Vega launch slides leaked

File: amd-deck-03.jpg (229KB, 961x540px) Image search: [Google]

229KB, 961x540px

>>59677761

File: amd-deck-06.jpg (369KB, 961x540px) Image search: [Google]

369KB, 961x540px

>>59677783

Oh shitttttt. Can't wait to see some prices.

File: amd-deck-01.jpg (292KB, 960x540px) Image search: [Google]

292KB, 960x540px

>>59677795

April fools bullshit

File: amd-deck-02.jpg (328KB, 961x540px) Image search: [Google]

328KB, 961x540px

>>59677848

>>59677761
>Under NDA until April 1, 2017

File: amd-deckfix1-04.jpg (688KB, 961x540px) Image search: [Google]

688KB, 961x540px

>>59677869

>>59677861
that a very real possibility, but a lot of the details are pretty subtle for a hoax.
for example, Doom having less of an comparative edge than the other games makes sense if Vega really does have a binning rasterizer, since idTech 6 already does spatial binning in the host software side.

>>59677761
>Rage
Yeah when benchmarks drop.

>believing amd hype ever
it's like you retards never learn.

File: 1060-3gb-hitman-1080-100680486-large.png (49KB, 580x908px) Image search: [Google]

49KB, 580x908px

>>59677947
yep

File: april1.png (298KB, 640x430px) Image search: [Google]

298KB, 640x430px

>>59677761
You guys are fucking morons.

>>59677861
It is not April fool's you europoor

>>59677761
this reeks of April Fools bullshit, but I want to believe...

File: 2017-03-31_23-48-24.png (4KB, 670x152px) Image search: [Google]

4KB, 670x152px

>>59678025

>RAGE
>FURY
>RAGE X FURY

Why does AMD name their GPUs after their feelings towards Nvidia's success?

>>59678083
>being indian
typical of a nvidiot

File: amd-deckfix3-08.jpg (243KB, 961x540px) Image search: [Google]

243KB, 961x540px

>>59677761
read the last line of the endnotes, faggots.

>fooling AMD drones into thinking Vega isn't going to be shit
Brilliant.

Pretty convincing for an April Fools joke

>>59678066
>haha amd btfo nvidia very funny
It's 4D chess by amd. They leak this stuff, everyone takes it as a joke, after some weeks everyone want real charts and start realising they were real since day 1

It has the same number of cores as the fury X with the difference that it is clocked roughly 55% higher. So it's safe to say that it wil be at the absolute minimun 55% more powerful than the old fury X

File: 1413505880014.png (161KB, 293x271px) Image search: [Google]

161KB, 293x271px

>Rage x Fury

>>59677761

Like, I want to believe. I do, but I don't until I see someone not retarded or a shill do benchmarks.

The other thing. . . RAGE X FURY? Really AMD? Naming your shit too gaymury just turns off regular customers IMO. 1060, 1080, 480, etc are fine, but if its constantly branded as RAGE X FURY instead of say, RX680 then that shit is lame.

>>59677761
So I'll have to choose between a cut-down card and a whiney water cooler? Deja vu.

File: SapphireRear.jpg (696KB, 2500x1064px) Image search: [Google]

696KB, 2500x1064px

>>59678429
The actual Vega will be ~225W and be on a Fiji-sized PCB.
Hopefully this means that non-X Fury cooling solutions become the norm.

>NDA until April 1st

yeah, totally legit.

>>59677761
Too bad this is all bullshit; the 4gb Rage would be pretty nice in my machine, especially if they can release it cheap and in the same small size as the fury nano.

File: totally_legit.png (14KB, 601x283px) Image search: [Google]

14KB, 601x283px

>>59677848
AMD strikes again

File: lulz.jpg (343KB, 2000x1125px) Image search: [Google]

343KB, 2000x1125px

ATI strikes back

The end notes literally say April fools>>59677761

>>59677761
Man I would totally buy a RAGE X FURY.

>>59678509
No, fuck that. We need smaller cards, not bigger ones.

>>59679199
bigger card for same TDP = cooler and quieter.

>>59677925
>but a lot of the details are pretty subtle for a hoax.
that is what makes a good hoax, it's all in the subtle details, otherwise it wouldn't be believable.

>I still want this to be true

>>59677761
> RAGE X FURY

If AMD pulled that name off, I'd buy it for the sheer balls it takes to use such a silly name.

Sure this all looks great but who's willing to throw money away on something like this and hope enough people make the same mistake?

>>59677761
So they're all small factor garbage?

File: Kharn_victory.jpg (138KB, 511x753px) Image search: [Google]

138KB, 511x753px

>>59679416

I endorse this product.

>>59677848
>rage x2 fury vs 1050ti

>>59677861
What im thinkin but shit, who knows

File: plan9bunnywhite.jpg (164KB, 1376x1492px) Image search: [Google]

164KB, 1376x1492px

>>59677761
Do you see what I see?

Where is buy now button

Will these cards have any CUDA support?

>>59678083
lolol, sides destroyed.

>>59678322
>>59678244
>>59678114
ITT:
>underage too young to know about the ATI Rage or even the R9 Fury

>>59677848
>rage x2 fury
>x2

dual GPU are meme

>>>/g/sqt

>>59680230
kill yourself you fucking r*ddit nigger

>>59678116
thread/

I knew you guys were retards but this is on a whole new level.

File: APRIL FOOLS.png (9KB, 893x26px) Image search: [Google]

9KB, 893x26px

>>59677761
April Fools.

Why would a company, who is in a desperate position of competing against two other powerful companies, make a fake slide release showing that their real upcoming graphics card has performance greater than expected, when their recent graphics card and cpu releases have generally been overhyped and underperforming?

File: 1476547804738.png (323KB, 452x581px) Image search: [Google]

323KB, 452x581px

>>59677761
Oh boy, the desperate housefire watercooling is back.

>>59681391

April fools joke.

This is why Wccftech is a fucking cancer.

Any "news" that breaks in the next 24 hours is going to be bullshit.

>falling for april fools joke

https://www.youtube.com/watch?v=UtyaGppWknE

if they're releasing dual chip variants from day one it would entail that they're quite confident with the gpu maturity, power and thermals.

>>59677861
ah crap. I forgot you burgers have this awful habit of going full retard on april's 1st.

>>59677783
it's a hoax based on this slide, that is the ugliest card I've ever seen

even their machine learning cards are more attractive than that

The clocks aren't fake though, their Vega enterprise card is clocked some 1550MHz.

Instead of taking all this as fake, AMD has mingled some truths in as well.

File: ahegao.jpg (53KB, 960x564px) Image search: [Google]

53KB, 960x564px

>>59677761
>mfw told everyone that 4096SPs 1526mhz Vega was just an engineering sample for the SERVER CARD and that the consumer one would be higher clocked. And that it would be geared as a card to get 4K 60fps minimum, not average, in a large number of games.

>mfw told everyone there would be a 1200-1400mhz 4GB HBM2 model between barely under 1080 performance at around 1070 price.

>mfw correct as always

It's an April Fools, but everything points to this being roughly accurate.

>>59683778
Yeah the enterprise is 1536.

It is likely the boost on the top end cards is really more like 1650-1725.
The process these are on is supposed to clock MUCH higher than Polaris, and Polaris clocks over 1500 fine.

AMD sandbags a lot. They're probably sandbagging here yet again.

>>59677761
>AMD's april fools joke is actually just their own upcoming products
Wow. That is incredibly pathetic.

>>59680161
That doesn't make it any less edgy.

>>59683822
I don't think the boosts are that high, but then again this is the biggest change in arch since GCN so I'm hopeful.

>>59677925
>AMD HDAO 2
it is official, shit is faker than Pamela Anderson's tits

>>59683778
How would a Fury X perform at 1600Mhz? That's your baseline without any architecture improvement, since at the same ALU count.

File: 1484813255780.jpg (46KB, 500x486px) Image search: [Google]

46KB, 500x486px

>>59677922

>>59678116
>nobody saw Jen Hsung Huang mentioned
>Or some references to youtubers and Island cards


Oh AMD, you're way too good at this.

>>59683877
I thought Fury X was clocked higher, but it's boost is under 1100MHz.

Even just clocked at 1600 would make it a monster, not to mention other improvements

>>59683854
It's not just the architecture that's changed.
The die is manufactured on a different process than the RX400 series and Fiji. It will clock higher.
It won't clock to 2150 like Pascal overclocks, but "1600+" is being very modest.

Or here's a better example:
Polaris 10 36CU Radeon Pro WX 7100 is 900mhz core, 1240mhz boost.
Polaris 10 36CU RX480 are generally around 1280mhz. They overclock usually to around 1500.
But for Vega, the passively cooled enterprise server card has a 1526mhz core clock that's higher than what Polaris generally even overclocks to, if I'm not mistaken.

The likelihood is that Vega hits 1526mhz at a core voltage similar to what Polaris hits 1240mhz at.

1600+ for the high end consumer card seems a given. But I wouldn't be surprised if it's more like 1650-1725.

>>59683877
>How would a Fury X perform at 1600Mhz?
good thing it's not fury x then?
LPP though, doesn't leave much to OC if ryzen anything to judge it on

>>59683919
To be fair, either the 1080 or 1080Ti at 2050 is probably going to be similar to the Vega at 1600.
So overclocking headroom on Vega may matter. 1600+ may be what it tops out on like how the Fury X basically topped out at 1050.

My hope is, that with an overclock, they can really reach 1750+.

>>59683928
>It won't clock to 2150 like Pascal overclocks
and throttles

>>59683877
60% better. dur.

Vega is looking to be more than 2x the Fury X performance at 50% higher clocks.

So 30-50% higher "IPC" essentially.

>>59683944
Yeah, I should have said 2050. I did correct myself in a later post to say 2050 as that's more reasonable.

I don't care, though. I hate Nvidia software and drivers now days. I hate how they gimp their older cards.
My 7970 has lasted me 6 years great and is only now finally unable to keep up with the newest games without dropping settings down. Seems like a no-brainer to go with Vega as long as their $600ish card has good performance.
I don't care if 1080s price drop under $400. They'll end up being shit in 2 years anyway, if the past is any indication.

>>59683937
It's the closest thing, they're both 64CU.
Talking purely clockspeed improvement, no IPC or pipeline improvements.
Just clocked that high would make Fury X a monster.

>>59683960
980 is already shit, so yeah, fury x is still decent-ish

>>59683960
>I hate how they gimp their older cards.

i cant take anything you say seriously because of this retarded meme

any thoughts on external cache controller?
is it there just to make server racks easier to build?
their MD demo is not very convincing.

>>59683983
neglect=gimp
nvidia cards worth nothing without game per game optimization

>>59683985

PCIe hard drives acting as cache over infinity fabric.

Should be fun to see.

>>59683990
they demo it with radeon pro already last year

>>59683974
Yeah, that's what I mean.
Over time, the Fury drastically outperformed the 980. Only exception is a few Gimpworks games, pretty much.

>>59683983
But it's true.
The 970 gets 1/4th the FPS of the 380X in RE7 maxed because of the 3.5Gb memory shit. Underperforms it in just about every other game now days, too.

>>59683985
No, the cache controller helps for gaming.
It makes the 4GB of HBM2 roughly equivolent to 12GB of GDDR5X in performance (so long as a game doesn't allocate more than 8GB, I am figuring)

>>59683990
No it uses your DDR4/DDR5 as cache.
Presumably, Optane may work as well.

>>59684012
>Optane may work as well.
is a flop.

>>59683985
It's for 100GB+ datasets, not much use for us consumer shits unless AMD has some crafty software for it

>>59684017
Optane is fine for specific workloads, its low queue depth perf is great.

AMD fan here, but you're just seeing what you wanna see with Optane.

>>59684020
well, they boast +100% minimums which would be amazing if it's even 30% more

>>59684028
look, they lowered characteristics literally x1000 in one month and kept the price

>>59684040
Marketing slides have little tl do with actual products.

Fact of the matter is:
It's much faster than NAND at QD=1-10
It has more endurance than NAND, slightly.
It has equal read speeds.
It has worse write speeds.
Its power consumption is probably higher, but we have no tests yet.
It costs more.
It's first gen

Optane is fine, it's biggest issue is needing Furnace/Delidlake for no real technical reason.

>>59684028

Optane is already BTFO by the open standard JEDEC NVDIMM.

It is DOA.

>>59684075
Talk when actual products are available instead of vapor.

I'm glad there's a somewhat technical thread on /g/ for once.

>>59684017
Optane has a very specific purpose which isn't for 99.99% of people, but I wouldn't say it's a "flop".

It might actually become incredbly useful, paired with Vega.

Vega allows 1TB of memory to be addressed per GPU. In a server with 4 of them, that's 4TB of virtual VRAM you can have.
To increase thoroughput, you want lots of RAM as well to go from SSD -> RAM -> VRAM
Optane makes easier to have 4TB of fast RAM and go from storage to RAM. So this may be one of the very cases that Optane is actually good for.

>>59684030
They're likely correct, but they're talking +100% minimums when you're VRAM starved, not just a general +100% perf increase all the time.
In that example, they cut the VRAM back to 1GB or something in a game that used 4GB+, which would usually be unplayable.

So it's true, but sort of misleading.
But it illustrates how 4GB of HBM2 is probably better than 8GB of GDDR5X in most cases.

>>59684084

JEDEC JC-45.

Intel trying to force proprietary tech on the industry is always fun to watch when the industry rejects it whole heartedly and develops and open standard.

Just moving to tile based rasterization should be a pretty huge perf/watt boost, that's practically Maxwell's biggest feature.

>>59684090
>+100% minimums when you're VRAM starved
oh, didn't understand that
so basically 4/8 Gb will be much more than just 4/8Gb
didn't nvidia talked about how they did that with better compression? AMD does it with actual hardware then

>>59684124

That's part of it, but look at the double precision floating point performance of the 480 versus the 1080. It's significantly higher on AMD.

>>59684111
Products, not fucking whitepapers.

>>59684124
doesn't polaris do that already?

>>59682104
So anyone have sauce on this character?

>>59684138
It's just a difference between 1/32 vs 1/64 fp64.
Both are useless

>>59684139

We have been hearing about optane for how many years now? And when it launches it lands with a ignominious thud.

>>59684153
So you got nothing? Optane is out, for consumers its wownothing unless Intel puts out a consumer version of their Pcie caed, but for servers its a good technology in its infant states.

Jedec's standard is not even a product yet, much less somethimg tangible in someone's pci-e slot.

>>59684129
They're calling it "high bandwidth cache" instead of memory because it's so close to the GPU and has a sophisticated hardware (not software) memory controller that makes it like cache compared to how Nvidia does it.
So essentially you'll have as much VRAM has you have DDR4/DDR5, but... there's still bandwidth limitations and so on going on there that performance is going to drop when a game address more than 2-3x as much HBM VRAM that you have.
But under 2x, it should be fine.

Better compression doesn't mean a lot when there's much of the graphics pipeline that needs things uncompressed.

>>59684140
No, Polaris does not.
Nvidia first did it with Maxwell, and kept it secret up until this year when they revealed the 1080Ti. Before then, they denied that such a feature existed.
It was only first found out that Nvidia does it, and that's how they got such high perf/watt improvements with Maxwell, by an engineer looking at the die under a microscope.

>>59684146
fp32 is most commonly used in graphics APIs (even when it shouldn't be)...

>>59684178

And no one is buying in to proprietary technology for hard disk cache.

The NVDIMM is why it is DOA.

It doesn't need a new product out to render Optane defunct because it's not intel dependent.

>>59684189
Intel advises with its partners when it makes new technologies, if they said " we need something new because NAND scaling is reaching a dead end"

They made Optane products, Intel wouldn't sell the cards if no one is buying it, don't presume you know better than Intel, Dell or HP

>>59677761
>>59677783
>>59677848
>>59677925
Dammit, so if this is AMD's idea of an April fools joke, what the fuck will Vega actually be like? Or maybe the joke is on us and all of this is all legit.

>>59684184
I'm talking about DPFP(fp64), not fp32.
The former is near useless on consumer cards, at least in gamimg

>>59684244
Read the thread, the last 50 posts for some educated guesses

>>59678083
>time in India
probably misinformation on purpose, but nice try

50% increase in clocks over Fiji?
Pascal clocks were certainly impressive, but 50% impressive they were not.

>>59684244
AMD is pretty well known for their sandbagging.

The one time they DIDN'T sandbag was with the Fury. They ended up having to put stock clocks nearly max due to the 980Ti. But they get to come in after Nvidia playing their hand and Volta being a year away, this time.

I'm not sure these are AMD slides, though, are they? I think wcftech made these.
But if that's the case, their guesses look pretty spot on. They're roughly what I've been noting for a while.

It should be clear to anyone with the slightest bit of technical literacy that the card they've been demoing with 12.5TFLOPS is the server card, and the high end consumer one will be higher clocked. It also seems very clear that there was going to be a 4GB HBM2 card priced more around the 1070.

>>59684184
>Nvidia first did it with Maxwell,
okay, I kind of get now why AMD made no volta joke in reveal
fiji already was a monster raw performance wise

not to give in to the hype again first priority now.

>>59684012
optane won't work because it's DRM'd to intel newest cpus

>>59684313
They can still use intel CPUS with Vega GPUs.

Not enough fanboy crap ITT

Being clocked that high would explain why AMD is taking its sweet time with Vega, since its IPC is far higher than Nvidias those clocks would have a far bigger effect.

>>59684361
A month or two ago there was a leak of vega at 1100Mhz being slightly faster than a 1080,so yeah, pretty much an ES.

>>59677869
>4GB > 11GB

Hahahaha no. Try playing Deus Ex with that VRAM

>>59684393
There are software and firmware techniques to mittigate the need for a large amount of VRAM, you should be glad that we'll start smartly using what we have instead of dumping PCB, size, power budget and money at more VRAM.

>>59683928
>P10 overclocks to 1500Mhz
Yeah what on fucking LN2? Nobody talks about ultra-niche unusable methods when they talk about consumer cards.
Even a typical XFX GTR has a hard time passing 1450Mhz (a 70Mhz OC from the stock 1388) as it is now.
Maybe, maybe the "P20" 5xx variants on 14LPP+ will be able to reach ~1500 on average.

With that said, the (minor) clockspeed gains from a better and more mature 14nm process are going to be diminished unless these new P20 chips are equipped with GDDR5X@10Gbps. P10 gets just about the same gains in performance from RAM overclocks as core overclocks. Nobody actually knows if Polaris had been or is now designed to handle GDDR5X. The 580/570 cards are gong to be stop-gap tier fillers until 2018 if/when AMD can finally release a mostly new GPU stack top to bottom. I suspect this is why the expected prices for the P20 based chips are lower than their original release cards - and why I think AMD has not added GDDR5X (or expensive 9Gbps GDDR5) to the design.

If any of you are on semiaccurate forums ( I know at least a few of you guys post here on /g/) feel free to put this into the Polaris thread.
Wish I could have an account there.

>>59684393
that's what they demoed with the new tech.

>>59684448
580 will be 1340 out of box, so who knows

>>59684393
So you prefer that a GPU hit a hard wall when it's VRAM starved instead of being able to run great despite being under the VRAM requirements for a game?

So once games start using 12GB of VRAM, enjoy that $700 1080Ti being completely obsolete while 8GB of HBM2 still works fine because it can cache in and out from your RAM instead?

Deus Ex is exactly what they demoed to show that tech, by cutting the VRAM down to 1GB or something.

Your post makes no sense. I assume you just come to such conclusions because you're retarded.

I'm guessing Nvidia is delaying Volta because it's a new arch as well, like Vega.

>>59684486
volta supposed to be scalable like navi
it is whole new can of worms, nobody has any idea how would both companies going to pull that off

>>59684471
so now 16GiB of RAM will not be a meme anymore?

File: 4gb.png (58KB, 653x409px) Image search: [Google]

58KB, 653x409px

>>59684393
Wtf are you on about you retard?

>>59684498
No, if anything the median amount of VRAM in high end GPUs will either stand still or decrease in the next few years.

>>59684496
Simple:
Both companies have been developing new interconnects for a few years now.

Nvidia will probably leverage some variant of NVlink scaled down for multi-chip comms, AMD is certainly going to enhance and expand upon their Infinity Fabric.
The difference is nobody has a damn clue if Volta has anything new at the table, while we all have a pretty good idea of what AMD is able to put together, albeit we have no idea of the differences between Vega and Navi.
>probably just minor improvements in instruction set, utilization and compression
>Navi will be on 14nm, though probably 14LPU allowing an extra ~150Mhz clock speeds over Vega

>>59684486
Volta hasn't even taped out yet, we're looking at Q1 2018 at soonest if it tapes out in the next 2 months.
Realistically something between q2 and q3 would be a better guess, especially if they're gonna use TSMCs 12nm

>>59684559
Some people were actually expecting a full Volta lineup this summer.
How delusional, or simply ignorant.

>>59684539
>Navi will be on 14nm
might be 7nm, navi is 2019

>>59684587
Some people here are dumber than rocks, they're usually the loudest tooo.

I'm hoping theres at least a 1070 equivalent in the lineup, ideally undercutting the 1070

AMD has a massive plus of vertical integration, when the start making profit they should leverage it, CPUs with fixed function blocks that can run critical GPU driver code faster, as an example.

>>59684498
Well. 16GB of DDR4 I'd say is the minimum.

But it's still hard to recommend 32GB due to prices.

I got lucky and got 2x16GB for $135. Hoping I can at least overclock it to like 2666 CL15. :/
I'll buy proper low latency DDR4 3200 2x16GB later when prices go down, or I end up upgrading to DDR5 in 2-3 years.

>>59684667
That would be the 4GB model, which seems likely to exist.

I'm hoping for a Nano successor that still has 8GB HBM2 but is just lower clocked or something. I don't care if it's $600 or whatever.

Weren't Nano's the best binned chips and overclocked well?

>>59684496
I don't think Volta is going to be scallable like that. Probably what comes after, like Navi.

The current assumptions are just that Volta will change architecture and bit and use HBM(2). But Nvidia is secretive, and surprises due to that sometimes, so who knows.

>>59684559
I really doubt Q1. Q2 at the earliest. If it was Q1, there'd be engineering samples and some sort of clues floating around.

>>59684539
>we have no idea of the differences between Vega and Navi.
I think we mostly know what Navi will be.

Same as Vega, same HBM2, same NCUs, same almost everything except:
The die will be broken up into multiple pieces for both better yields and better scaling.

It's IPC will likely be lower due to the slight latencies incurred there, but there will be massive cheap "dies".

Considering the paper AMD engineers published about their ExaFLOP node APU...
It seems what they plan to have there is simply complete GPU dies. So while it's an APU, the CPU and GPU are separate dies on the same socket.
So they have a 16 core Zen CPU in the middle, and 8 GPU dies and stacked memory surrounding it. Separate dies, but on the same socket.

It's basically like Crossfire, or a 295x2 but it looks like a single GPU die to the graphics API.

>>59684618
We still have no idea if Vega is LPP or not, which is odd. It has to be something different, considering its clocks, right? I was under the impression that Polaris clocks are silicon limited, not architecture limited.

>>59684706
Doesn't work when people use compilers that optimize for intel.
If compiled programs simply took advantage of AMD's 4 complex instructions per cycle that it can do (instead of 1complex+3 simple of intel) Ryzen already gets higher IPC than Kabylake.

I'm surprised AMD doesn't bitch about that more. It's just not good marketing to play victim, I guess.

>>59684709
Nanos were underclocked FuryX, and you can drop a FuryX's power consumption by 70 watts by lowering the stock voltage and keeping the same frequency, while being stable.
Funnily enough nobody ever mentions this.

>>59684725
GCC with current lofty zen support shows Zen performing much better on Linux than Windows, even in multithreaded stuff.
Llvm should be little to no different.

But as you say Intel has control with ICC, even Ashes uses ICC, and it's supposed to be AMD sponsored

>>59684727
Ah okay. So better binned than Fury, but the same binned as Fury X.

So you could overclock the Nano to the same as the Fury X? Yet the Nano was smaller, and just didn't come with that (honestly ugly) AIO?

Yeah I really hope they come out with a "Nano2". I'll buy that. You'd think they would, unless Nano didn't do well?
ITX is so huge now, it seems like a no-brainer.

I'm tempted to get the x2 card, but I'll wait for Navi which should have the whole "x2" power without the drawback of it only working in games that support Crossfire..


I was strongly considering the Nano at the time, but decided it wasn't worth it over my 7970 which was a non reference card that would clock stable up to 1325mhz over the 1000 stock boost. :/

>>59684759
Yeah, better instruction scheduling for Zen should improve things another 10-20% over what it's at now on Linux, too.

But AMD has always allowed 4 complex instructions per clock, haven't they? With programs that can take advantage of that, it can amount to over 50% higher IPC over Intel. But everything is just compiled for Intel on Windows.

That's aggravating me, though, that games just do one compile. Oh, and fucking Windows, too. I'm going to be using Linux for my 1600X, but it's not like I can compile most games from source.
If they're not going to compile for the system, they should at least have multiple compiled binaries and install the optimal one, for fucks sake.

Why are CPU manufacturers forced to be in charge of power efficiency when computers would be more efficient if things were compiled properly for the CPU?

>>59684782
Maybe someone will make a cloud compiler and installation packager to do just that.
Like 'oneclick' but not shitty.

That'd be pretty huge.

>>59684706
The ironic thing is OEMs will probably sell AMD/Nvidia machines because Nvidia are still seen as "the" GPU brand to get

time to sell my 1080??

I hope the actual top tier card doesn't come forced with a shitty AIO cooler

More waste to throw away when putting it in my custom loop...

>>59684828
There's software to port Cuda to OpenCL and shit now.

>>59684861
Doesn't work properly

I'm stuck with nvidia solely because of CUDA

Might as well make peace with it and get a goysync monitor...

>>59677761
GOD FUCKING DAMNIT IT'S WATER COOLED AGAIN

>>59679368
if its a hoax then amd done fucked up bad unless its more powerful than what they are doing, any thing hoaxy would just make people pissed they aren't actually doing it or aren't actually that good.

>>59683876
you do know amd bought ambient occlusion to games right?

File: 1484952748297.png (33KB, 420x420px) Image search: [Google]

33KB, 420x420px

>>59677761
> /g/ is taking this REALLY seriously

Everyday I feel like everyone is shiposting and baiting on this board, it's impossible THAT many people are retarded at the same place

>>59684452
didn't they cut it to 2gb of vram?

>>59684910
Speak English. Nigger.

>>59684932
can't see what you can't comprehend. do i need to speak to you like a child?

>>59683876
AMD is the one that invented whole thing fyi.
HDAO1 runs better, but I think HBAO+ looks better

>>59684931
to 1Gb

>>59684855
That's another reason to hope there is a Nano model that will overclock to the same. Heh.

>>59684930
It's being taken seriously because it roughly lines up with expectations of anyone technically literate that:

-There will be a lower clocked 4GB HBM model priced around the 1070.
-There will be a lower clocked "nano" model.
-The 1536Mhz clocks are for the server engineering sample, and the real high end consumer 8GB HBM2 card will be 1600+ mhz.
-There will be an X2 card (though I'm not expecting it on launch, and I think it'll be same clocks as the "nano" version")

>>59684931
>>59684962
I don't recall if it was 1GB or 2GB. Anyone have a citation and not hearsay?

>>59684515
No, the vram will likely stay the same or get bigger, the only difference being once a game that demands let's say 16gb of vram comes out, the old cards won't shit the bed like they would now if you tried it.

You would prefer the vram but your gpu isn't hard limited if you don't have it, its a value added kind of deal rather than a fuck it now we can decrease ram. It does alow the low end cards to have less and less ram though, that will be a huge plus to entry level, or even apus.

File: af.jpg (56KB, 680x382px) Image search: [Google]

56KB, 680x382px

>>59678116

>>59684978
its an hour-2 hour long presentation that should be on amds youtube, I don't want to scour the damn thing right now to find it, reading thread. if someone hasn't done the leg work by the time im done, ill do it.

>>59680230
maybe maybe not.
If amd really does make BETHESDA games better by having a dev work on them, then why not crossfire when amd is clearly better at it then nvidia when it works?

you want 4k 144, well there is only one way to get it, multigpu.

>>59684981
Yeah. Just 1GB or 2GB of HBM2 on an APU will be insane.

Next year, we might actually have APUs that are as powerful as a 7700+RX470 is (if AMD will bother to put 16-20 CUs in a consumer iGPU) for the price of what one of those were alone.

>>59684990
I know, I saw it. I just don't recall if it was 1GB or 2GB, and I don't want to look again.

>>59684987
>JHH

>>59677922
amd is fairly good with marketing, this would get them a solid week of speculation, with everyone thinking its fake, then the reveal shows OH SHIT IT WASN'T, apposed to if they if they said it at any other point and everyone just said it was fake because 'amd can't beat nvidia'

This way one release gets them multiple ripples, kind of like the ryzen price, they told us the cost in december but speculation kept them in the news for weeks. then at ces, more weeks.

>>59678195
shit, just what I said.

I'm still thinking fake unless they fixed their gpu usage.

>>59685012
>you want 4k 144, well there is only one way to get it, multigpu.
not really

SLI = stutterfest
Crossfire = still not enough HP

>>59685065
look at adoreds rise of the tomb raider bench he did with crossfire 480's, now imagine that done with cards that are actually powerful.

>>59685084
>now imagine that done with cards that are actually powerful.
imagine being the operative word

Vega is launching in May or even near the end of April.
If I was buying a highend GPU I'd wait another month.

>>59685094
I have no doubt that vega is at least a 1080 at its worst, at least the uncut version, amd would have to either fuck up, or get fucked hard for it to be worse, and I don't believe they did, or will.

>>59685104
I think that anon just means "imagine" because Vega isn't actually out yet.

>>59685102
I just hope they actually unveil something about it by the Ryzen 5 launch...

I can't help but hope/wonder if they pushed the Ryzen5 launch data back just to be closer to Vega. Since it seems BIOS are mostly getting fixed now, and the CPUs exist in large numbers, but retailers are just being told not to sell them yet.

April 10th announcement for the launch, specs, pricing, and a late April launch would be amazing. But late May sure looks more likely.

>>59685104
The 1080 is like 30% faster than a Fury X, AMD can leapfrog that with clockspeed increases alone.

People who think the 1080 is competition for a highend Vega are absolutely clueless on how this works.

>>59685127
we know a vega is said to be 12.5 flops, at amds worst they needed to be 33% more powerful to match nvidia, that puts it just above a 1080

polaris current efficiency is around 25-22% to match nvidia, that puts vega closer to titan xp then it is to 1080

to beat the 1080ti they need 17-15% performance lost.

Amd seems to realize they have a gpu that is underutilized, and have made great efforts to fix that, low and behold, it may pan out, but im expecting no more then uncut to just beat out a 1080

>>59677761
>RAGE X2
nvidia on suicide watch

>>59685127
Good point, Fiji boosts to 1050mhz, while 1080 boost is somewhere around 1800mhz.
For having that huge difference in clockspeed, the 1080 isn't all that much faster.
Now before someone screams IPC, AMD has plenty to grow in the clock department, while I don't expect Nvidia to go much higher in clocks than they do now due to physical issues with the silicon.
The enterprise vega being clocked at 1530mhz is a very good sign of the performance.

>>59677795
>It is got a brain
damn pajeets

>>59677761
>RAGE X FURY
Need to fire people in their naming department

>>59685160
This is probably why Volta is nowhere to be seen, Nvidia can't fight on clockspeed alone, they need to get their stagnant IPC up.

>>59684393
hey just add moar rams and increase the price.
are you in marketing?

File: file.png (916KB, 1723x1431px) Image search: [Google]

916KB, 1723x1431px

>>59677995
>hitman

>>59685143
12.5TFLOPS is the passively cooled 180W server card.

The consumer gaming card is likely going to be higher clocked to deliver 13-14TFLOPS.

>>59685160
What? The 1080 is way faster than the Fury X.
The Fury X is about 80-95% the performance of the 1070, which is significantly lower than the 1080.

If you mean it's not 75% faster despite having 75% higher clocks, okay that's true. The 1080 has less cores, though.

>>59685177
Nvidia's IPC hasn't increased since Kepler.
Not a very forward thinking action at all, which is weird because the situation was reversed from 2007

>>59685222
no, unless amd says anything about power, I am assuming this is polaris all over again for gamers, where the higher binned chips all go to embeded and pro gpus and consumers get the shit that needs to be overvolted to get enough samples working.

We know polaris chips that needs 160 watts to run, and we know there are chips that only need 90 to do the same fucking task. we got the 160 watts till late in the game and xfx gtr oc came around as the only one to get binned gpus, at least at the time.

>>59685222
https://tpucdn.com/reviews/NVIDIA/GeForce_GTX_1080/images/perfrel_2560_1440.png

Some 20 games test, can't really say it's that much faster than a Fury X.. Or 980ti that's 73% of its perf, while clocked almost 75% more

>>59683988
>>59683983
look at the halo wars 2 beta that had the 780ti above a 980 and tell me why the 780ti usually performed below a 970 and in some cases a 960

>>59683877
about 45% better on paper, but fury x could not saturate itself to save its fucking life, for all we know that 45% better number could be 70-90% better if vega can fully saturate itself.

>>59685246
To add to this, while the 1080 has less cores, it has far more ROPs, which are fixed function units are generally more important than pure ALU.

>>59684378
potentially lower clocked buy up to 22%
potentially thermal throtteling
potentially just working good enough to not hard crash

>>59677761
>these names
wtf amd

>>59685277
ROP count is the same IIRC, but directly comparing the shader pipeline with numbers like that is nonsense since you're not taking into account the size and complexity of the shaders, which historically Nvidia had larger so by nature, less units.

>>59685276
Both a ROP/L2 size issue on top of lack of VRAM.

Fiji was full of bottlenecks, compared to that GM100 was quite balanced.

>>59684539
well, we can look at it this way with navi. 8gb of vram cost 80$ at launch of polaris, so remove that, you now you have the gpu that costs 160$ assuming literally everything needs to be doubled/trippled/quadrupled besides vram

240$ gets you mid range gpu, bin lower gets you low end
400$ gets you a 1080 preformance
560$ gets you the titan xp/1080ti killer
720$ gets you better then nvidia can produce

this is assuming alot of things, like amd not developing a smaller but easier to mcm gpu that scales better, assuming that they don't have 1 bran box chip and multiple retard without the brain render dies, so on so forth.

If amd makes a brainbox die, and then makes retard dies, they could scale those VERY easily allowing many skus with 2 slightly different setups.

>>59685334
oh no shit, the worst one was the 20nm process falling through, and needing to go back to the 28 with a 20nm made gpu

the reason vega is so interesting to me is amd seems to have solved most if not all the issues.

>>59685342
They aren't making an HBM2 card for $240, not even a 4GB one, you dumbass.

That's what the RX500 line is for.

>>59685362
is your reading comprehension that fucking horrible?

>>59685352
Now that I think bout it Vega is the first AMD highend GPU Raja gets to work on without any fab or time constraints, Fiji had huge promise but those fucking bottlenecks on top of low shader utilization.

What's with this civil technical thread deprived of fanboy cancer?

>>59685378
if, and big fucking if, amd has a good driver team work ground up for vega, and they manage to get the flops to actually mean their performance, then nvidia cant touch them, but i have my sights on this.

the worst amd ever was was 33% more powerful to match nvidia, vega is at a 1080 there
25-22 is where polaris is
and to match a 1080ti they need 17-15%

This this indian motherfucker get an extra 7% efficiency out of the new arch? Did the driver teem not fuck up horribly, find out in t minus 3 months.

GUYS IT WILL BEAT THE 1080ti AND THE TITAN X PASCAL

JUST WAIT...

NVIDIOTS BTFO

>>59685398
Flops are pure arithmetic math capable by the GPU, it doesn't work for pixelpush workloads where the job is spread through the memory controller, ROPs, TMUs and shaders.


But I get what you mean flop/real perofrmance should improve since this is the first time AMD acknowledged the issue and was a major point in their vega slides.

>>59685445
FLOPS does matter way more now days than before, though. That's why TFLOPS performance is increasing so greatly year after year with both GPUs.

Games are so shader heavy.

>>59685445
Flops are pure math and what they can do, if nothing else bottlenecks it, that is the tell of performance, but there are bottlenecks fucking everywhere.

>>59685454
the reason its getting that way is because the only way to make perceivably better graphics is to make them orders of magnitude better.

see ps3, it had Resistance and then in Resistance 2 they doubled the poly count, but everyone had to be told because you could not tell otherwise.

File: krypto-mainimage.jpg (49KB, 731x718px) Image search: [Google]

49KB, 731x718px

>>59677761
BUT CAN IT MINE KRYPTO AT MINDBLOWINGLY HIGH SPEEDS?

>>59685454
As said, you need balance to push 4000 shaders through the ROPs in games, and 64 rops aren't gonna cut it, for compute rops don't matter, so tje full flops of the GPU are used.

>>59685481
what year is it

>>59677761
Wasn't Nvidia going to release HBM3 with Pascal?

>>59685512
It's never too late to mine cryptocurrency when your electricity bill is fixed if spend less than 20,000 watts a day.

>>59685531

Nvidia never backed HBM - they backed HMC (which was a competitor to HBM) but JEDEC decided on HBM thus Nvidia ditched HMC and went with GDDR5X.

>>59684709
>>59684667
that 4gb if it has the efficiency of polaris is a 1080 competitor.

>>59685738
The "1070-priced" Vega with 4GB HBM2 very likely will be better than the 1080 in some games, worse in others.
If not always better.

I don't really understand how people doubted that. AMD has been working on this since Fiji. Polaris was a minor stop gap to get something out before then. This is the biggest leap forward in GPUs since GCN 6 years ago, if not bigger. (Maxwell was pretty big too, though)

>>59685013
ok, caught up and found the part of the video

https://youtu.be/bDl6xJJqIAU?t=21m3s

2gb

>>59685429
it does stand a good chance.

>>59677761
>watercooling again

Fucking dropped. What a load of garbage.

>>59685847
>again
You know you could have just bought the R9 Nano and overclocked it to Fury X speeds, right?

Don't be stupid. They sold a watercooled and non-watercooled version.

>>59685803
Ah okay. So that's good, but maybe could have been better.

>>59685893

>R9 Nano

The nano is really the unsung hero of the 28nm process. Its effeciency (i.e power draw vs performance) is possibly the highest of any 28nm card ever made and shows how good binning can give surprsing results. Its the same sort of deal the R7 1700 has - take a relatively minor hit in clockspeed but a huge drop in power draw.

I bet the retarded netizens will still believe that this time it's really gonna be overclocker's dreams.

And then there are those super empty-brained guys who believe that those cards with HBM2 will have a better price/performance than Nvidia.

Fuck off

The only argument for AMD is gonna be the freesync1/2 support.

>>59677925
>Doom having less of an comparative edge than the other games makes sense if Vega really does have a binning rasterizer, since idTech 6 already does spatial binning in the host software side.
what? post source or gtfo. you probably made these slides yourself

>>59677848
>it's their top tier watercool-only card vs a GTX 960

Time's up.

>>59686063
APILS FOODS IT'S JSUT A PRANK BRO HUEHEUEHUEHUEHUEHUE

>>59685342
What this is assuming is the ability for all GPU chips on the MCM to share a single pool of VRAM. Perhaps the HBCC is a step in that direction.
Much the way we see Ryzen link L3$ to IF to IMC, perhaps Navi would link HBCC to IF to VRAM.
Each chip's L2$ would still be directly linked to the VRAM pool assuming Navi isn't a vast departure from Vega, although that would get very messy scaling beyond 4 chips.

>explanation for those who don't understand
Otherwise, with the way GPUs are now designed...
Each chip would have it's separate VRAM pool, by necessity because each chip could only house a certain number of IMCs.
So like current SLI/CFX RAM would be limited to what each GPU can connect to, else the chips would shuffle massive amounts of data through each other in complicated crosslinks that eat up efficiency.
Not only are these options completely batshit they would be detrimental to all performance and efficiency metrics.

With all this said, we need to see small Vega to know what Navi could potentially give.

>>59685362
We're talking about "designed for scalability" Navi, released late '18 or early '19.
Navi is expected to be, instead of very large chips, smaller GPU dies connected together using AMDs new stuff we see on Ryzen and Vega.

S-so is it april's fool or not guys I'm confused

>>59686128
yes

amd said vega is """"just around the corner"""" (which means several weeks/months) as late as march 27

>>59686018

> this time it's really gonna be overclocker's dreams.

Oh look, another person who never actually watched the event and listened to what Joe was saying. Since i'm in a good mood I will educate you. The conversation with Joe Macri at the time was about the pcb design for the fury x - not the actual chip - and the reference pcb is, infact, an overclocker's dream (see: buildzoid's breakdown - the board can take, way more abuse). The issue is the chip has nothing more to give.

Now since listening and understanding is fucking hard y'know (all dem words be tuff) people saw the fury x, heard "overclockers dream", put 2 and 2 together and got 5 and ran with it as a headline.

>>59685893
No you couldn't. It's common knowledge that the nano throttles like a bitch.

>>59686202

It throttles to maintain its power draw limit, not because of thermals. Once you up the power limit (which goes all the way to +50%) it'll fly along at fury x clocks all day (note I did not say it would do it quietly - that little fan needs soem serious rpm to cool the chip).

>>59686161
>""""just around the corner"""" (which means several weeks/months)
If it was weeks we would have actual leaks.

They need time to process the R5 launch as well even if it's not the same division of the company, it's still all "AMD" and big launches very shortly after one another aren't good from an investor relation perspective.

I would assume R5 launch in mid april (well that's confirmed), and Vega announcement/paper launch end of June, with regular availability in early to mid August

>>59686221
Why don't they just put another extra thermal pad/gel between the heatsink and gpu? Yes, above the grease

>>59677783
>1600mhz
>when the Radeon Instinct Mi25 is 1540MHz

Doesn't look like an april fools joke to me.

>>59686270
>with regular availability in early to mid August

>literally over a year since Pascal launched and AMD launches a slower card

J-J-JUST WAIT

>>59686272

Wouldn't do anything. Generally speaking the best way to get a fury-x from a nano is set a custom fan curve so it ramps up earlier but not as aggressively.

>>59677783
look at the jagged edges and the overall unappealing aesthetics (the angle etc), some retard made this in gimp, not even photoshop

File: 1489090612227.jpg (61KB, 1280x720px) Image search: [Google]

61KB, 1280x720px

>>59677761
THIS IS NOT A LEAK YOU FUCKING RETARD CANT YOU READ?

>CONFIDENTAL UNTIL APRIL 1.

ITS APRIL 1 YOU MONKEYFUCK.

>>59686188
>talk about hypothetical stuff, when in reality it is not the case

No matter who spit it out, it's a stupid statement.

Nobody is interested in abusing the board, it's about the whole package, fucking hard to understand.

>The issue is the chip has nothing more to give.

Which makes the statement invalid and really dumb.

>>59684393
They demo'd Vega running Deus Ex on Ultra with 2gb vram

>>59681391
The joke is on Nvidia, the GPUs will be way better

>>59686335
Well it's not like there's much demand I'd say.

A 1070 will demolish almost any modern game on 1080p and a 1080 will demolish almost anything on 4k if you don't mind playing on high instead of ultra while AA is useless for all but the biggest 4k screens anyway.

Even the 480 can probably do anything you want on 1080p.

The only thing AMD needs is a 1070 and 1080 competitor and better drivers

>>59686471

Yes, just like the Fury and Polaris series.

If that is your understanding of "way better".

File: 1487463821177.png (2MB, 1440x900px) Image search: [Google]

2MB, 1440x900px

>>59686046
You know they mean an equal MSRP card.

AMD is not that misleading.

>>59686079
Then yes, for Navi, we might see a much cheaper cards instead of this split like with Fiji and Vega.

>>59686128
>>59686317
It's an April Fools but also close to reality.
>>59686391
Yes, that's been pointed out. Over 90% of the posters in the thread realize that, but people keep pointing it out anyway.

There was like 10 posts of people realizing it's an April Fool's but simply discussing that it simply looks right and just talking about Volta and Navi in general before the first person did a retarded
>LOL DONT YOU FAGS SEE ITS AN APRIL FOOLS
post like you.
We're 200+ posts in but here you go.

>>59686581
fury was/is better than gtx 980

>>59686581
Yes. Like how Fury was better than the 980, and became better still with Driver updates.
And like how the 480 is better and cheaper than the 1060, and the 470 is better and cheaper than the 1050Ti.

That's exactly what people are looking forward to. I'm glad you understand.

>>59686847
The difference is I've been playing on a 1080 for more than half a year now while you're STILL waiting on a card that might or might not even catch up.

>>59678025
It's very well made, but fore example Radeon RX Vega Rage X Fury is vey too long to work as name.
The render of the card looks slopy.
And look at the End Notes slide.
Plus I think it is significant that the article has disabled comments.

>>59686949
Yeah, you get 10% more FPS than a 2 year old 980ti.

Congratulations.

File: 1476574884037.jpg (86KB, 736x770px) Image search: [Google]

86KB, 736x770px

>>59684143
Yun from New Game.

>>59686079
But what improvements will Navi have to get better performance?

>>59686949
The difference is that I didn't need to buy a 1080 because my non-reference 7970 that overclocks amazing and that I bought 6 years ago still plays every game fine.

Whereas everyone who bought an Nvidia GPU that long ago has been forced to upgrade since they were shit.

File: 1481454591018.jpg (119KB, 796x805px) Image search: [Google]

119KB, 796x805px

T-T-THIS IS NOT REAL, THIS IS A BAD APRIL FOOLS JOKE AND FRANKLY I DON'T THINK IT'S VERY FUNNY

>>59687432
lol fuck off retard, you can't play max settings 1440p in new games like gtx 1080 can, you have to play at 1080p and turn the settings down just like people with older nvidia cards to get 60 fps

>>59687708
>can't run 1440p 144fps
Cool. I have a 1920x1200 monitor that's mainly for work.

And no, on most games I hardly have to turn the settings down.

I was running For Honor at over 60FPS maxed out. That's a new game.

>>59687708
Don't lie, you play at 800x600

>>59687708
>lol fuck off retard, you can't play max settings 1440p in new games
when the fuck did this
>LOL UNPLAYABLE UNLESS 4K60FPS MAX SETTINGS FINAL DESTINATION
come from recently? It's so fucking retarded.
I swear this wasn't a thing back in the days

>>59685481

Apparently it has a soul. So you must first convince it that it wants to be a miner.

>>59687798
>7970
>for honor
>maxed out

Why are you lying? There's literal video evidence of oc 7970's on YouTube barely managing over 30 fps maxed out at 1080p and yet you're magically getting over 60 fps at 1920x1200? Fuck of shill. Nobody believes your lies.

File: 1479618237612.png (325KB, 522x593px) Image search: [Google]

325KB, 522x593px

>>59687881
It's all shills have to justify their new overpriced purchases against how well high end GCN GPUs aged.

This is the life of an Nvidiot. Buy new GPU every year. Why do you think Nvidia sells more GPUs than AMD even during the times when AMD GPUs were objectively better? Because a large number of their customers buy a new one every year due to how poorly they age but they still think Nvidia is the best.

>>59687928
>Why are you lying? There's literal video evidence of oc 7970's on YouTube barely managing over 30 fps maxed out at 1080p and yet you're magically getting over 60 fps at 1920x1200?
Literally none!
... except for the first result when you google "for honor 7970"?
https://www.youtube.com/watch?v=YIMRN0_oc3o

Considering he averages about 40, and I get about a 35% overclock over stock on mine, seems right.

I also didn't play those gay ass babby mode NPC fighting maps. I played the 1v1 and 2v2 like an actual man.

Enjoy your 10fps in 6 years with your card, tho.

>>59687928

>Believing JewTube
My FPS never drops below 60 with the 7970.

>>59686545

AMD has way better DX12/Vulkan drivers. Recently shown that XF480s will beat a Titan in DX12 in RoTTR. Nvidias DX12 drivers are broken or half assed.

>>59687131

I think its a joke, but isnt. A real teaser on performance packaged to look like an April fools joke. Everyone laughs it off, then once the card is released it actually is a monster.

> I hope its this because I own AMD stock

>>59687881
pretending that your 7970 is on par with a gtx 1080 is fucking ridiculous. they're completely different tiers

File: 1489789244241.jpg (33KB, 300x392px) Image search: [Google]

33KB, 300x392px

>>59687983
did you even watch the video. he's in the 30's for 90% of it. are you claiming 7970's can oc to gain 100% performance now? get the fuck out of here.

>>59688108
but he never did?
he simply said that 7970 still holds up just fine

>>59688137
he's saying that with nvidia gpus you're forced to upgrade but with 7970 you aren't. that's inane bullshit. 7970 is still comparable to gtx 680 depending on the game
>Whereas everyone who bought an Nvidia GPU that long ago has been forced to upgrade since they were shit.

>>59688171
there is some truth to that, as AMD's drivers matured way better than NVIDIA's, not to mention the infamous gimping of older generation cards

>>59688186
nicely meme'd

https://www.youtube.com/watch?v=yuMZxXlX9ys

>>59688171
it's not though. the ghz edition is, not the regular 7970. HUB tested this literally 2 months ago and found the 680 to still be slightly ahead of the 7970 ghz.

>>59688129
I said
>I also didn't play those gay ass babby mode NPC fighting maps. I played the 1v1 and 2v2 like an actual man.
>mine is overclocked

Why are you posting still instead of killing yourself?

>>59688207
my sides

amd tards on suicide watch

File: 1459295212343.jpg (63KB, 600x1200px) Image search: [Google]

63KB, 600x1200px

>>59688217
>b-b-but muh 6 yr old gpu can reach 59 fps once every 10 minutes in this non intensive game mode!
>gtx 1080 owners BTFO!!!1!!1!!!1


aww there's no need to be mad

Man, I'm torn. Because I upgraded from a 7950 to a Fury X, I got a massive performance boost and gained the benefit of Freesync. However, I pre-ordered the Fury X from the very first day and got mine on the very first day (thankfully it was a revision 2 card, not that it matters now its on a open loop...)
Now VEGA is coming out, a part of me wants it, badly. More performance, especially for gameswork heavy titles like Fallout 4, would be fucking lush.
At the same time, I don't really have the spare money to waste on an upgrade right now. I could, but I shouldn't.

So part of me obviously wants AMD to do really good, because even if I don't upgrade, it provides better promise for the future.
A part of me however, looks at it as if I had been willing to stick to 1920x1080~60 and my 7950, I could've instead chosen to upgrade to Vega.

Woe is me.

File: IMG_20170401_165928456[1].jpg (2MB, 1920x2560px) Image search: [Google]

2MB, 1920x2560px

>>59688284
>Living the life of a true manpatrician streamer
Glory to the old girl

>>59685165
>It has got a brain

>>59688207
I don't know about that one, but I saw the Babeltech one from last year
They cherry picked older games when it's the newer games that are using the superior compute of the 7970 that has allowed it to hold up well.

Here's the list, for anyone interested:
>Serious Sam 3
>Wolfenstein
>Left 4 Dead 2
>World in Conflict
>Crysis 1
>Far Cry 2
>RE 5
>Just Cause 2
>Battleforge
>Alien vs Predator

These are 7-10 year old games. And this was from a 2016 article. I wouldn't be surprised if the other one is cherry picked.

Now if you take a list of actual new games and compare them (without comparing them with a 4c/4t on the AMD part and 4c/8t for Nvidia, that'd be nice too), the 7970 Ghz comes on top. Most 680s sold with 2GB of VRAM, too.

>>59688284
If you're on 1080p, there probably isn't much point to Vega.

>>59688375
I'm on 2560x1440~144 w/Freesync.
What I meant was that if I had stayed at 1920x1080, I wouldn't have needed to move on up from my 7950. I was happy lowering settings to suit, so long as the game still looked reasonably good.

>>59688375
the HUB video tested new game though and still came to the same conclusion as the 680 being slightly better.

>>59688384
Ahh, I see.

Well I want to move up to 2560x1600 or 3440x2400 or whatever, so I'm waiting to see how Vega performs.

>>59688423
Ah hardware unboxed.

I don't know how my experience with the 7970 and of many others I know that have one is so different.
I do have a non-reference card that had top binned dies and a different PCB that overclocked better.

https://www.youtube.com/watch?v=BfSi-Z8r12M

Seems like the 7970 is better ...
The times where the 680 is significantly better are games like Overwatch where both get over 100fps anyway.
But when things are lower and more fps matters more? 7970 is faster or barely behind.

And it seems like my results are much better than these benchmarkers in most games. And it seems to be the experience of so many 7970 owners that they feel they still run things fine and don't need to upgrade. Subjective, but whatever.

>>59688689
AMD is for poorfags so they're more likely to be content with sub-60 framerates and lower graphics q. nvidia cards are still doing "fine" by your standards but nvidia users are more likely to upgrade to 1440p or even 4k and be able to play the newest games at a solid 60+ fps or even 120+ fps with max/high settings

File: old man huang.jpg (25KB, 475x324px) Image search: [Google]

25KB, 475x324px

>>59688893

I endorse this post.

april fools, and WHO >>59688888

>>59677761
> 8 gb
into the trash it goes

>>59687983
the very first number giving you the >60fps is actually the temperature

>>59689154
it taps from system ram, look into high bandwidth cache

>>59688893
R9 Fury is great for 1440p, tho. It does better there, proportionately, than it does at 1080p.

And Vega is right around the corner.

You seem to not be making a point at all there.

>>59689154
Would help if you actually read and knew anything.

>>59689459
kek. I have an artic accelero on it.

>>59684244
It's made by Currytech, not AMD.

>>59683886
priceless picture

>>59677761
The last one doesn't even have the pcie interface in the right spot. Its in the middle of the card for some reason. This is fake.

>>59677869
That can't be real because it's not how vram usage works

>>59686035
the slides are fake as fuck, but that logic is plausible.

There are a couple presentations Tiago Sousa did about the idTech 6's renderer structure, and one of the most substantial features it has is slicing the scene into screen-space volumetric tiles (logarithmically sized in the Z-axis) that it uses for draw call batching.

The upshot is that shaders have great cache locality even when there is no hardware-assisted rasterizer tiling/binning, so GCN systems including consoles can run things well. Vega claims to add binning rasterizers, so just like D44M runs proportionately better on GCN compared to Maxwell/Pascal, a Vega GPU would be expected to show larger improvements in game engines with pre-tiling.

>>59691486
Can you not fucking read?

The "memory controller" in Vega works like a cache controller and can use your system memory as VRAM and the HBM2 more like cache.

Due to the speed of it, having 4GB of HBM2 is probably going to be similar to 8-12GB of GDDR5X.

This should have already been apparent with how well Fury scaled to 1440p. With the driver updates, it often beats the 980Ti in 1440p when it was priced around the 980 and handedly beats it.

>>59691623
thanks. i was interested in knowing more about it but i couldn't find what anon was talking about on google. do you think Tiago Sousa talks about it here?

https://www.youtube.com/watch?v=4mr5uezENlg

>>59692214
haven't seen that talk, but here's a slide deck from one of the presentations:

> http://advances.realtimerendering.com/s2016/Siggraph2016_idTech6.pdf

>>59692687
sweet, thanks anon

>>59683900
le ebin references redditbro xDDDD