant API

File: kekekek.jpg (7KB, 184x184px) Image search: [Google]

7KB, 184x184px

>>62471652
AHAHAHAHAHAHAHAHA

>>62475319

>it is so relevant they didn't enabled it

so they are just monkeys throwing shit at themselves, makes sense since pajeet raja and his pajeet team.

what was I expecting from someone from a country like India, they bath in their own shit, baka

File: 1310483412100.jpg (36KB, 413x395px) Image search: [Google]

36KB, 413x395px

>>62471652
AMD fags BTFO

>>62475371
t. retard

>>62474398
>because only a idoit would think 8GB GPU can compete with 11GB gpu.
Only an idiot would think video memory is any way related to raw GPU performance

File: 1309858401577.png (9KB, 142x99px) Image search: [Google]

9KB, 142x99px

>>62471652
Well they fired rajesh poonjab so who knows when now

>>62475590
They did what?

>>62475620
keep up with news my friend

>>62475648
There's nothing about AMD doing any kind of layoffs.

File: 128mb card.jpg (184KB, 1878x954px) Image search: [Google]

184KB, 1878x954px

>>62474398
"Hey! look! 128MB of ram! that's like TWICE what you find on a Geforce 4ti! i bet this card is a monster!"

>>62475668
tfw fell for the r9 390x 8gb vram meme

>>62474032
>Suggesting people discussing Vega on /g/ should actually read the white paper for the product they are mindlessly shitposting about is shilling
The absolute STATE of neo-/g/

>>62475668
>NV30
How the fuck did they manage to make FP32 shaders THIS slow?

>>62475678
still a decent card so dont understand how it would be a meme

>>62475678
At least games are very playable on it.
FX5200 was made out of pure disappointment.
Was like getting a Geforce4 that can boot more games but run em worse, because Nvidia crippled the T&L capabilities due the 4MX being fast enough in T&L to do CAD as well as the expensive quadro cards.

>>62471652
never, even Raja gave up.

>>62475678
390x is mazing my friend, especially since it can get double its value as heater and winter is coming. If your room is small enough, your PC can practically heat it to a cozy temp by istelf.

>>62475694
>>62475697
cos it runs worse than an equivalently costing card, atleast in Australia

also the card grinds to a halt if you try to use anything that approaches 4GB vram or more so what's the point?

>>62475689
Skimming over the internet, it seems that the case is that the Geforce FX only have a 16bit ALU, and when you try to use 32bit, it takes two cycles to perform it.

>>62475711
>Raja gave up
Dunno "primitive shaders" are strictly his input.

>release a GPU that performs the same as a Fury X clock for clock a year later than the competition
Why do people here shill for ATI/RTG when they can't even release a finished product after a year delay? Vega 64 would be a shoo in for worst product of the year if Intel hadn't released Kaby Lake X. Is it really just muh underdog?

>>62475770
>Is it really just muh underdog?

it's about competition, people don't like being gouged by monopolies

>>62475770
It's only slower than Fiji clock for clock because of 4tris/clock limit imposed by legacy pipeline.
And you know it yet still shitpost.

>>62475770
AMD bet the farm on Zen. That left RTG with minimal resources to develop Vega, and so they don't have enough driver monkeys to get the drivers done in anything resembling a timely fashion.

Vega behaves like an OC'd Fury right now because it's still subject to GCN's 4 triangle per clock front end geometry bottleneck. This is because Vega launched without they key uarch feature designed to overcome this limitation enabled yet. If primitive shaders can deliver even half of the claimed benefits, Vega's geometry performance is going to literally double when prim shaders hit the public drivers.

>>62475682
discuss what? that shill has posted the same pic a hundred times now, anything that there was to discuss has already been discussed

until the super duper mega giga epic performance uplifting drivers come out, I will keep calling vega SHIT

S H I T
H
I
T

>>62475907
>That left RTG with minimal resources to develop Vega
Why does this bogus keep getting parroted

>>62475917
>posting a page from whitepaper is shilling
?

>>62475907
>That left RTG with minimal resources to develop Vega
What is this meme?
The changes Vega brings are too hueg for "minimal resources".

>>62475934
>whitepapers increase gpu performance
?

>>62475921
I dont know. The most problematic thing about VEGA is that they couldn't make seperate lines for compute and graphics cards. Which would have been impossible with any kind of money as a purely graphics card wouldn't sell as well without an Nvidia sticker.

>>62475921
It's obvious how understaffed RTG is for driver monkeys or they wouldn't have launched Vega FE with pre-alpha tier drivers and Vega RX with half finished drivers missing core uarch features.

>>62475959
>The most problematic thing about VEGA is that they couldn't make seperate lines for compute and graphics cards.
They did exactly that.
Vega20 is their HPC offering with 1/2 FP64 and ECC in LDS/GDS/caches.

>>62475962
when have they not been understaffed and underdelivered when it comes to release drivers

>>62475952
They will.
Now steal me 17.320 branch.

>>62475962
>with pre-alpha tier drivers
>he actually believes that a magical driver update will come and obliterate nvidia
lol, see you in 2 years, when vega manages to beat a 1080 by 5% because the gpu is still shit

File: 1298635737876.jpg (24KB, 586x357px) Image search: [Google]

24KB, 586x357px

>>62475962
hopefully someone who can actually program applies to work there

we'll just have to wait

>>62471652
The real question is, when is GPU mining going to be not worth it

File: facepalm.jpg (34KB, 400x400px) Image search: [Google]

34KB, 400x400px

>>62475984

>>62476002
>better performance in internal branches means it will never perform better in public ones
?

>>62476017
do you do it for free?

>>62476023
>reeeeeeeeeeee its shit because i say so! shill!

>>62476040
>draws more power than a fury x
>1080 tier performance

yeah it's definitely not shit lol

>>62475907
>Vega performance improves two years later after everyone has already moved on
>Vega beats 1080 but no one cares except AdoredTV
>AMD fanboys start screaming FineWine in Youtube comments section
>Nvidia continues to dominate in marketshare

>>62476073
>Nvidia continues to dominate in marketshare
That was still the case during Fermi so it's a moot point.

>>62475985
Read the white paper. Core uarch features well described in the white paper are not "magic."

>>62476079
>dude it doesn't matter if AMD fucks up the launch of a really hyped up GPU lmao

>>62476088
I don't remember AMD fucking up the launch of SSG.
Do you?

>>62476103
>moving the goalposts this hard
hahaha fuck, you're adorable
dude, it's ok to admit that vega is shit, I own an AMD cpu + gpu combo

>>62476132
>really hyped up GPU
Did anyone hype up RX Vega?
Stop moving the goalposts.

>>62475779
I don't see why people should support a company with an incompetent driver team just because of competition. If anything this just proves that we need a third GPU company, but of course this will never happen due to the patent nightmare.

>>62474398
>>62475291
>reddit spacing

>>62476143
>Did anyone hype up RX Vega?
umm yes, AMD themselves did
don't tell me you don't remember "poor volta"

you know what, you're a shill, that much is obvious, so you go on and shill some more on /g/, I'll stop wasting my time on you

File: 1301147807686.jpg (25KB, 175x154px) Image search: [Google]

25KB, 175x154px

>>62476143
>Did anyone hype up RX Vega?

>>62476160
Support is one thing, but hoping a company does well is another

>>62476143
Yes actually, /g/ has been telling people to wait for RX Vega for over a year. Now they're already saying wait for Navi.

>>62476073
>Two years
The current public driver is based on the 17.300 branch. Primitive drivers are working in the 17.320 branch. That's not 2 years away. More like 2 months tops.

>>62476166
>don't tell me you don't remember "poor volta"
How is this in any way related to RX Vega my child?

>>62476181
Current public driver is 17.20 at best.
DSBR is *supposedly* working.

>>62476216
You wouldn't be able to see any benefits from DSBR until after the geometry bottleneck is fixed anyway.

>>62476143
>Did anyone hype up RX Vega?
are you for fucking real, dude?

>>62476236
Not completely true.

>>62476143
>Did anyone hype up RX Vega?

This is the face of the AMDrone.

>>62476236
I dunno you can load something shader heavy and hammer the GPU for a few minutes.
Too bad no one bothered with it just like no one besides some germans ever bothered with B3D suite.
Modern """"""""""""""journalism"""""""""""""" my ass.

Vega is shit. Efficiency is not bad but Rajaeet had to clock it up to 1500-1600Mhz in order to compete. For Mobile offerings vega graphics and their full support of the DX12 feature set is certainly nice. I just hope they can get die size under control.

>>62476308
>Efficiency is not bad
"Efficiency" meme is straight up relative to ALU count.
>For Mobile offerings vega graphics and their full support of the DX12 feature set is certainly nice.
What the fuck does this even mean?
>I just hope they can get die size under control.
Vega10 is only 486mm^2 for 4k ALU design.

>>62471652
Tomorrow 2PM

Making hardware and wait for software to catch up. this why amd keep failing hard.

>>62476384
That's ATi (or ArtX) for you.

>>62476404
it same with bulldozer, hopeing for dev to adapt 8 core back then is dumb move

>>62476473
Vega is exactly opposite of that.
It gives devs nothing.
Everything it implements is absent from modern APIs.
Heck "primitive shaders" are the biggest "FUCK YOU" possible to both Khronos and MS.

>>62476329
>"Efficiency" meme is straight up relative to ALU count.
I am missing something here. Isn't 3d rendering FPU intensive? Or because of the high alu count vega is inefficient at 3d worloads? Explain.

DX12 contains many "Feature levels" with the most basic being 12.0; I'm sure there are going to be more but right now 12.1 is the most recent one with among others support for asyncronous compute at the hardware level (Nvidia implemented it in software which will become a problem for them) and rasterising.

All it means is that the chances of needing to fall back to DX9 or 11 is slim with vega

>>62476710
Vega has too much ALUs for GCN's legacy geometry pipeline to handle efficiently.
It's a front-end bottleneck.
>
DX12 contains many "Feature levels" with the most basic being 12.0; I'm sure there are going to be more but right now 12.1 is the most recent one with among others support for asyncronous compute at the hardware level (Nvidia implemented it in software which will become a problem for them) and rasterising.
>All it means is that the chances of needing to fall back to DX9 or 11 is slim with vega
Two words: Vulkan exists.

>>62476501
No it's not. How the fuck does it make sense to give devs complete control over that stuff when AMD hasn't even figured out how to implement it themselves? Knowing AMD they'll probably open it up eventually. Way too fucking early to do that now. Besides, it's not THAT much of a big deal if they never do.

>>62476833
That's what i said.
The spec for whatever they call a "primitive shader" is not ready.
They are not giving anything to the devs.

>>62472869
>>62472869
>this
So 15th of October when leather jacket man announces it.

>>62477260
GTC Europe is mostly about memes but maybe he'll announce it.

>>62477260
why do people think there's any sort of substance to the rumours of that card being real

>>62477385
It's the most obvious product they can launch without it sounding like damage control.

>>62477385
https://www.techpowerup.com/237016/nvidia-readying-a-geforce-gtx-1070-refresh-gtx-1070-ti

>We've spoken with well placed sources.

>>62477401
It would fill such a tiny goddamn niche, performance- and price-wise. They better make 1070 cheaper.

Nvidia will probably do it because they fucking can. They like to rub AMD's nose in it. plus they don't sit on thier R&D like Intel does.

>>62477434
Is that why Volta isn't coming out anytime soon?

>>62477424
It's still something and it does not sound like absolutely rampant damage control (hello Intel, never go full retard).
>>62477434
Companies don't launch products because they can. Especially when it comes down to refreshes or different die configs (hello GTX260 216).

>>62475994
China is already cracking down on it. They don't like their worker ants getting rich away from the yuan. It completely destabilizes their economy.

>>62474949

Even if AMD delivered GPUs made of gold with coupons for a free blowjob from your local roastie Nvidia would still outsell. The only way AMD can get mindshare back is if Nvidia flops an architecture and AMD nails it, for two years in a row.

>>62480460
>The only way AMD can get mindshare back is if Nvidia flops an architecture and AMD nails it, for two years in a row.
That happened already and nVidia STILL maintained 60% marketshare.

>https://www.reddit.com/r/Amd/comments/70nejc/what_happened_to_the_amd_radeon_vanguard_program/
Oh boy, NDAs.
>inb4 lebbit

>>62475732
>Australia
>being relevant, at all
>ever

>>62473699
what kind of crpytic /pol/ post is this.
Just gonna toss a date out their? Your guess is as good as mine.

>>62480662
Hey, does that mean 17.320 might be out sometime soon? That would be nice.

>>62482546
We don't know.
Everyone who *might* be testing the batch is under the fucking NDA.

>>62480662
That's pretty shitty. It's not like it's some super secret new space-age architecture. It's fucking drivers for cards people already own. What a bunch of faggotry

Must not be going well

>>62482633
>It's not like it's some super secret new space-age architecture
It is.
They literally killed traditional fucking shader stages we've all got used to in favout of custom ones.
That's R300-tier change.

>>62482687
I'll believe it when I see it

>>62482704
I too love being pessimistic about something i can't understand.

>>62482687
How many times is /g/ going to overhype some ATI product or feature only to get disappointed? After the Fury X I'm never getting excited about a GPU until I see benchmarks.

>>62482749
LOL OK, BUD. AYYYMD GONNA CHANGE THE WORLD AMIRITE?

Weren't you just defending their small R&D budget in the last thread?

>>62482772
>How many times is /g/ going to overhype some nVidia product or feature only to get disappointed? After the GTX 480 I'm never getting excited about a GPU until I see benchmarks.
Aka who the fuck cares someone telling both Khronos and MS to fuck off is cool.
>>62482800
Now you're shitposting.
Have a (You).

Never most likely.
Vega's just shit. Accept it and hope Navi won't be, (but it probably will).

>>62482809
Did you spend all of your allowance on AMD stock?

>>62482829
I wish i did back when it was $2 per share.
>>62482828
(You)

can buy a new Vega 64 reference version for less than Vega 56 and lower than the RRP. should I?
liquid cooling on my 295x2 just died and I need a replacement until I get that sorted.

>>62483894
It's a reference card still, gonna be loud.

>>62483948
295x2 is loud as well, it doesn't bother me much

>>62484083
I think you can get one.
It's pretty meh right now but it might get better.
RX Vega was only launched to please shareholders, that's why you see no AiB boards.

>>62474630
GPUs are SIMD machines, single instruction, multiple data. That means if you have 32 threads doing the same work, then you can do all 32 at the same time (32 threads with their own data, one instruction "applies" to all of them, like 32 adding operations). But that's a perfect scenario, in reality most of the work a GPU does will change and go down different paths. For example, 4 out of the 32 operations trigger some if statement and do multiplications while the other 28 keep adding. That means they're no longer all working at the same time. Now you have two different steps, one step 28/32 threads get to do work on adding, next step, 4/32 threads get to do work on multiplication, the rest of the threads just wait while the others do work.

It doesn't seem bad but it can become a serious problem with multiple threads diverging, so you can have like 12 steps, sometimes of only 1/32 thread working during a step, meaning 31/32 threads idle for a step.

What independent scheduling does is try to sync up the threads as much as possible. So if you have 12 different steps independent thread scheduling will try to find steps that are working on the same instruction and group them together, or they'll temporarily pause or skip steps until other steps start working on the same instruction. To put it simply, it will try to turn the 12 steps back into 1 step so all threads work in 1 lockstep and there's no waste.

>>62475743
It's not just that, they went for full 32bit precision as a minimum, while Ati used the partial, 24bit version that was allowed by for DX9. This allowed them to be more efficient with the silicon budget. IIRC they also had a better process node too, allowing for higher clocks and way less heat.

>>62484376
Actually R300 used inferior node versus nv30.

>>62476710
>Isn't 3d rendering FPU intensive? Or because of the high alu count vega is inefficient at 3d worloads?

GCN is a compute beast, but it is not good enough at rasterizing, leaving many shaders idle. They introduced Mantle/Vulkan/DX12 for this very reason.

They attempted to improve the rasterizer in Vega, by adding tile based binning, basically the card determines which triangles are overdrawn, and skips drawing those as they don't show up. This is done by cutting up the screen to tiles, because those tiles fit in L2 cache, so it can be done extremely fast.

PowerVR chips in phones have been doing this for a decade now, because it is a very efficient solution, and Microsoft was experimenting with it back in the 90s too but the brute force attempt of the 3dfx cards was more popular at the time. The Nvidia 9xx cards introduced this and it is why they got such a boost over the 7xx ones.

Problem is that this is a complex task that can break shitload of existing stuff. So they put a legacy path in Vega, otherwise the card would ship half broken. So basically Vega is Fury X but with a new super efficient rasterizer, that remains disabled until they can get it working.

>>62484487
Eh no. Vega's problem lies not in raster, but in front-end.
Right now it's using legacy geometry pipeline with standard shader definitions we've all been used to.
>PowerVR chips in phones have been doing this for a decade now
ATi also did it in Imageon.

>>62484531
Well the pipeline is more complex than just rasterizing triangles, you have geometry shaders, vertex shaders, rasterization, and pixel shaders (iirc). Obviously if you determine that some task is unneeded as it gets obscured later in the data path, you save that many more resources, not just in rasterizing but in the tasks before it too. But that's a difficult task to do without potentially breaking everything, hence why drivers are taking so long.

I'm not sure if the 9xx has a legacy code path as well, or if they just got drivers 100% right by launch.

>ATi also did it in Imageon.

And the Sega Dreamcast in 1998 also used tiled based deferred rendering (using a PowerVR2 chip).

>>62483894
1080 performance for less than 1070 price? Of fucking course you should buy it.

>>62484597
>you have geometry shaders, vertex shaders, rasterization, and pixel shaders (iirc)
You forgot hull and domain shaders.
Vega kills every one of them (besides Pixel, that one is still decent) in favour of something novel; it combines everything but Pixel shader into two stages, primitive shaders for all things geometry, and it's subset - surface shader, for tesselation tasks.
Think of it as a big "FUCK YOU" towards MS&Khronos straight from the Raja himself.
>I'm not sure if the 9xx has a legacy code path as well, or if they just got drivers 100% right by launch.
It has a fallback mode for the cases where tiling is useless and only creates additional overhead.

File: 1499234569267.png (2MB, 1822x1224px) Image search: [Google]

2MB, 1822x1224px

>>62482516
You'll know why soon. remember the pic and look for the signs.