More info about the RX Vega launch

I can wait

>>62079630
Production of gddr6 won't even start until 2018. It'll take probably half a year to see it in products. I really doubt we'll see volta q1

>>62079630
>well-informed sources of DigiTimes
ha. ha.
why is tech journalism such a joke? I guess what can I expect from people that write about corporate products, still...

>>62079630
The most annoying part is Fiji and Vega didn't even need to exist, they were both shit parts filling in the gap until Navi.

Ellesmere XTL (Big Polaris) existed, with 2816sp/176 tmu/64 rop/384bit GDDR5
But AMD shitcanned it for Fiji because 'muh HBM'

RX Vega sucks, give me my RX 590.

here is a question, I can see several vegas through google, price is retarded, but theya re there
why is everyone complaining about supply then?
shouldn't we complain about retailers jacking up prices for no real reason? What is even point keeping price that high if nobody buys the cards? It's like they will lose money, they got them dirt cheap under MSRP, probably for ~450.

>>62079898
vega 56 is rx590

>>62079914
No, it isn't.
RX590 is Ellesmere XTL, Which was Big Polaris, not Vega.

The only device that will ever get it is Xbox One X...

>>62079630
How can someone post these trash claims when fucking AIB partner cards are scheduled to hard launch in September? Like, they expect their audience to just forget that that supplies of reference cards are going to be irrelevant once AIB partner cards hit?

>>62079630
>they were named Vega 56 and 64 because that's how many they produced.

>>62079936
what I mean vega 56 costs like possible 590, and probably performs better, why complain? big polaris would have exact same problems, if not worse if launched right now

This will be amd fine wine. Expect these to be more powerful than the novideo gtx 1080 in a few months.

>>62080023
It would actually have far less problems.
Less shaders (and an equivalent drop in TMUs) but the same ROPs.
Great card for gaming.

Shit for Mining and Compute.

>>62079945
>Blame AMD for making asuccessful product that is flying off the shelves
nice try leather jacket man

>>62080051
Possibly, but never higher than 1080ti.
AMd just can't compete with 96 ROP.

>>62080065
But you don't need moar ROPs.
These were never the problem.
Big Polaris won't fix inherent bottlenecks in front-end.

>>62080076
You don't need moar ROPs.
ROPs on Maxwell-like are accidentally overkill due to how they work.
Aka GPC must stay the same for every die there is.

>>62080065
it would not be able to feed all CUs properly, it would literary be fiji 2.0 without any hope, vega still got some hope left

>>62080070
I will, they could have released RX590 and sold even more cards.
Ellesmere XTL for gamers
Vega for compute and 'future proofing'

>>62080081
You do if you want decent performance over 1080p.
Polaris 10 sucks for anything over 1080p simply due to be limited to 32ROP.
That's why Hawaii beats it at 1440p in most everything.

>>62080105
Polaris 10 also clocked barely higher than Hawaii.
Vega10 clocks much higher.
No, you don't need moar ROPs.
Hell you don't even need moar bandwith due to DSBR and larger L2$.

>>62079630
What prevent miners buy it again at october?

>>62080099
No it wouldn't, it would literally be 'Hawaii 2.0' Electric Boogaloo.
Hawaii still rocks today becuase it was well balanced.

Fiji and Vega are shader heavy for compute.
Great for those markets, shit for games.

Both markets can exist.
(Especially since Vega problems right now are with packaging, not die supply - turns out Interposers are STILL a bitch)
Extra capacity while waiting for packaging on Vega could have made short work of mid-high end Ellesmere XTL orders.

RTG made a bunch of stupid decisions that are now biting them in PR (only, money is fine)

>>62079941
really tempted by asus this time around
they finally did proper radeon 2.5 cooler
all these complaints about reference are by people that not going to buy vega anyway, anyone sane waits for aibs

>>62080144
VEGA LITERALLY USES FUCKING SHADERS TO CHEW GEOMETRY.
YOU NEED MOAR ALUS.
LITERALLY.

>>62080136
Nothing. AMD is creating mining drivers.

>>62080144
vega is not selling contrary to current belief, and what is selling doesn't give AMD penny more than normal margins

>>62080136
etherium difficulty will jump again in october to the point where GPU won't do shit

>>62080173
they released one mining driver and 2 gaming ones in two weeks

File: 1430236296498.gif (112KB, 255x231px) Image search: [Google]

112KB, 255x231px

>yfw hollywood and apple are buying all vegas, not miners.

>>62080236
fuck apple, they are probably the reason why gaming drivers this late because if they didn't prioritize pro drivers mac pros would burn
considering how vega eats geometry I can see why apple chose it, in power saver it is 4k ALU 200w

>>62080236
>hollywood is a lot of people

They will keep their quadro's tho

>>62079651
>it's okay when AMD does it

>>62080317
Eh nah.
SSG stronk.

>>62080317
no, they will replace all machines in the studio with one project 47

>>62080330
>It's okay that nvidia delayed Volta, we can wait while we make fun of amdfags waiting for their card

>>62080390
t. 815mm^2 die

>>62080340
Wait the drivers (tm)
>>62080342
Not even related. It's for compute and deep learning

>>62080426
They literally showcased P47 working as virtual werkstation.
P47 is mostly for that.
Cheap and VERY dense. Low TCO.

>>62080426
>It's for compute and deep learning
no it's not, go watch siggraph presentation
they demoed it specifically for rendering
meme learning is nvidia domain, amd branches out thanks to whole package with epyc

>>62080461
>>62080469
>believe amd propaganda

you did not learn the last time?

>>62080558
Are you motherfucking retarded?

>>62080558
there was never "a last time" in enterprise sector

File: AMD-VEGA-CUBE-750x400.jpg (34KB, 750x400px) Image search: [Google]

34KB, 750x400px

>>62080579
>

>>62080618
>125 TFLOPS
>TENSOR CORES
>800mm2
>but ehm, no cards, guys, but you can preorder right from me, let me take my notebook...

>>62080673
It's 120 meme FLOPS and the die is actually 815mm^2.
But yes V100 is fucking retarded.

File: ProjectQuantum1.jpg (252KB, 1200x782px) Image search: [Google]

252KB, 1200x782px

>>62080673
>

File: image_tech_A_1037180104.jpg (141KB, 580x386px) Image search: [Google]

141KB, 580x386px

>>62080751

What's in the box?

>>62080751
>gaming
>concept pc
you see the problem?
also why did you post the cube? I don't get what they lied about there? it is true that 6 vegas got that much performance.

>>62080778
i7 and two fury x everything water cooled

File: download.gif (189KB, 200x200px) Image search: [Google]

189KB, 200x200px

>>62080812

Hot.

>>62080070
>Manufacture a handful of housefire GPUs
>They get bought up by miners
>Proclaim, "Our GPUs are selling like HOTcakes!"

I can see through your lies, pajeet.

>>62080844
t. sopa eater

>>62080788
AMD talk talk talk but don't delivers irl

>>62080844
And I can see through your 0.00000017% yields.

>>62080070
>die bigger than 1080
>housefire
>expensive
>slow on gayming
>no drivers
>need to be undervolted
>shitty overclocking capacity
>low stocks confirmed
>no AiB
>fake ass launch price
>1.5 years late

Rrrrrright

>>62080926
>shitty overclocking capacity
Let's not pretend you can really OC Pascal.
It's 100mhz over Boost3.0 tops.

>>62080926
And yet its still flying off the shelves. Imagine what is going to happen when they actually get the drivers more than half finished!

>>62080945
>ship out minimal cards, basically none overseas, tons of retailers don't even get a single card
>"IT'S FLYIN' OFF THE SHELVES!!!!!1!!1! "

>>62080926
I'll bite.
>die bigger than 1080
relevance?
>housefire
it's gpu, nobody complained when 390x or gtx580 were even hotter, other thing I don't understand somebody explain it to me, how come AMD pulls 50w more but has temperatures exactly like nvidia?
>expensive
true.
>slow on gayming
not true.
>no drivers
true. but if it performs like this with bad drivers how will it perform with complete drivers?
>need to be undervolted
such hurdle, moving the sdlier to the left
>shitty overclocking capacity
hbm OCs 180Mhz, core goes to 1600 on air, but just like pascal it gives zero performance gain and only increases pweordraw
>low stocks confirmed
not confirmed, bunch of rumors spawned by lying brit, all online stores have them for stupid price, but they are in stock
retailers will come to their sense in a week or two
>no AiB
asuss. early september. asus just tends to announce earlier thane everybody else
>fake ass launch price
fake ass rumor started by lying brit
>1.5 years late
irrelevant, if you needed 1080 last year you have one, if you need gpu now, it's irrelevant

File: 1475939426731.jpg (30KB, 421x473px) Image search: [Google]

30KB, 421x473px

>>62081109
>none overseas

File: ....jpg (22KB, 684x453px) Image search: [Google]

22KB, 684x453px

Could someone tell me when the AIB cards of Vega 56 will first show up? I kinda want to wait but I've been without graphic card for two whole weeks and I really really want to play but can't...

Should I get a GTX1070 at around 430€ with shipping Monday, or is there hope for a Vega 56 with decent cooling as early as first week of September?

I'm in pain pls send help...

>>62081266
get 1070 if you need card right now, first week of september won't happen
which 1070 for 430 though? do not buy gigabyte or blower please

>>62081266
Doubt it and it'll be nearly impossible to get even if it does come out that early in September. But Vega 64 aib models are definitely coming early september

>>62081327

>Inno3D iChill GeForce GTX 1070 X3

Was thinking of this one, litteraly the cheapest though.

>>62081333

Yeah, that's what I feared too. I don't want to spend too much money on graphic either, 500 bucks would really be streching it.

> More info about the RX Vega launch

NOOOOOOO STOP ALREADY

>>62081366
>Inno3D
eh, I have same associations bout this brand as gainward
learn more about it before buying

>>62081387

Yeah, thanks I'll try find reviews for whatever I wanna buy before I order it anyway.

File: Untitled.png (14KB, 856x139px) Image search: [Google]

14KB, 856x139px

>>62081387
>>62081327

Fuck it, I'm weak lads.

File: Screenshot 2017-08-22 at 9.52.53 AM.png (68KB, 607x621px) Image search: [Google]

68KB, 607x621px

>>62081652
>Buying a 1070
>Three (3) days before Vega 56 launches
LMAO

>>62081682
And 56 will be out of stock in a mere seconds, pal.

>>62081682

Sick of waiting brah, been waiting two whole weeks since my GTX770 fried.

I'll get a Vega in four years.

File: Screenshot 2017-08-22 at 9.38.54 AM.png (61KB, 598x408px) Image search: [Google]

61KB, 598x408px

>>62081721
For good reason.

>>62081744
>>62081682

Looks tempting, but fuck it. I don't like to be teased, either drop on your knees and lick my balls or I'll just get the slightly uglier but less proud card.

>>62081744
I'm sometimes grateful that I live so far from the western consumerism that high end video cards do not sell out here in 3 days and I have time to chose.

>>62081777
honestly dude, I don't know how you did survive last two years with 770
1070 going to be x3 it's performance, so you get good deal for a year at least

>>62081804

Well tbqh I'm still playing on a xl2411t so in 1080p and my go to games are mostly Arma3, Mount&Blade, Day of Infamy and Rising Storm 2.

I recently finished DOOM, and it was as smooth as silk and looked fantastic. Shadow Warrior 2 and Just Cause 3 ran pretty good too.

Only game I still need to finish than ran like shit was Deus Ex Mankind Divided.

I wasn't planning on upgrading until at least Coffeelake came out and maybe even Volta.

I never asked for this....

>>62081682
>>62081744
It's going to run at 90C with that stock cooler. And an AIB 1070 is going to be 5-7% faster than a FE and you can OF another 10-15% on top of that. Still not on the undervolted, oc'd vega56's level but the difference won't be nearly as big because vega56 will be stuck with a shitty cooler

File: Screenshot 2017-08-22 at 9.56.06 AM.png (68KB, 611x621px) Image search: [Google]

68KB, 611x621px

>>62082028
You understand that Vega 56 is delivering this kind of performance compared to a 1070 on half finished drivers still missing core uarch features, right? Once RTG implements primitive shaders in drivers, the gap is going to get wider than this.

>>62082080

So nothing guaranteed. It all sounds like make believe bs so far. I don't mind saying.

>>62082098
>So nothing guaranteed.
>It all sounds like make believe bs so far.
Literally whitepaper but whatever.

>>62081112
>relevance?
Vega design is doa

>1.5 years late
>not a big issue

>vega rx56/64
>good on gayme
Its a good on cumpute but terrible for gaymers

>>62082121
>Vega design is doa
What?
Why.
I mean it's obvious you're absolutely clueless when it comes to GPUs but at least try to answer why do you even think so.

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

906KB, 793x877px

>>62082098
The clearly explain what primitive shaders are expected to do in the whitepaper. Increasing Vega's effective polygon throughput to up to 17 per clock would solve the fundamental front end bottleneck that always held Fiji back. It would give Vega > Pascal geometry performance.

>>62082121
>Its a good on cumpute but terrible for gaymers
Vega 56 undervolted on half finished drivers utterly dumpsters the 1070, and Vega FE's performance in pro 3D rendering applications proves that with feature complete drivers Vega excells at graphics workloads, not just compute.

>>62079630
What? So they AR not rushing shipment because they want to wait for volta?

Even if volta is nothing more than a 5-10% improvement over current Gen nvidia will wipe the floor with amd.

>>62079903
Prices are high because quantities are so low. If a store only gets 5 in they will chare $680. If they had 500 they would charge $499.

>>62081682
How is 56 so close to 64 in every benchmark. Are they literally the same chip?

>>62082267
The front-end is choking resulting in heavy ALU underutilisation.
You can trim it to 2.5k, or even lowe ALU count and the performance would stay roughly the same.

>>62082287
Is that something they can fix in drivers or did they fuck the hardware until the navi redesign happens?

>>62082267
What >>62082287 said. RTG needs to get primitive shaders implemented in the fucking drivers because that feature is specifically designed to fix this problem.

>>62082311
Read the fucking thread.
Some anon even posted a whitepaper page related to that. >>62082164

>>62082267

Memory can be unlocked to same level, only difference is 8 less render computer units and slightly lower frequencies.

File: 1469805074878.jpg (35KB, 602x393px) Image search: [Google]

35KB, 602x393px

>release unfinished product
>everyone shits on it
>fanboys angrily defend to no avail
These threads have been great

File: 1493501647689.png (217KB, 484x557px) Image search: [Google]

217KB, 484x557px

The only thing that matters is:

Is it good for mining shitcoins?

>>62082455
No.

>>62082206
1. what is the point if they can't sell any at 650
2. they would jack up the prices hoping miners swipe everything anyway, you can't be sure of anything right now

File: uv.png (34KB, 638x422px) Image search: [Google]

34KB, 638x422px

>>62082417
You mean besides getting the editor in chief of Anandtech to confirm on the record that primitive shaders weren't enabled yet? Or getting Gamers Nexus to confirm these undervolted Vega 56 results by matching them on his most recent Vega Livestream? Because those seem like they are of some avail.

>>62082548
>replying to the most obvious bait ever

The wait feels painful and I don't feel like I need this shit. May as well just grab a rx 480.

>>62082567
they sell these somewhere? haven't seen one since january

>>62082164
No game needs to push 6 billion primitives much less 17 billion. It's a fake problem they're solving.

>>62082588
That's per second, not per clock.

>>62082136
>bigger die than 1080ti
>more expensive vram than 1080ti and less of it
>higher power consumption than 1080ti
>hotter than 1080ti
>more expensive to manufacture than 1080ti thanks to yield loss from packaging the gpu and hbm into the interposed
>late by over a year
>performs significantly worse than the 1080ti

Tell me how it's not dead on arrival? Even fiji was in a better spot

>>62082601
it says right there 17 per clock

>>62082588
vega chokes on 6000 triangles, pascal does 10000

>>62082601
So for 60 fps a game would need to push 100 million primitives per frame. Yeah, sounds like bullshit to me. I'd like to know where we're getting these Hollywood blockbuster quality models from

>>62082675
It's not about pushing MORE primitives per frame, it's about doing it FASTER you moron.

>>62082649
Anon that say 6 BILLION. Giga means BILLION

>>62082577
I saw some new ones popping up here and there locally now that the buttcoin storm has subsided.

>>62082675
what happens when you can do more in one tick anon?

>>62082740
Nothing. Nothing happens, just like dsbr and accelerated discard did nothing

>>62082820
Fucking hell you're either retarded or shitposting.
Probably both.

File: vega.jpg (63KB, 720x1280px) Image search: [Google]

63KB, 720x1280px

>>62081129
>specify category:
>Toilets, pissoirs, bidets
>Fireplaces and stoves
:^)
I hope Vega 56 will be very affordable.

>>62082837
>amd "solves" two problems
>nothing happens, performance the same

>JUST WAIT THEY'LL GET IT RIGHT THIS TIME

>>62082641
>>bigger die than 1080ti
So what?
>>more expensive vram than 1080ti and less of it
Except HBM2 is much faster than outdated GDDR5 ram?
>>higher power consumption than 1080ti
Again, so what?
>>hotter than 1080ti
Wait for AIB models.
>>more expensive to manufacture than 1080ti thanks to yield loss from packaging the gpu and hbm into the interposed
If it was more expensive it wouldn't cost $200 less than the 1080 ti, don't you think?
>>late by over a year
Completely irrelevant. Besides, the 1080 ti just launched a few months ago.
>>performs significantly worse than the 1080ti
It isn't competing against the 1080ti but the 1080 and 1070, which Vega utterly crushes.

>>62082588
Wrong.

https://www.pcper.com/reviews/Graphics-Cards/AMD-Vega-GPU-Architecture-Preview-Redesigned-Memory-Architecture/Primitive-Sh

"AMD gives an example from a 4K scene in the latest Deus Ex that starts with 220M polygons of data though only 2M of which are viewable."

Don't confuse the amount of geometry in the displayed frame with how much were in data but subsequently culled.

>>62082879
Are you really THAT retarded?

>>62082887
And it's not even about pushing more tris, it's about doing it faster.
And primshader contains some other neat tricks i won't bother explaining because it would be too long.

>>62082820
>Nothing. Nothing happens, just like dsbr and accelerated discard did nothing
How would you see the benefit when the front end is bottlenecked before you even get to PDA, much less DSBR? You have to fix the front end bottlenecked before you can see the benefits of improvements later in the pipeline.

>>62082699
>Blind idiot whom cannot read a chart
It's clearly ~7.2Gprim/s according to AMD's paper, compared to ~4.5 for Fiji

>>62082879
Anon is right though, less processing time and/or lower overhead frees up resources, hence "removing a bottleneck"

If prim shaders are a single command to do multiple actions compared to multiple commands to do multiple actions, and since we know prim shaders combine culling and shading stages into a single stage, it's both lower overhead and less processing time.

Doesn't matter what the ultimate throughput is, it doesn't. That was never the argument, just the example.

>>62082934
>If prim shaders are a single command to do multiple actions compared to multiple commands to do multiple actions, and since we know prim shaders combine culling and shading stages into a single stage, it's both lower overhead and less processing time.
And it does it with lower buffer swamping/less cache flushes.
Good engineering.
Tricky to implement though.

>>62082934
one other thing everyone ignores, prim shaders are not just for geometry and culling
this giga number can be applied for everything that is rendering

>>62082887
Culling is already active. The 2m is the 2mil that the primitive shader throughput won't make a difference on. And for that matter Vega doesn't do substantially better than fiji, which doesn't have as advanced culling features compared to even tonga. And Vega's higher throughput without primitive shaders doesn't scale up properly. 50% more geometry throughput yet not even 30% more performance? Hmmm. So much for that geometry bottleneck. Solving fake problems.

>>62082913
And yet they prioritized dsbr instead of fixing the front end first. Funny higher works. One would think they would tackle a bottleneck that is supposedly present since fiji yet they go for a gesture that did nothing when enabled; Vega FE performs exactly the same as RX despite Vega FE not having dsbr at .

Who knows when primitive shader will be active, a month? Two? Shows how important it is

File: 1486612293568.png (14KB, 295x497px) Image search: [Google]

14KB, 295x497px

>>62082934
Learn how to count first.

>4 * 1.75 = 7
>1663 / 1050 =/= 1.75
>????

>>62083214
>Culling is already active.
?
>And yet they prioritized dsbr instead of fixing the front end first.
Maybe, just maybe, implementing one is easier than the other?
Please stop being retarded.

>>62083273
>?
Culling is a type of work geometry shader are used for. Throughput is the amount of primitives the shaders can do work on every cycle. Get it? The primitive shader's unique function of increasing throughput isn't active but the vs and gs functionality is working otherwise Vega would perform like dogshit and not even look right

>>62083427
Well of course VS and GS would be fucking active, otherwise the legacy pipeline in Vega won't even exist.
Fuck no one would launch Vega then, even.

>>62079815
GDDR6 production will ramp up in early 2018 but availability starts in Q4 of this year from both SK Hynix and Micron.

>AMDrones still posting "m-muh drivers" bullshit
lol by the time vega actually catches up with maxwell, volta will be coming out

>>62084540
>volta will be coming out
>dies so big they are hitting the reticle limit
Whoa, impressive.
0.00000000000017% yields are not.

>>62084583
>actually thinking the dies will be large
they have zero competition, gv100 will never make it to the consumer market.

vega needs to be gp102 size to barely compete with gp104 while consuming 3x the power, what a fucking disaster

Fuck another 2 months with a 390x

Could be worse

>>62084614
Oh, look, our resident shitposter is back.
Say 'Pajeet'.

>>62084630
you literally have no leg to stand on, first you were shitposting about vega being gods gift, then FE game out and the game changed to 'b-but RX Vega will be much better, FE is not for gaming!!!!' then RX comes out and it's no better than FE in gaming then the 'm-muh drivers!' bullshit comes around.
holy fuck you drones are so laughably pathetic and AMD are even taking you for a ride, making these bullshit 'packs' to 'deter' crypto miners then releasing TWO rounds of drivers for blockchain compute while the consumers get fucked in the ass.

but i guess you love being fucked in the ass so you're being done a favor

>>62084670
No, say 'Rakesh'.
Or something.

>>62084676
(You)

>>62083732
Are we seriously going to skip HBM now in favor of GDDR6? I guess I should have seen it coming when GDDR5X cards were able to match the bandwidth of the Fury X with HBM.

>>62084670
https://forum.beyond3d.com/posts/1997699/
>Quick note on primitive shaders from my end: I had a chat with AMD PR a bit ago to clear up the earlier confusion. Primitive shaders are definitely, absolutely, 100%not enabledin any current public drivers.
That is Ryan Smith (the editor in chef of Anandtech) confirming on the record directly from AMD that primitive shaders are not enabled yet in Vega drivers.

Without that uarch feature enabled Vega is massively front end bottlenecked in games, so it won't even possible to really begin assessing what Vega is capable of until after it's enabled.

Nice attempt at furiously shitposting though!

>>62084992
Yes, nVidia are much less of a friend with Samsung or Hynix to consider switching to HBM2.
They are friends with Micron, which fabs G5X and G6.

>>62080469
>AMD not going for meme learning also
Did you not watch the whatever presentation in which Raja talked about that shit with the mountain, going for compute, gaming and meme learning?

>>62080751
I'm really annoyed that this got announced and then basically erased from history in the very next moment. I wouldn't have bought one since I already owned an i7 4790k and pre-ordered a Fury X from the very first chance, but the concept was really cool -- especially the way the case was designed.

>>62086784
Meme learning is Nvidia's domain (for now),
What AMD aims for is VFX and visualisation in general.
Their new Vega-based offerings are MORE than stellar for that,

>>62086821
Not saying you're wrong on their focus, but just saying that Raja did talk about it, is all.

>>62086838
Well of course they would.
ROCm needs more libs and maturity in general to offer competition to CUDA.
But in visualisation - Quadro is more or less kill, especially since AMD is very active and always ready to provide support there.

>>62082884
>Except HBM2 is much faster than outdated GDDR5 ram?
The 1080Ti has 484.4GB/s of memory bandwidth, the RX Vega 64 has 483.3GB/s of memory bandwidth.

>>62087238
1080ti uses G5X, which is faster GDDR5.
No density improvements, which is a shame but whatever.

>https://forum.beyond3d.com/posts/1998059/
The thing sounds more and more interesting with every passing day.

According to whitepaper the primitive shader is a certain type of compute shader. And NGG path is described as programmable geometry pipeline.
It really really smells like AMD pulled a Larrabee that actually fucking works.

If that is the case, then the delays for Vega and current shitfest are pretty reasonable.
They are literally writing shaders into driver.

>>62087753
>AMD pulled a Larrabee that actually fucking works.
Is this the Hindu magic of Raja?

>>62087810
Considering the dude made the first 8pipe design?
Nyes.

Like really that kinda makes sense.
"Primitve shader" is described as general-purpose compute shader. And Ryan's statement about the complexity of exposing it to devs makes sense.
It's even lower than low-level APIs.
It's a pipeline-level API.
Also 45megs of SRAM suddenly make sense.

>>62079630
I fucking called it 2 months back, Vega is paper launched, custom versions launch a month or two later, and actual availability even later than that.

Every one of you shills kept saying that AMD will launch with full inventory because "that's what AMD promised", despite the fact that two months before that it was supposed to launch at Q2 because that's what they mentioned before hand.

The only surprise is the performance, which is slightly better than expected, especially if they can indeed fix up TBDR with drivers, and it isn't just fucked in the silicon.

>>62088072
It's not DSBR you braindead monkey.
Read the thread.

>>62088072
>Vega is paper launched
I can buy it right now it in stores, Im waiting for aib.

>>62079630
Remember the exact shilling Intel was doing before Ryzen?

Here it is again. I will never buy or recommend any Intel products ever again because of this ham-fisted shilling.

>>62088894
But we have the vega reviews and it's mostly a turd

>>62090298
Well it's not working properly yet.

>>62090713
When will it be working? If you don't know, then why are you shilling/defending Vega?

Because until they get it fully functioning, you should not be advocating hopes and wishes for an unfinished piece of hardware to get finished.

Should I buy a vega card now, and not wait for aib if I am going to watercool it anyway?

>>62090925
I wouldn't recommend buying Vega until they deliver on driver promises and fully functioning hardware.

But if you're dead set on it, yes, just make sure it's a full cover waterblock.

>>62090925
Stock cards have locked bios but I am fairly sure aibs will have the same issues.

Just wait for someone to crwck it there is already software reghax

>>62090978
That's another stupid fucking thing about Vega. Why did they lock the BIOS?

>>62091017
Sheckles from ms for secureboot cert

Blame Ms for fking forcing this shit.

Won't be long till it's cracked

>>62091017
>>62090978
Vega have hardware level bios check. Having a softmod to unlock vegas bios mod will be unlikely.

File: Honeyview_serveimage.jpg (199KB, 400x600px) Image search: [Google]

199KB, 400x600px

>tfw no sub $300 consumer gaming card that supports sr-iov

>>62079630
just wait™

>>62086546
>t-trust me guys 1080 ti will be BTFO when vega is working!!

>>62091729
(You)

>>62091423
Why the fuck release a enthiast card and lock it down? For all of nvdidas memes they ain't this bad

>>62080307
>Vega 64 is 56/Gflop/w at 1.4Ghz

>>62092152
ask nvidia why do they lock down voltage on theirs?

>>62092991
Mostly because Pascal gains nearly nothing from massive voltage increases. Hardmodded +250mv on the core (beyond the stock limit) gives up barely any clock ceiling, in some cases single digit percents unless under extreme cooling.
Pascal is a very heat intolerant design. You'll see the chips auto-dropping clock bins with every so many Celsius because everything already runs at the bleeding edge of latency.

This is the direction modern IC designs are going. Advanced power control techniques and operating characteristics that already approach the limits of fabrication tolerance are necessary for competitive edge.

Heat at these feature sizes is noticeably expanding the gaps between transistors, slightly enough that so many degrees of warming causes the chip to require a longer cycle delay (less clock speed) to remain stable.

All that aside, Pascal also gains less performance the higher clock speed goes - VRAM can't keep up with cycle access time of the cache so the GPU begins to stall waiting on the memory controllers. Relatively low-end chips seem be more imbalanced than higher end chips in this regard.
In a perfect world the 1050Ti should have been 192-bit, the 1060 256-bit (8GB 256-bit 1060 would have stomped all over the 480), the 1070 GDDR5X, and the 1080 a faster G5X.

On the plus side, think we'll see enough driver improvements and feature activations that when third party cards drop, reviews will start looking different to the stock cards?

Wait for NAVI

>>62094363
I'd wait for AMD to properly explain us mortals what the fuck "primitive shader' is.

>>62094377
EVERYTHING!
Sweeney at AMD event makes perfect sense now. First who makes engine that needs no optimizations wins the market.
Also explains bethesda and ubisoft.

>>62094396
>EVERYTHING!
Pixel Shader is still separated.
But yes it implies do-everything-graphical type of compute shader that modifies itself according to input data.
That's nuts.
Like really, that's Larrabee but it's with actual shaders.

>>62094411
>>62094396
Sounds like Vega can be powerful but the feature implementation is at least a quarter away.

>>62094595
We don't really know.
Hell we don't know what "primitive shader" IS.

>>62094595
wolfenstein 2.
october.
fp16.
25tflops.
will see.
predict dirt4 situation.

>>62094627
Oooga booogaaaa
>Bangs rocks together
Ooh ooh uuughhh

>>62094664
Not everything CAN use FP16 math.
Hell you don't even need it, really.
Just a neat bonus.
What matters is readiness of their new geometry pipeline.
>>62094671
Thanks for the explanation, DeShawn.

File: IMG_0401.jpg (220KB, 1242x1196px) Image search: [Google]

220KB, 1242x1196px

>>62094674
Mudniggashaders booboobaaah!

Lil nigga

>>62094752
Hands off the crack, Jamal.

>>62094771
Seriously though does anyone know what the fuck prim shaders even?

Why can't ayymd explain it better? Why isn't it enabled? Where are the tech demos? Where is their dx12.x and vulkan titles? Why are they focusing on dx11 that's 8 years old? So many questions

>>62094782
Well they sorta did it, but very-very generally.
Think of it as their personal secret sauce.
And it isn't enabled because programming exactly primitive shader requires algo of decent complexity. And hours and hours of testing after that.

>>62094805
>>>62094782 (You)
>Well they sorta did it, but very-very generally.
>Think of it as their personal secret sauce.
>And it isn't enabled because programming exactly primitive shader requires algo of decent complexity. And hours and hours of testing after that.
If they had it working they would be demoing it at the very least but all they are doing is pissing around it seems.

>>62094821
How do you "demo" it?
They showed the discard rates in whitepaper.

>>62094829
Could a non brainlet translate for me? I'm dumb.

So basically it's super aggressive culling combined with extremely low level access? Does that mean we won't need per game optimization anymore like mantle ? Or does it do it all automatically meaning devs nor and have to do anything?

>>62094849
It's entirely new (probably fully) programmable geometry pipeline. Think of it as a very general purpose compute shader that does all things geometry (yes, it's a COMPUTE shader that chews geometry).
And it's invoked by the driver.
How?
I dunno. Maybe it reads parameter data and invokes it if it fits.

>>62094867
So why isn't it working then? What about the new hardware on Vega cards is that still turned off as well?

Will it change the gflops ratings for the better? Then there's floating point 16 isn't there as well?

The more I learn about Vega and what it's supposed to be capable of the more I want one.

Then I look at its actual performance, power consumption and prices and I suddenly don't want one anymore.

>>62094888
Because it's needs to be PROGRAMMED.
That also implies very throughout testing with variety of game engines.
Everything else is turned on, but it doesn't even matter when Vega is still geometry bottlenecked.
FP16 is only partially relevant for gayms.
It's good for low precision compute effects and checkerboarding but that's about it I guess.

>>62094894
Don't worry you won't be able to get one anyway.
So sit here and learn for fun.

>>62094908
So AMD have literally remade the cell broadband smid based spe/CPU into a gpu? Why the fuck use this for gaming? Didn't Intel can larrabee back in 2009-2010 because of shit performance?

I wish they made big Polaris in the meantime because vega clearly isn't read if half the card is locked off and essentially bricked outside of compute non gaming related tasks

>>62094944
No, these time its actual fucking shaders.
Larrabee was supposed to be a second coming of GPU Jesus and it would be one if not for it using a bunch of low clocked in-order cores running HUEG vectors instead of actual shaders.
>bricked outside of compute tasks
Hwhat?

>>62094944
that's the point, it's larrabee that works with shaders, maybe, probably

>>62094969
So AMD made the holy Grail of GPUs and won't actually turn all the features on because they literally don't know how? Fuck me!


You vegacucks must feel so frustrated but I know exactly how u feel.

T. Early 290-390x adopter
>>62094964
Is prim shaders dsbr and all the other fancy hardware based new shit api dependant? Or can old games on dx11 benefit from this wizardry as well?

>>62094894
>power consumption
Reminder that you can turn it down by 200 watts, putting it in the range of 1070/1080.

>>62094994
Pretty weird AMD didn't do microcode updates like they did with 480 polaris

>>62094991
Every game benefits from moar tris/clock, or faster lower latency raster.
And they KNOW how to turn it on, but it requires testing.
A lot.
Think of it as writing shader from scratch: you define input data, shader itself, output data, and in that case, culling algo.

>>62095005
all power benchamrks can go into the trash right now, it didn't even have voltage profiles release day

>>62095009
So basically they need to code machine learning ai into the driver for prim shaders so it knows what to do or what? Why the fuck can't they just explain what the fuck it is what it does how it works and show it off ffs
>>62095014
updated benchmarks? Where

>>62095005
I hope that they are still tweaking things for the disabled features. It's not worth to tweak the actual microcode if they are throwing it away soon (tm).

If that's the case. why wouldn't Nvidia rush Volta to the market, or something.

>>62095033
Meme learning has nothing to do with that. The algos for culling in shaders must be pretty simple and strict.
They are not going to explain it as long as the feature is not fully ready for both driver stacks present.
No, simply saying
>we just pulled a working GPU Jesus
won't do it.
Nu-AMD knows it very well.

>>62095047
Because it would take time, considerable time, to polish Vega.
And they CANT rush Volta, 12nm FFN is still VERY costly, and GDDR6 is also not even in HVM.
They simply can't do that.
They can only tell you to buy Pascal.
It's good enough.

>>62095047
1. they physically can't
2. it's cat in the bag, nobody knows the heck is it, people started understanding what is going on only couple days back

people in industry probably know something because contacts with AMD increased drastically over last year

>>62095064
Oh yes, cocksucking Sweeney appearing on AMD's stage is more than weird.

>>62095053
So it won't require new apis and shit to use it if and when it comes?

Dx11-12.1 level is fine or will Vega be future proof like gcn one point zero and support dx13 and beyond?

>>62095080
It's literally all API-agnostic features.
Like "primitive shaders" do not exist anywhere besides fucking Vega.

>>62095072
Fucking Sweeney I hate him so much backflipping faggot
>>62095045
Why would they throw an entire uarch away? Stop meming

>>62095005
They wanted but people figured out the undervolting stuff before they could unfuck the cards.

>>62095087
So technically it's non api dependant? That's all I wunna make clear.

What I don't get is how ryzen launched ok out of the box and had memory scaling issues and other shit fixed quick compared to Vega that's almost been out for just as long in the non gaming side of the industry.

And if Vega indeed sucks ass at gaming will they just dump it and drop Navi early?

>>62095106
So the cards work good out of the box now? A 64 will boost up to or over 1600-1700mhz or is that bullshit because Nexus said they where hard locked at 1750 or something

>>62095108
No, it's not.
And CPUs are very very different to GPUs.
CPU launches are much easier.
There's more to GPU than simply hardware, and in case of Vega there's much, much more.

>>62095108
>ryzen launched ok out of the box
haha. it was better than x99 launch, but no where near okay

>>62095117
It depends on silicon lottery but you can both undervolt and decently OC Vega right now. The cards are overvolted to hell stock.

>>62095117
Don't knew about the OC, but the power saving mode uses 200 less watts and the card loses about 4% game performance.

>>62095127
>X99 launch
>people but super ebin high clocked DDR4
>it sits at 2133 even after years of updates
Man that platform was pretty fucking bad.
At least 100 series that followed was much better.

>>62095108
>Vega that's almost been out for just as long
it took 2 months to fix ryzen more or less
vega pro performance is fine
rx vega is one week old

>>62095139
Fuck it's only been 1 week?

October is ages away reeeee

>>62095158
Yes, it's only been 1 week.
And they even fucking managed to fix voltage control.

>>62095175
it should've been working to start with, but it's amd what can you do

>>62095182
Why did they even talk about dsbr and prim if they don't work? Imagine if Nvidia did this /g/amers would blow their top but because amd does it it's ok.

Then there's the whole other slew of issues like pricing availability and shit performance.

T. 390x owner

I never understand these threads. Every time I browse through it's just amd fanboys getting completely systematically destroyed.

>>62095223
Because these are inherent uarch features of Vega.
And DSBR works anyway.

Fucking finally someone has the balls to move towards more flexible, software-defined GPU. We've been milking good old unified shaders for too long.
>>62095238
I don't think people talking and trying to guess how certain uarch features work fits into definition of "getting destroyed", shitposter-kun.

>>62095238
Ironically by other amd owners who aren't fucking retarded.

Imagine all those stupid fucks that went from Hawaii xt to Polaris loool
>>62095239
Yet it's still slower than a 1080 unless u do tonnes of tweaks

>>62095274
>It's slower than 1080
If course it would be, this is your big Polaris. The actual Vega will come later with proper programmable pipelines.

>>62095279
How pissed off would /g/amers beif this was the case?

>>62095274
>went from Hawaii xt to Polaris
nobody did that, what are you on about?

>Yet it's still slower than a 1080
if 3% ahead and 3% behind in other case can be considered slower, okay

>>62095298
What do you mean by that?
Current Vega behaves exactly like big Polaris would. The actual Vega requires some particular software efforts to work.
But in return it offers something that we've never seen before: flexible software defined graphical pipelines.

>>62080357
What company in their right mind would release volta right now when Pascal is already king of the hill?

That would be the most retarded thing ever

>>62095315
>>62095315
Big whoop only 2 games even run properly dirt rally and doom 2016 and even still the 1080ti is cheaper uses less power and beats Vega in every other game

>>62095323
They also simply can't release it.

>>62095328
What the fuck are you talking about?
I literally said their new, now yet implemented geometry pipeline, offers something we've never seen before.

>>62095328
come on, make an effort, post a graph or two, I don't know what happened
rx580 launch graphs as far as eye can see trashing it, vega no graphs, I liked to get some actual info from these threads now it's shitposting without any info

File: Screenshot_2017-08-26-14-43-24-16.png (444KB, 1280x720px) Image search: [Google]

444KB, 1280x720px

>>62095351
Add 18% to this for vega and there's ur grpah

>>62095383
so one fps slower than 1080 in UE4 least favorable scenario with potential to grow more? amazing.

>>62095457
oh wait, it's aftermarket 1080, even better, thanks anon

So wait for Vega?

Tempted to go get a 64 now and water cool it since aib prices are fucked

>>62095323
So it's not okay when AMD delays a product, but it's okay when their rival company does the same on the reasoning that "Lol, we don't need to, we can milk these retards for another year"?

>>62095498
I have a feeling prices will calm down in a week or two because there is not real mining interest for vega.

>>62095589
56 will probably be better for that anyway undervolt and shit to 150watt we I mnot a minertard

>>62095543
Yes, it is okay, are you to dull to understand the different position both companies are in?

File: 32.png (126KB, 776x521px) Image search: [Google]

126KB, 776x521px

NANI?!

>>62095457
At twice the power consumption.

>>62095457
1070

>>62095642
Curse you for tricking me

>>62095682
yeah well sorry it might just be germany
>https://www.mindfactory.de/product_info.php/8GB-XFX-Radeon-RX-Vega-64-Black-Aktiv-PCIe-3-0-x16--Retail-_1186799.html

>>62080191
And what they will use to mine then? ASICS? Oh wait, you can't use those on Etherium

Tfw friend and I both walked into Microcenter on a whim and got a pair of Vega64 reference models by sapphire for 641 a pop no problem. Was going to wait for the AIB cards but my tard impulses are strong.

Fucking love Microcenter

File: Capture.png (197KB, 1032x541px) Image search: [Google]

197KB, 1032x541px

>>62079630
I cannot wait :^)

File: trick.jpg (324KB, 2560x1313px) Image search: [Google]

324KB, 2560x1313px

>>62095689
I'm from Germany too and it say 649€ for me.

>>62095661
what does it say about the game if 1070 is 3 fps away from 1080?

>>62095742
click on 'Mindstar' you fucking brainlet

>>62095699
you won't be able to use GPU either, ROI won't ever happen no matter how many gpu you have

>>62095759
I don't fucking care about your brand-wars bullshit
Just pointing out some people apparently can't read

>>62095759
That means it's unoptimized as fuck

>>62095772
what is 18% from 56?

File: undervolt.png (105KB, 1204x1407px) Image search: [Google]

105KB, 1204x1407px

>this kills t.letherjacketmanlet

File: 1482430140430.jpg (85KB, 1621x433px) Image search: [Google]

85KB, 1621x433px

>>62095742
here is real kraut vendor you should use

people are still going on about these gpu? it's dead. you can stop shilling for it now.

>>62079854
>tech journalism
it's not just tech, it's pretty much most journalism today, it even came to show hard during the election:
clickbait and sensationalism, a bunch of shit being spewed all for $$$.
Video game, tech, political, you name it, everyone is biased as fuck.

journalists manipulate information and lie to you.

>>62095807
I'm honestly still baffled how AMD managed to fuck it up so much. Why wouldn't they clock it at that out of the box.

Buy AMD for worse graphics and worse compute.
It's cheap, cheap in price and performance. Cheap stock too because AMD are bad at making money.

1000 bitcoin whales keep the entire company afloat.

>>62095826
>3% off
>two stars
kek

>>62095884
i guess they only set Voltage to 1.2V and do a test run. it's totally shit, everyone knows. bad PR hurts

>>62095634
Nice double standards, shill. Now fuck off, your sopa will burn.

>>62095770
Right. Found it.

Not sure I want to buy a reference card though. How complicated is it to mount an aftermarket cooler on reference models?

>>62095915
that's why i'm saying prices will drop

>>62095998
well look on youtube for this i guess

>>62095998
as complicated as on non reference models

File: 1503561450718.jpg (437KB, 2560x1440px) Image search: [Google]

437KB, 2560x1440px

Guys, just wait Navi

File: 1463726086702.jpg (64KB, 709x538px) Image search: [Google]

64KB, 709x538px

>tfw 1080
>power efficiënt powerhouse
>been enjoying it for over half a year
>AMD releases another trainwreck which takes a nuclear plant to power even with HBM2, with GDDR the power draw would even have been 50-100w higher

Glad I didn't listen to you dumb cunts

>>62095981
Applying the same standards for two different situations just shows you're a simpleton with a lack of understanding of basic business strategy

>>62097360
>it's okay to hold off technology to maximize profits

>>62097415
Do you mean Volta?
Because they can't rush it to consumer, their half-node is barely ready and GDDR6 is not even in HVM yet.

>>62097460
I know, but for some reason some people so sure that intel and nvidia have secret tech hidden away somewhere and can release it any moment

>>62097537
Eh nah.
Nvidia can only make bigger dies.
Intel can only throw moar coarz.

>>62097537
To be fair Intel hasn't felt any sort of pressure in years. They could afford to come out with these marginal upgrades to Haswell year after year.

>>62097813
To be fair making a better CPU is muuuuuuuch more difficult than making a better GPU.

>>62097847
You know Intel is sandbagging though.

>>62097928
Oh yes, totally.
I s-swear they totally have some secret jewish tech.
Hopefully it's not Tejas.

>>62095689
tempting, but I bought a Radeon RX 580 two months ago. I could sell it without loss, but I would prefer a non-reference design and more mature drivers.

>>62097949
You're seriously suggesting Ice Lake will be another marginal upgrade?

>>62098071
Yes, you fucking moron.
Core is tapped out.
*lake suggests that very well.

>>62097310
This.

Fuck amd at least for GPUs

>>62098092
>"Core" and "-Lake" meaning anything at all
Ah, yes. Truly I am the moron, and it's just another evolution of Nehalem.

>>62098169
It's just another evolution of SB: moar vectors edition.

File: Screenshot_20170822_at_9.55.12_AM.png (71KB, 607x615px) Image search: [Google]

71KB, 607x615px

>>62097310
>>62098121

>>62099279
Can't beat a 16 month old 1080 with a tonne of mods

Still front end bottlenecked

If prim shaders fix it we would be seeing Titan xp pascal oc perf but it's not happening is it?

AMD fucked up hard it's gcn1.x fury x all over again

>>62099326
They didn't.
It's a first GPU with actually programmable graphical pipeline.
It's fucking awesome.
But needs programming.

>>62099348
Just the primitive shaders are. Replacing two fairly simple units with one general purpose unit sounds like a lot of unnecessary work with few practical advantages, just like HSA turned out to be a cool but half-baked tech that no one wanted to use. And one unit needs to do the work of two units in a given time span.

>>62099348
And what makes you think AMD is up to that task? Throw in Navi and it has to be completely transparent from chip to

>>62099393
There's not strict definition to what "primitive shader" does.
Right now it's only supposed to do culling, more - later.
>few practical advantages
U wot.

>>62099393
That's what they said about the 8800, x1900 and larrabee

>>62099431
If only Larrabee used actual fucking shader instead of small in-order cores meme.

>>62099414
Navi is just mcm Vega on a 7nm fab 4+ core config screenshot this
>>62099440
It was a different time we had power PC and the power of the cell ffs.

So 56 (at $400) will be good bang for buck at least or nah?

>>62099472
>It was a different time we had power PC and the power of the cell ffs.
Ye in-order cores running HUEG vectors was the hottest meme back then hands down.

>>62079630
Its yet Another Major Disappointment. Honestly might go for the Volta 1x80 even with my freesync monitor

>>62099497
yeah, you can make it run at 64 with free memory oc and undervolt

>>62099531
>even with my freesync monitor
But you dont have one.

>>62099497
Nope hbm yields prices and stock are horrific expect too wait until 2018 for normal prices
>>62099531
Reeeesync is a meme

>>62095543
Yes, if you are the market leader, it is OK to delay products.
If you're lagging behind, you should be delaying products.

>>62099556
so, delaying is always okay? good to know

>>62099541
You think I would go on the internet and tell lies? Its a viewsonic 144hz monitor. I figure high end Volta should be able to hold that in most games at 1080p so freesync wouldn't matter anymore

>>62099556
That's why their Pro lineup is right on track.

>>62099568
>viewsonic
my condolences

>>62099589
Dis

Company went to shit a decade ago

Get a Vega 56 or should I keep my R9 nano?

>>62099589
Haven't had any problems with it so far. I don't see what the problem should be.

>>62099602
Keep Nano until Vega starts actuall working.

>>62099602
keep your nano for now, wait and see if magik drivers happen or not

>>62099602
Nano

Vega fucked up bad they bet everything on prim shaders and even after a year long delay and hbm2 fuckup it's still not working

>>62099642
>year long delay
?
>hbm2 fuckup
?

>>62099673
Polaris was a stop gap

Vega was meant to come out q1 this year and now we won't get stock till q4

Hbm is in short supply as production is fighting space for other shit

>>62099716
>Polaris was a stop gap
>Vega was meant to come out q1 this year and now we won't get stock till q4
And Maxwell was supposed to be 2013.
Delays always happen when the arch is at least somewhat new.
>Hbm is in short supply as production is fighting space for other shit
Not exactly.
But memory in general is in short supply.

>>62099716
>we won't get stock till q4
there is plenty of them around, nobody buying them for retarded prices

>>62095771
And CPUs are even less efficient in that sense.

File: 1441666291626.gif (2MB, 288x210px) Image search: [Google]

2MB, 288x210px

>>62099753
Mining meme will burst soon anyway I can't wait

>>62099804
Yeah, cheap used GPUs are worth the wait

>>62099810
I wouldn't waste money on that, only if you have no other choice. Plenty of manufacturers silently added "no warranty" rule in ToS if used for mining.

>>62099620
http://www.tftcentral.co.uk/
Please refer to this next time you need a monitor.

>>62090298
>reviews
most reviews are shit, but morst reviews are by gaymers, the raw computational ability of the thing appears to be excellent though

>>62100335
The Pro performance is more than decent.
It's just Vega's software stack is blatantly unfinished.
For a GPU that desperately needs one.

>>62099837
And you can just say that it wasn't used for mining. Boom! Your warranty is magically back