VEGA HYPE THREAD

VEGA HYPE THREAD
if you live under a rock watch this video: youtube.com/watch?v=hp92_yx9t78

discuss all we know and what you hope for with VEGA

Other urls found in this thread:

youtube.com/watch?v=AHtu2ACL77M
youtube.com/watch?v=I-wFVIlMY_0
twitter.com/AnonBabble

why did polaris not have HBM2 or even HBM?

Your videos are shit, stop shilling them, no one cares.

Is AMD going to deliver?

i know the 570 isnt part of the vega line but what does this mean?

When are the cards coming out

I'm about to buy the 1070 today if AMD doesn't announce anything

watch the video
first half of 2017

Last time it was said in may

kill yourself you piece of shit

V for Vega and R for Ryzen makes...

V.R = VIRTUAL REALITY AHHAAHAH WE FUTURE NOW BOIS

THIS.
HAFL LFIE 3 CONFIRMED FOR VR

EVERYONE GO NUTS

>notebook
Mobile variant of polaris. Generally speaking something like an r7 350 or 360 would become an r9 490m.

just go buy 1070, and wait for volta and prob navi next year or two, fuck vega

Any speculation on power draw on idle and load?
I really don't want to be memed on by an AMD CPU that uses 300watts idle and a amd 250watt GPU idle

No idea, probably around new titan X

i dont think performace will be on titan x level that card is 1200 dollars
i think it will be more like 1080 or 1070 like

I say power draw not the performance

Honestly if hl3 came to vr I'd probably sell the car to get the stuff needed to play it

oh sorry man. the TDP will be around 225W according to the video

there will be a HLVR game 100%
youtube.com/watch?v=AHtu2ACL77M

>When are the cards coming out
enterprise first, depending on their demand, we'll get some scraps

Vega goin to be bad, but im goin to buy it anyway who with me

Waiting for vega pro cutdown, for 90% of vega xt performance for 70% of the cost,

>why did polaris not have HBM2 or even HBM?

HBM1 is too expensive, requires 1 module per GB and a massive silicon interposer if you want 4 modules, and 4 GB/s would provide 500 GB/s of bandwidth, which is way more than a 5-6 TFLOPS card needs.

HBM2 wasn't ready yet, and would have been too expensive anyway even if it were slightly further along in development.

>what does this mean?
polaris rebrand

>When are the cards coming out
>I'm about to buy the 1070 today if AMD doesn't announce anything
H1, most likely May. Just buy the 1070 now, faggot.

>speculation on power draw on idle and load?
225W for a stock single card clocks.
300W for a dual card in H2, so a Vega Nano might get down to 150W.

Fiji had some unexpected bottlenecks, but I might get a Vega this round as revenge for how Nvidia has treated my GTX 670 later in its life.

Probably a Polaris 10 rebrand with better binning.

>Hype
>Even biggest AMD advocate out there gives up on it

BTFO HOW WILL Sup Forums POOJETS EVER RECOVER?

>die size 80% bigger than the 1080's
>uses expensive HBM2 instead of GDDRX
>performs roughly simmilar in Doom
>probably uses much more power which is why they taped off the power connectors in CES to hide it has two connectors.

Sounds like AMD will release another great card!

Because of this, I'm afraid that Vega will have astronomical prices just like the Fury did. HBM and its interposer destroyed yields so bad that they could afford to include nonfunctional chips as memorabilia bundled with each Fury.

Stop samefaging the same fucking shit-tier YouTuber.

How G.N. sounded like in his AMA there won't be a HL game, atleast not this decade.

It's hilarious how this faggot spins his words trying to blame consumers for not buying amd while amd has nothing to put against gtx1080. Meanwhile amd not delivering a high end product is considered "growing up".

you realize that Fury's astronomical prices were below what the GTX 1080, a card with a 256b GDDR5 interface, costs, right?

he said multiple times theyre working on a big VR game.

Vega looks like an enterprise GPGPU with a few afterthoughts to mainstream gamers.

AMD wants corporate shekels first and foremost, leaving gamers to get anally raped by Nvidia monopoly pricing.

It won't be HL. They probably are experimenting with tons of new I.P.s

jesus fuck he is retarded. if AMD releases good products, people will buy them. nothing else to it.

polaris is proof enough.

>AMD wants corporate shekels
too bad they're like 5 years late to the party.
nvidia already has a developed ecosystem around cuda

This decade has been so shit if you don't buy secondhand or get a great deal.

they're trying to get a fresh start with the fp16/DL craze.

the number of customers with GP100s in their actual hands is still only in the dozens, so Nvidia strangely isn't already entrenched there.

The fp64 simulation market is going to be a bloodbath between Knights Landing and GP100, with Vega 10 apparently sitting this round out.

That's not the main problem. The problem really is with the consumer. It's anecdotal, but I know two meatbags that bought 1080s when they had 970s and are still running 1080p 60Hz monitors. They don't think at all, they just buy whatever seems like the newest greatest thing. See Apple for example.

>performance roughly equivalent to 6 month old nVidia cards

Meh.

Why can't one of these companies make some massive breakthrough and just suddenly double performance out of the blue? At least that'd be somewhat exciting.

AMD doesn't have the budget.

>performance roughly equivalent to 6 month old nVidia cards
It will be a year old card when vega is released.

Because if they figured out a way to do that they could cut all research for the next 10 years and just slowly scale up their cards every year using that technology until they're forced to invest more money in research again.

These are businesses. They don't care how performant their stuff is, as long as it's competitive enough to sell.

If AMD could figure out MCM and make small dies act like CPU cores, then we could have very high performing cards at much lower prices.

I'm surprised there isn't a polaris10 tier fp64 part, giving Hawaii fp64 performance on a much more affordable platform. A dual gpu board is much more manageable for a 300w 5tflop part with easy yields and simple memory (only 256bit bus). Obviously I know a whole new die, even a shrunken hawaii is a big deal. But di hawaii or did hawaii ever compete well in the market?

Stop being tech illiterate. They demoed some Vega chip (reminder: there are two) with unoptimized driver at unknown clockrates.

The big Vega chip is guaranteed faster than what they demoed.

inb4 video presentation featuring Raja Poo in the Loo Kodookie staring at twitter talking about how 2x 490 in crossfire are cheaper than a 1080ti in "insert beta game here"

>unironically believing this

>The big Vega chip is guaranteed faster than what they demoed.
and you know they didn't demo the "big" vega how?

Feels bad when liking AMD and being associated with these kinds of meatbags. They should big Vega, it's a fat chip. You can stop now.

>should
I >should leave, I haven't slept in almost 40 hours now. See you in 8 hours.

>should big Vega

They said it was flagship Vega in their promo video with Doom. What are you expecting? 30-60% performance increase at launch compared from said demo? At max 20%. And that would still be disappointing.

chips are already about as good at milking out raw FLOPS per watt or area as they will get on a giving lithography node.

all the remaining improvements can be divided into the following areas:

> higher utilization rate inside ALUs
- use bigger register files in SIMD units like GP100 for more warps/wavefronts in flight at a time
- try doing variable-width SIMD like Vega purportedly is adding
- get SIMT to maybe handle non-coherent branching better someday, though this is a known Tough Problem

> less demand for memory bandwidth
- better L2 cache crossbar designs
- compress ROP writes (delta color compression) like Nvidia has done well for a while
- add cache coherence between ROPs and shaders like Vega is doing
- use tiling/binning rasterization to maximize working set locality characteristics

> easier access to broader memory pools/other compute nodes
- use coherent physical memory space like everyone's doing
- add a shit-ton of external IO like GP100's NVLinks
- treat at least part of local HBM pool like an enormous L3, like Vega's High Bandwidth Cache Controller probably is
- allow rasterizers to queue fragment work to SMs/CUs in remote GPCs/SEs, if you ever want MCM scaling to work right

> adding further HW acceleration to particular operations
- add new ALU intrinsics for special math operations
- make existing fixed function units more programmable to have more direct acceleration of new effects (Vega's primitive shader)
- crank up tessellation acceleration to 11, flood market with HairWorks titles

> doing less work where you can get away with it
- more aggressive culling of non-visible triangles/fragments/pixels as early as possible in the pipeline
- return of 2x fp16, starting with GP100, then Vega, then probably the Volta line
- have your driver team rewrite game engine pipelines to do less work for hopefully visually equivalent results

There was an ano quoting an AMD exec saying that Vega would be competing at the highest end.

They might not deliver, for sure, but they will go for it anyway.

stop talking out of your arse..
Search for AMD and Nvidia R&D disclosed costs. They are very close to each other.

I never took him as meaning that. Pretty sure he blames AMD.

my personal prediction is that the Vega hardware will be solid but like before it will take another 3 or 4 years before software can use it competently.

So, when is the stock price going to go up?

True.

adoredtv is like Gartman, if they say something will be good it'll be disappointing, if they say something is bad it'll be great.

Ive had AMD GpUs before and they're all godawful, those Indians cant code drivers for shit

Their CPUs on the other hand can be a mad decent cheap alternative to Intel which is great

Yeah, no FP64 product till Vega 20

When is Vega coming? I'm getting bored of waiting desu.

They said it was Vega 10, and Vega 11 is known to be smaller, clock rates weren't known but you're probably the same fuck who copypastad constantly that they were using Fury X drivers.

Vega 11 is gonna be a Polaris 10 upgrade/successor, so I don't know how anyone thinks the 4k demos were running on some card with probably 256 GB/s of bandwidth.

I'm more optimistic about Vega than most of Sup Forums, but there's no way it wasn't the 500+ mm^2 big Vega being shown.

I wouldn't be surprised if they kept Polaris 10 around as a lower end chip, certainly Polaris 11 and whatever the fuck Polaris 12 is.

That mystery 2560 shader AMD chip is still MIA.

>videocardz.com
And today's shilling cycle is complete.

A 2560 shader Polaris card would be a waste of money to produce.

Most of the gossip was by people who assumed that even the RX 480 was a cut part, which is pretty clearly not the case from a die shot.

Why do you even come into GPU threads, all you own is a laptop with a GT 430 iirc, and before that a 9400 GT. You don't even play games.

>I know two meatbags that bought 1080s when they had 970s and are still running 1080p 60Hz monitors.

Are people really this stupid?

Titan X performance would not be disappointing

>Vega 11 is known to be smalle
[citation needed]

Nah there was a 2560 shader (specifically 40 CU mentioned) part in some shipping manifests and driver code a bit before Polaris launch, along with the 38 CU Polaris. At the time people assumed it would be Polaris 10. Vega 11 being that size would make sense, but the release timeline (11 after 10) doesn't fit. AMD has been known to make pipecleaner products, perhaps early architecture test chip?

When they skipped a whole generation with even more dev time it is.

36 CU Polaris***

Roadmap confirms it bruv, sorry. There's not going to be a magical 96 CU part like some people were shilling a year back.

hey OP, how about a video that's not complete shit?

> youtube.com/watch?v=I-wFVIlMY_0

Of course. Look at spec threads with anons having 980Ti or 1080 with shit monitors. At least some of them say they're upgrading their monitors.

What the fuck AMD why are they so objectively shit?
I really want to keep buying AMD but there are literally no comprises. It draws nearly double the power of an Nvidia card and is still outperformed by one. How can you fuck up this bad

Sup Forums - Shillers and Shitposters.

Not even a shill I've bought nothing but AMD since it was ATI but the literally Poos have been ruining it since

The Ryzen CPU they demoed was running at 95W

There was a time when Ati released cards that were both more powerful and cheaper than their Nvidia counterparts. People still bought Nvidia in droves. It doesn't matter how good your product is if your marketing is awful.

About the only thing that a 2560 shader chip would make sense as would be a Vega 11 replacement for Polaris 10, which used either 256b GDDR5(X) or 1024b HBM2.

It's way too close to coexist with Polaris 10.

Some simple math based off what was given about the MI25 card, which is Vega 10, gives a clock of 1500+ mhz. If they get any IPC improvements then a 2560 shader chip would have sufficient difference for both to be viable products. Remember HBM increases fabrication costs quite a bit so it's not just die size cost.

sorry meant to say that it's way to close to Polaris 10 to be another Polaris.

11% more shaders plus 20% more clock and better IPC would definitely be enough to put it sufficiently ahead of RX 480.

>why don't they try something drastically different

That is literally what the entire GPU market has been about since forever. "Video/graphics accelerators" of yore all had really similar design schemes, namely a combination of pipelines and texture units.

Pretty much all they did was draw in texture pixels in accordance with polygon geometry the CPU calculated and maybe some minor shading and post processing.

It wasn't until the geforce 256 video accelerators could start doing some proper calculations of their own not only well, bet potentially better than a consumer CPU. The big change was the inclusion of a hardware transform and lighting engine. Hilariously it was rarely supported until another 2 generations later when it became a requirement for the newest DX.

I'll spare the long details, but from there we've reached a point where GPU's are so good at compute they have their own special place in the enterprise market. And their underlying architecture is so vastly different from where they started they can hardly be considered in the same realm.

Each generation of architecture has more or less brought something big and fancy and new to the table. Sometimes it worked really well, other times it was a dramatic failure (lol Fermi). In the case of AMD's GCN arch and its savvy async compute ability, it has been limited by how DX dictated tasks be handled within the system, completely ignoring its compute prowess.
And even now with the low level api's, the GPU's true strengths are only played to if the game's engine is designed to take advantage of them.

There was a Curr tech rumour of a chip using gddr5x, assuming that's vega11, 2560 cores at 1500mhz by my guess leaves it competing with a 1070, leaving fat vega10 and it's cut down gunning for titan and 1080 tiers.

What's also up in the air is if polaris 10 has a spot in the 500series or what polaris 12 is, possibly a polaris 10 with gddr5x

why would they make another Polaris at this point, instead of just an even smaller Vega that used GDDR5(X)?

This fucking retard even said in the video
>it might have been throttling
>The drivers are in early development

>titan x just needs HBM2 (which, it fucking has) to destroy AMD even if they added another 50% performance

This guys a tard, thats all I gotta say.

which Titan X does he think has HBM2, TXV?
it's sure as fuck not TXP/GP102.

Why would I wait for Vega when I could get a 1070 for $380?

He probably is mixing up GP102 and GP100, consumers will never see GP100

Vegas will shakeup the market

Any reason to leave Hawaii til Navi? Will have this card for over 2 years come this summer,hope the new cards aren't fire spitters at heavy loads.

Please be Portal

Because Polaris 10 and 11, by design, are meant to be cheap as dirt to produce, have high as fuck yields, and give decent margins despite being (we'll ignore the supply/demand markup since AMD only sells the chips to third parties and the extra cash likely goes to to them, and maybe just a bit of a bonus commission determined by sales volume goes back to AMD) targeted at a $200-240 price point.

Like they did with the entire rest of the GCN 1 family they could just scale it up with a couple tweaks.
I don't know if they'll bother or not, though I'd imagine they haven't done so yet because they've got their limited resources tied up in zen and vega.

AMD has some crazy talented people under their wing, but they can only handle so much at a time. If they got too ambitious and tried to expand byond what they can reasonably handle they'd go the way of 3dfx.

>>probably uses much more power which is why they taped off the power connectors in CES to hide it has two connectors.
JUST

this

But the best of the best still comes from Nvidia and intel.

Titan line up has no competitor from AMD.