More sandbagging? AMD had claimed for the longest time that

really makes you think

>>60773193
it was the raw throughout that holded back the gcn their polygon drawing was fine

are they giving shader engines the bulldozer treatment?

>>60773193
further analysis:
It appears it could be the geometry processors and ROPs as the wide rectangles at the top and bottom of the 8 shader engines.
So it would appear the instructions are coming in from the outside, then finishing down into the middle, instead of the top-down of GCN.
It also appears there are 2 geometry processors and ROPs per shader engine.

But.. again that doesn't make it clear whether they consider this 4 shader engines or 8.
I do think they could be considering this as four, but with the data splitting up into multiple geometry processors on either end for whatever reason.

It also looks like the global data share is on one side, and L2 on the other side, and there is over double the area for asynchronous compute engines and the global data share is fucking massive.
Bottom is obviously memory controller.
Top right appears to be hardware accelerators.
The middle between the Shader Engines... I have no fucking clue.

>>60773341
Na it was not enough ROPs for the compute that meant lots of CUs would be idle as there was much more polygon drawing work needed compared to compute work.

>>60773354
More like the Zen treatment. They can more easily cut this down for mobile and smaller GPUs.
GCN was fixed at 4 shader engines and you had various degrees of lopsidedness with ROPs and CUs.
Same reason you get issues where the RX580 is hardly better at 1080p than the RX570, but is much better at 1440p in screenspace heavy games.

amd's strategy is to clearly blindside their competition

keller already shitstomped intel and theyre in a complete panic

amd knows exactly what failed with fiji

theyve poached the best parts of maxwell/pascal and are about to get a comparable performance from FLOPs with respect to nvidia

>>60773370
Oh. It could actually be the geometry processors and ROPs in the middle, and the skinny rectangles above and below them could be the async compute engines.
That actually makes more sense because the CUs should be short and wide.
And looking at it that way, I do see 4 pairs of geometry processors+ROPs.
But if that's the case, the gap between the row of CUs (looking at it turned 90 degrees) makes no sense.

>>60773370
it was the culling process that was bottlenecking gcn
they literally didnt had one for a long time and tried to force everything to pass via the ACE to parallerize them so that they can gain much perf
didnt really end up well for them

>>60773376
infinity fabric will be used on vega..
i think that amd will clock the memory lower than expected just so that they showcase the full potential of it

File: cook_hero.png (359KB, 575x657px) Image search: [Google]

359KB, 575x657px

So did this asshole buy all the Vega from AMD? Is this why we've been waiting?

>>60773488
its not the first time tho..
apple always buys the higher bins every single chip they can get and its top notch they will pay insane money for it

>>60773474
Yes. Lack of culling really hurt Fury. But more polygon thoroughput on top of culling doesn't hurt.

amd always have multiple lineups of preformance chips

>>60773496
Aren't the higher bins not saturated until later in the production? I mean yields improve over time right?

>>60773488
he is fighting with powerpc and intelligent right now so he probably will run to amd lol

lets hope he gets a apu with a seperate igpu and hidpi screen with display port and freesync

>wishful bsd dev

>>60773542
they used the same technique as ryzen we dont know the yields but im pretty sure its way too high

>>60773193
Dont regret the wait now. Shouldve never doubted AMD

I just found out that nvidia can only use 1 of the wayland compositors

amd supports 7+

once again nvidia wants to create their own shit instead of a standard protocol

just like fucking egl
reeeeeeeeeeeeeeeeee

Yeah and? Its still slower than a 1080

>>60773602
>reeeeeeeeeeeeeeeeee
Just sell it second hand.

>>60773636
>Yeah and? Its still cheaper than a 1080
>Yeah and? Its still freeer than a 1080
>Yeah and? Its still less gimped than a 1080 in 2 years

>>60773636
>>60773636
Fury X at 1600Mhz can't be slower than a 1080.

Stop shitposting.

>>60773193
Didn't this happen with Maxwell, splitting the SM into smaller units for better utilization?

>>60773193
https://www.youtube.com/watch?v=owL_KY9sIx8

this looks really promising. Vega will deliver.

>>60773193
>More sandbagging?
I doubt it. If they had something impressive they'd show their hand by now. Simple as that. They have nothing to gain at this point by Sandbagging. VEGA is going to disappoint you if you think AMD are still hold back.

>They still consider it 4 shader engines, but each one is split in two groups
This how they arrived at the 2.75x increase in the first place obviously. Pulling an extra 2x out of your ass is pretty desperate.

It's a cool die shot though, you can see all the space dedicated to the infinity fabric logic pretty clearly.

If it has the same tflop/perf as Tahiti, it will be a monster.

>>60773781
>they'd show their hand by now

No, because they learned their lesson on what happens if you over hype something with Ryzen.

>>60773920
You meant Polaris and Fiji.
If Vega is half a good arch as Zen is, it would be a miracle.

>>60774028
Vega is what Fiji failed. considering Fiji's perf after seeing Vulkan-DX12 and is LN2 overclocking numbers, I am very hyped for Vega.

Someone fire up paint and mark the hardware components there, please.

What's that between the shader array? Looks quite significant in size.

But it's still only 64 rops, bottleneck....?

>>60774221
see>>60773776

Is this design more scalable?

Has AMD got another winner on their hands? Is this their RV700?

>>60773193
I want release dates and prices

>>60774572
July.
500$+ starting.

>>60774562
>Is this their RV700
it seems so.

>>60774591
You telling me low end Vega is 500?

>>60774610
What lowend Vega? Thats months away, you got big vega and bigger Vega now.

>>60774610
we don't know the line up.
we know there is 56 CU Vega (big vega) and 64 CU(bigger vega) Vega thanks too Apple. no price, nothing.

>>60774647
So and is pulling a fury/Titan meme release? We're they rebrand a third time for everything else?

>>60774562
I feel this is more Tahiti than Terascale, Tahiti was really good, could OC like mad, no bottlenecks.

im pretty happy with all their offerings but i hope they have some lower power consumer stuff with aes and iommu for muh servers

intel only cares about the enterprise

>>60773193
????
Where's the AMD BTFO DEAD BANKRUPT line?
No..No. You're not are you.
No way.
No way, get fucked.
No way get fucked cunt, fuck off.

You can't be seriously having a serious GPU discussion on /g/ of all fucking places.

>>60774695
Fuck off fag, this thread is pure cause summers fags are asleep

>>60774079
agreed
vega vulkan is going to be great

>>60774700
Hold me as I weep tears of joy.

How does geometry throughput of polsris/fiji compare to Pascal? I think I saw techreport doing the tests, but fuck me If i can find em while at work on a phone.

File: deutsche.keltic.federation.png (1MB, 1124x711px) Image search: [Google]

1MB, 1124x711px

>>60774713
bask in thine glory for it is short lived

inb4 it ends up like other amd cards with high theoretical power but it will still be shit at gaymen

>>60774720
Around two to three times lower.

>>60774742
And so summer awakens

>>60773193
>trying to interpret anything from this """"""die shot""""""
uhh
>AMD had claimed for the longest time that Vega had 4 Shader Engines
no they didn't
in fact they basically said that it would work over more than 4 engines
see: http://www.anandtech.com/show/11002/the-amd-vega-gpu-architecture-teaser/2

AMD just needs some good value chips to retain marketshare, it can curbstomp Nvidia once it gets enough money from their CPUs

>>60773602
Yeah. I don't understand when people say AMD support on Linux is bad. Maybe if you're on any early version 3 of the kernel or earlier.

>>60773542
Now days dies usually sensors on them to easily test and bin themselves very quickly and easily.

>>60773636
Even a Fury X with a 50% clock increase beats the 1080 by a lot. Vega is much more than simply just a 50% clock increase.
Retard.

>>60773739
A lot happened with Maxwell. Maybe. idk. I don't follow Nvidia architecture much because they are so secretive and you often don't figure out features until YEARS later of reverse engineering.

>>60773776
Yes, I disagree with his analysis, if you read the OP.

>>60774221
IPC per ROP is 275% higher. Also don't know if it really is 64 ROPs.

>>60774797
Not really. Halo products sell the midrange. Consumers are dumb.

Way too much arch changes to gauge performance off a 1600mhz Fury X.

These frontend changes can be anything from 10% to 40% increase in games alone, the latter for particularly shitty ones lupike Fallout and Dishonored

>>60774831
No one is doing so really.
It's just the baseline minimum performance increase without arch changes would be ~50% over Fury. But obviously it's higher than that.

The cherry on top would be if it can actually overclock.

>>60773920
>they learned their lesson on what happens if you over hype something with Ryzen.
How is saying a 40% IPC increase but delivering 52% overhyped?
Unless you're one of those people that listened to the shills claiming the world about it just so they can pull this trick, that it was 'overhyped and didn't deliver'.

Well, Samsung shat out a very impressive quantum dot freesync monitor a few days ago, makes me think there's some meat to Vega.

>>60774902
Haven't most of Samsung's monitors these days been FreeSync supporting?

>>60774902
http://optocrypto.com/2017/06/05/samsung-chg70-first-hdr-monitor-freesync-2/

File: $R1B3XXA.gif (80KB, 350x263px) Image search: [Google]

80KB, 350x263px

>>60773193
WHO WAS IN THE WRONG ?

WHAT DID THEY MEAN BY THIS ?

REALLY MAKES YOU THINK

>>60774931
Average quality though.

This one is 144Hz, quantum dot, VA panel with 3000:1 static contrast, and freesync 2, meaning HDR10 and 125% sRGB coverage.
If it was 4k, it would have been literally perfect.

And will probably cost a good $700 at least.

>>60774974
Wat is this quantum dot meme

>>60774974
Doesn't it still have black shift issues since its a VA panel? That stopped me from buying a CF791.

>>60774974
Personally if I had to choose between 1440p with that contrast and HDR and 4k, I'd pick the contrast any time of the day, most 4k high refresh screens are still around 1000:1

>>60773193
>175%-275%
Hnnnnggg, give me some budget Vega cards for that sweet altcoin mining

>>60775005
Wait for tftcentral reviews.
I don't think they'd be advertising it as a premium gaymer monitor as is, and IPS has its own issues.

>>60775028
anon what altcoins are worth it?

>>60775073
I don't know any current ones, but there are always coins worth mining. I sold my mining rig a good while ago, I now only trade crypto, but I'm looking into mining again.

You can hype and use fancy marketing crap as much as you want, it still won't change tje fact that it's barely faster than a 1070 a year later

>>60774676
No, it's literally RV700 2.0, this time Pajeet made Fiji actually work.

>>60774654
No, it's probably Q4 2017-Q1 2018 for Vega11.

>>60775399
Hello, fuccboi. AMD gave you the die shot explicitly to shut up niggers like you.

>>60775514
unfortunately those fuckboys will be chimping out until we see vega.

then they go REEEEEEEEEEEEEE

>>60775399
You can keep crying as much as you want.

So what's the term 'sandbagging' in this context.
>4 but actually 8
Sounds good?

>>60775702
It means presenting something worse than it is, probably to slow down the competitors response time.

A good example is Zen, they promised 40% IPC, we got 52%.
Sandbagging.

>>60774974
>>60774984
Good question, what is it? Sounds like a marketing gimmick

>>60775734
Its just the name of a specific type of pixel structure used in the display.

>>60775734
http://www.anandtech.com/show/10243/qd-vision-color-iq-and-the-philips-276e6-review/2

>>60775733
Thanks. I was confused as in today context it's to stop flooding waters. so comes off as preventing the inevitable.
Like they're lying to themselves that everything going to be alright.
If anyone wants to drop origins of the phrase you're more than welcome.

>>60775702
It's exactly that. Everyone assumed Vega still had 4 geometry engines, while in reality, as seen in the dieshot, they went for more balanced approach.

>>60775778
Thought so. More is usually better. Any good resource that would help the layman look at dieshots and understand what they're looking at. Lurking moar can only get one so far.

>>60775774
My guess it's alluding to using sandbags as jettisonable weight. So sandbagging would be intentionally downplaying upcoming products so you can then "jettison" the "weight" holding you back and shoot ahead of the competition, taking them by surprise.
That's just my understanding though

>>60775858
This tactic won't work much longer, Nvidia knows what RTG is capable of, and they're not slowing down.
Unlike Intel who has their head so far up their own ass they only noticed after AMD ate their entire SKU lineup overnight.

>>60775858
That's a very good explanation and also reverts back to the beginning of the technology era where the balloon was invented. I'm pretty sure I've herd that expression used in papers at that time.
Not that I'm of that age. I've researched many things.
Often turns of phrases can be deduced with logical thinking.
Thanks again.

>>60775920
>they're not slowing down.
Ye, they bolted fucking ASIC on oversized die in desperate attemp to get into datacenter market.

>>60775943
Fixed function and specialized hardware is always a good thing.
Pretty much everything in large scale is moving from general purpose to specialized hardware.

>>60775961
That's why everyone will move away from GPGPUs to FPGAs and ASICs. GV100 move reeks of desperation.

>>60775920
>Unlike Intel who has their head so far up their own ass they only noticed after AMD ate their entire SKU lineup overnight.
Their management, yes. But please don't label their engineers as being complacent in status quo. Well, not because they don't strive for better.
We're living in times where such a company has political persuasion. And aiding their products gives them power.
That's not to say a smaller company does not have the ability to persuade government. It's just some do it for worse outcomes.

>>60775976
Intel has one good engineering team: the dudes behind EMIB. That's it.

>>60775028
The window is already gone. Ethereum will be mined up before the end of the year. Card prices skyrocketed back up with the BTC boom this year.

Buy a 1070 right now today and it might pay for itself. A vega card would never pay for itself. Frontier edition will be 1000USD, 30 MH/s, 250W calling it now.

>>60775988
Their NICs are good and Optane is really good if it wasn't so unholy expensive.

>>60773193
They don't want Nvidia to start selling Titan Xp for 400$ before Vega release.

That's why they are pretty much airtight about it's performance except computational applications.

>>60776033
>Their NICs are good
Ah, yeah.
>Optane is really good
Fuck no it dies overnight. And we're talking R&D.

>>60776033
Optane was codeveloped with Micron though.

File: 1496550384158.png (123KB, 240x286px) Image search: [Google]

123KB, 240x286px

>>60775971
>away from GPGPUs to FPGAs and ASICs

GPUs and FPGAs aren't ASICs?

>>60776067
>modern GPU
>ASIC
2007 called.

>>60776060
First gen issues, NAND was pretty pathetic when it started out.
Optane has very low latency and amazing IOPS at low queue depths, that is great.
But it's still too young, give it time.

>>60776085
>just wait
Brian pls. You promised magical super-memory and it's literally worse than current NAND.

>>60776075
GPUs and all modern silicon has ASICs, you got the DSPs, Tensor cores and fp64 in Nvidias case, also count as ASICs, UVD and other decoding/encoding hardware.
That's all fixed function hardware.

>>60776116
ROPs are fixed function too? ASIC?

>>60776153
ASIC literally means application specific integrated circuit.
What do ROPs do? Mesh texels and pixels.
They're not general purpose like shader/ALUs

>>60776103
Hyperbole, you're acting like having equal sequential speeds, far more IOPS and lower latency and slightly better endurance than NAND ia a bad thing for a first gen product

Xpoints only misses are its price and immaturity.
And yes I agree they promised the moon and fucked it, but nobody sane expected their claims for fucking non volatile storage.

Aww, the die show was fake.

https://videocardz.com/70162/amd-and-nvidia-preparing-graphics-cards-for-cryptocurrency-mining

Even Nvidia wants in

Reminder to reporr shitposters.

>>60776338
Ye, shame the die shot was fake.

>>60774898

Because 10% less performance in gaymes means the chip is worthless. You should know better by now.

>>60776384
Go away IMG_XXXX

>>60776397

Shush anon, don't you know? Any benchmark where Intel™ loses doesn't count. Now be a good goy and buy the latest industry leader Kabylake-X™

>>60776474
But anon, what about nVidia™?

>>60776297
If AMD cards are so good at mining, why doesn't AMD mine?

>>60776514
anon, do you understand how margins work?

>>60776514
How do you know they don't?

>>60776533
whole 570/580 lineup sold out for mining and no restock in sight

>>60776497
Nvidia is fine, if they supported FOSS and had little less goy practices like GTX drivers shutting off any VMs, they'd be great.

>>60776598
But what about Goyworks™?

I have been waiting for vega for a long time and meanwhile 1080 prices has sunk from €650-750 to 500-550... I'm getting really tired of waiting, should I just get a 1080?

Power is important to me, it needs to run on a 500w like a 1080 does.

Or will I feel like the people that bought kaby and didnt wait for ryzen? I waited for ryzen and im incredibly happy I did.

>>60776622
>Power is important to me
Vega uses tiled binning, has larger L2 and L2 <-> ROP assocation.

>>60776611
It's a good framework for Nvidia GPUs, I guess? Not that Vega will have any trouble with Goyworks thanks to its geometry throughput

File: I can't even, even if I could.png (273KB, 640x480px) Image search: [Google]

273KB, 640x480px

>>60776598
>less goy practices like GTX drivers shutting off any VMs
Why am I only hearing about this now.

>>60776622
>buy from amazon
>when vega releases come up with a bullshit reason to send your card back
>enjoy your full refund because amazon is too lazy to check if your card is actually broken

>>60773193
that shot is tinted, right? color gradient looks too clean for just interference.

>>60776685
It's just a styled picture of the actual die, dies don't look anything like that.

>>60776685
It's fake.

>>60776640
https://warosu.org/g/thread/S60125659#p60127495

It's not a big issue but really tells what a company thinks of its users.

File: Screenshot from 06-06-2017 15:49:09.png (34KB, 599x533px) Image search: [Google]

34KB, 599x533px

I guess something good happened.

>>60776726
apple happened.

>>60776637
What does this mean for me?

Vega frontier edition has a 300w TDP, which would mean I need to replace a perfectly good 500w PSU and probably get a case with more fans.

>>60776679
As you can see from the € sign im a europoor, and no amazon here. The companies are pretty anal about returns

>>60776753
You don't know TDP.
Nobody but AMD does.

>>60776753
Vega FE has <300w TDP, which is actually ~225W due to it working on 8+6 pins.

>>60776753
>Vega frontier edition
it is a 1000 bucks card. you are not going to buy that.
yo would have to see cut down versions. late july.

File: 1496328708889.png (745KB, 1260x673px) Image search: [Google]

745KB, 1260x673px

>>60776726
Stop capping your pointer. I've seen you? even recap an image that had your pointer in it posted in the same thread. So you could repost said image in argument without tripping the previously posted image filter.
Pic related was the image you decided to recap.

File: Screenshot from 06-06-2017 15:58:03.png (69KB, 860x450px) Image search: [Google]

69KB, 860x450px

>>60776793
I don't really care if someone identifies me, I'm not hiding anything.

>>60776760
here's some sites that claim 300w
http://wccftech.com/amd-radeon-vega-frontier-edition-radeon-pro-vega-ssg-detailed/
https://www.nordichardware.se/nyheter/gamingversion-vega-snabbare-an-frontier-edition.html

>>60776760
If it peaks around 300w I'm still fucked.

>>60776767
But will they be cut down in TDP? They also claim the gaming ones will be faster (for games)

>>60776497

But goy-I mean valued customer, why settle for lesser brands when you can have the latest technology of the Intel® HD Graphics™ 630 paired with our Intel® Core™ I7 7700K? Don't forget to also get an Intel® Optante™ drive for all your fast storage needs...yes, it's fast, what idiots told you otherwise?

>>60776825
>wccftech
Oh wow. Please don't do that.

>>60776825
>But will they be cut down in TDP
gaming one will be faster than frontier in games, and probably need the same wattage with frontier.

And now im yet again hooked at WaitTM and i almost bought nvidia, fucking Ayyyemd. I hope they will have enough units in shops.

>>60776825
Those are not very good sources, TDP is still unknown, and if it fit into Apple's shit, it sure as hell ain't 300W

File: 1485421775097.png (66KB, 421x247px) Image search: [Google]

66KB, 421x247px

>>60776823
No, you care. Because you didn't want to be found out so soon. You wanted Anons in future time to ask back at why so many images contain a black hand pointer.
It was your master plan after all. To insight mouse equality. That no one truly cares about the color of their pointer.
I've stifled your crusade, Black Hand Capper. once and for all!

>>60776892
I would have gotten away with it if it weren't for those meddling kids.

>>60776726
Cause apple bought all the vega

>>60776847
I know. But we don't have anything else.

>>60776850
That's what i'm thinking.

>>60776881
They are also claiming a 18 core xeon that's gonna be at least 140w. I dont think 300w would be impossible, and apple usually doesnt hesitate to let their GPUs at max temperatures and throttle them. or they could be using a cut down version.

amd is 480 and 580 is more like 210w in practice, could vega really be as efficient as pascal or better when polaris was so far behind?

>>60777001
Wut? Undervolted Polaris is king of perf/watt.

>>60777001
The lower clocked version of 580 is 65W.

>>60777001
nvidia doesn't have hardware scheduler, that's why they draw less power than amd counterparts.

but polaris was made for lower power draws, and works amazingly efficient at lower wattage.

>>60777026
And it's also, comparatively very weak.

It's relatively easy to make a low performance chip power efficient, the hard part is making something that beats a 1080 or 1080ti and is still efficient.

>>60777050

What "version" exactly? even rx 460 is 75w

>>60777104
>What "version" exactly? even rx 460 is 75w
new gaming laptop of Asus for example.

>>60777119
Well that's cool in theory be doesn't help my desktop.

>>60777104
>the hard part is making something that beats a 1080 or 1080ti and is still efficient.
That part is piss easy. Just enlarge L2 and introduce tiled binning.

File: ASUS-ROG-STRIX-GL702ZC-AMD-Ryzen-and-RX-580_1.jpg (241KB, 1616x1080px) Image search: [Google]

241KB, 1616x1080px

>>60777104

>>60777104
Polaris is engineered around cheapness.
Everything about it is cheap.
It's whole big promise was 1080p 60fps for 199$

Vega is their take at top end rather than sweet spot.
Vega's motto is 4k 60 fps which is way above what 1080 has to offer.

>>60777168
Vega11 will substitute Polaris Q4 2017/Q1 2018.

>>60777197
I'm really eager for this one really. hbm2 in these cards would be fucking great.

>>60777219
It'll probably use GoyDDR6.

>>60777228
don't think so. hbm is partly amd's ip. also using gddr6 would need changes in architecture.

File: 1484334953923.png (418KB, 600x802px) Image search: [Google]

418KB, 600x802px

>>60776910
they're not kids they're animes

>>60777249
>changes in architecture.
Resuing Poolaris' memory controller is not that hard, pal.

>>60777251
Go back to Germany you slut

>>60777219
Vega 11 will go in APUs and bottom end gaming cards.
I bet it wont run on HBM

On other side I expect it to deliver above polaris performance in an integrated solution so 1080p and maybe even 1440p on a single APU.

But it's fucking 2017, about fucking time to bury 1080p

Vega die shot look like a door

>>60777272
>about fucking time to bury 1080p
Wat? 1080p is still perfect for 13-14 inch laptops.

>>60777282
It's not a die shot, it's a diagram.

>>60777134
Desktop Polaris are the worst bins. All the good ones go to MBP and embedded.

File: AMD-Fiji-GPU-Die-Shot.jpg (142KB, 787x659px) Image search: [Google]

142KB, 787x659px

>>60777282
This is a die shot, that's merely a render

>>60777272
>Vega 11 will go in APUs

Isn't Ryzen mobile a single die?

But since Vega has IF, it might be possible to get Vega11 + Zeppelin MCM.

>>60777351
>Isn't Ryzen mobile a single die?
Yes. It still has Vega11 CUs on die.

>>60777292
>13-14 inch laptops.
Garbage.

>>60777418
Shit taste.

>>60777418
for you. it would be god send for my work.

>>60774974
>2560x1440
Looks nice, but so far I'd rather get the C34F791. Hopefully they release an updated 3440x1440 monitor before this holiday.

>>60777362
It's anywhere from 11 to 16, we don't know.
Ryzen mobile is unusually large chip at 210mm^2, the CPU portion(cache and uncore) + memory controller take less than 80mm^2, so a lot of space goes to the GPU

>>60777495
Freesync 2 or HDR10 support?

>>60777510
RR is supposed to have 768 VEGA shaders, which means 12CU.

>>60777520
I'd love both desu, and I'm considering waiting for a proper 3440x1440, 100Hz+, HDR10, Freesync 2, IPS/VA monitor.

Any word on if Freesync 2 supports borderless fullscreen?

>>60777539
Supposed, that's old slides and things can easily change considering RR is probably not far off from its intial tapeout

>>60777549
I think it was part of the ReLive update.

>>60777558
We'll see. I also want to actually see it in the laptops. Because Intel loves bribery.

>>60777346
I see, plz post render for Polaris and Fury

>>60777598
that's fury.

>>60777598
That one is Fury.

This is Polaris
http://images.teraknor.net/polaris10/Die%20Shot%20-%20AMD%20Polaris%2010%20-%20Annotated.jpg

>>60777637
Man DDR memory controllers do take a lot of space.

>>60777699
No shit, 2 HBM controllers are not even a third of all those 64bit PHYs put together

>>60777713
Fuck DDR. Low-cost Samsung HBM when?

This got me thinking, what would Vega at Polaris die size be? Performance is competitive as is, but perf/watt is terrible, and clearly Polaris can't even begin to clock as Vega does.

I think I'm gonna be really surprised by midrange Vegas

>>60777734
>would Vega at Polaris die size be?
It'll look like beefire GP106. Less shaders, more performance and higher clocks than Polaris.

File: my prayers have been answered.jpg (37KB, 518x120px) Image search: [Google]

37KB, 518x120px

>>60777578
Thank you, that was literally my only issue with Freesync.

From googling, however, it seems like many users are having some issues with it not working properly. Hopefully I won't have that same problem.

>>60777734
knowing that they will in q4 2017/q12018, I would expect mid range vega to perf similar to gtx1070.

>>60777734
Polaris is a weird fucking chip, it has similar perf/watt as Fiji, which I don't have to remind you is almost 3 times its size and still on 28nm.

>>60777785
Vega 10, 4096 shaders.
Vega 10 cut, 3584 shaders.

Vega 11.. 2816 shaders? (I hope)

>>60777812
>Vega 11.. 2816 shaders? (I hope)
That would be around the performance of a 1080, and more than potent enough fend off the 1070 Volta replacement.
Especially since AMD isn't a stranger to price drops.

>>60777812
Considering GCN is inherently denser than Tesla-derivatives, they can allow themselves to throw more shaders.

>>60773193
That looks extremely weird, like there is backlighting behind the chip. This is a fake image and fake news, isnt it? Someone made this in photoshop didnt they?

>>60777812
Anything smaller than 44 CU leaves a large gap and the 1080 without a competitor.

>>60777884
It's a layout diagram, not an actual die shot. They even colored different parts of the chip differently. Find the HBM PHYs.

>>60777884
It's in Apple product pages.
It's a fucking render of the die, not a real die shot.
Real die shots look like the fucking surface of the moon

>>60776760
>~225W due to it working on 8+6 pins.
Don't forget the 75W you can leech from the PCIe slot.

>>60776767
>it is a 1000 bucks card. you are not going to buy that.
I'm tempted to buy one if it's a semi-pro thing like the Titan used to be.

>>60777908
Vega Frontier Edition seems similar to the Titan, but - unlike nVidia - AMD are upfront about the fact that Vega will be both cheaper and better performing in games.

I'd like to remind you guys that the previous iMAC GPU(tonga based) had a TDP of 125W

Just some foot to chew on, I personally wouldn't believe Apple would double the TDP for a single part in that tiny chassis.

>>60777935
A*ple said they redesigned cooling. It'll throttle like crazy and sound like a jet engine anyway.

>>60777947
They always say that, physics are still physics and you're not gonna cool 450W+ in that fucking AIO with fucking air.

>>60777894
>>60777903
So it's basically marketing material then. Am I unfair if I say that only literally idiots would try to guess things about Vega from some shitty picture some sales assistant fiddled together in photoshop to look cool?

>>60777935
Vega in the new iMac is clocked something short of 200MHz less(11tflop) if that thing is 125W, AMD has a Zen on their GPU, 160W is what I wager though.

>>60777961
It looks similar to Fiji but has a distinct difference of twice the shader engines.
So no, this image isn't just for kicks, you think Apple doesn't know how the Vega die looks like?

>>60777961
It's a die layout picture. Like, we've alreadt seen GV100 layout.

> "In the in the Ryzen, Ryzen processor is just as an example, we've launched seven versions of the Ryzen processor to-date and if you look at those seven versions relative to their Intel competitor that's most nearly priced competitor you'll see we're bringing anywhere from 30%, sometimes over 70% performance advantage. And when we launch our Vega graphics processor that's coming very shortly, it will come later this month, you'll see similar level of competitive performance in high-end part of the desktop or high-end part of the graphics market."

https://seekingalpha.com/article/4078955-advanced-micro-devices-amd-management-presents-stifel-nicolaus-2017-technology-internet-and

t. Jim Anderson

>>60778043
30-70% faster than 1070 for $450 confirmed! NVIDIA IN THE ICU

Seriously though, AMD should really price its shit cheaper at equal performance, they don't have Nvidia's mindshare.

File: best girl.gif (2MB, 480x270px) Image search: [Google]

2MB, 480x270px

>>60778043
>30%-70% better performance for the same price or cheaper

>>60778043
Can AMD really undercut Nvidia as they do Intel?
This is the question, AMD can live with 35% margins, so probably, it could just be a way to get marketshare on low GPU margins while they sell their $40 CPU silicon for $500, where the real margins are.

Complicated game they're playing

>>60773727
To the extent that the bottleneck isn't ALU throughput, it sure can be.

>>60778166
There's a FuryX at 1450MHz image floating around, it scores between a 1080 and 1080ti

>>60778187
I don't think anyone has actually managed to overclock Fury X to 1450 MHz. Are you sure it's not just a stupid extrapolated image?

>>60778258
It's on Ln2 run, anon.
It's the best baseline of performance we have.

>>60778258
ln2 possible

>>60778122
They will undercut nVidia to get market- and mindshare.

>>60776941
So it's gonna be Tonga all over again, where they scooped up the top binned dies and the chaff gets sold off in the AIB options for consumers?

T-thanks, Apple...

>>60778292
The iMac is only releasing in December, I doubt it has that much of volume that it would leave AMD dry of chips.

Actually, that would be better for AMD, they sell everything they got.

>>60773193
If AMD can actually pull off anywhere close to 250% perf of Fury X, I'll immediately upgrade from my Fury X.

>>60778283
not with mining on the rise again

huge drop happened with 970 for AMD when mining was big and 290 got sold out everywhere
it's happening again

>>60778270
>>60778268
Cool. Was the memory also overclocked, though? That would certainly skew the results.

>>60778316
I don't expect Vega to be that good at mining per buck, its largest improvement is in the frontend, which is generally useless for mining.

>>60778330
Yes, from what I remember.
There's also a issue of Fury X's memory controllers topping out around 350GB/s instead of 512GB/s due to the bug, so who knows if that memory overclock improved much.

>>60773727
>>60774827
We already got test benches of Vega, in Time Spy it is lower than a 1080.

And it doesn't really matter because Nvidia will just do a price cut and they'll have Vega beaten in price/performance - and still make a profit, due to the absurd markup they can put on their cards.

>>60778316
You pay for better frontend with Vega, though. Miners don't care and buy Polaris.

>>60778342
out of curiosity do miners undervolt for better long run perf/watt?

>>60778353
Yes.

>>60778351
>We already got test benches of Vega
What Vega?
>>60778353
Of course.

>>60778351
>We already got test benches of Vega, in Time Spy it is lower than a 1080.
there is three consumer vegas fyi and two enterprise

>>60778351
I seriously hope you're not looking at 1200MHz Vega engineering samples from late 2016 as an example.

>>60778379
Those benchmarks were really popular shitposting material a few months ago, it's amazing how people think something Fury X with 50% more transistors than a Fury X is as fast as Fury X.

>>60778409
>expecting intelligence in /v/ermin-infested GPU wars threads
Oh no you don't.

>>60778418
But this thread seems fine.

>>60778440
I don't really know why. Did 8 geometry engines scare /v/ermin so much?

>>60778452
Nvidia's largest advantage is its geometry units, Vega nullifies that.
And from what I know Nvidia still loses in texture and shader throughput

>>60778452
They only understand all-caps and cherrypicked benchmarks.

>>60778470
Goyworks is pretty meaningless now, except for HBAO+ i guess. They should probably open-source it or merge it with GPUopen.

what's consumer variant predicted ETA?
Would be nice if it's september.

>>60778498
Late July launch

>>60778497
HBAO is good
I have to say AMD HDAO look worse for some reason

>>60778504
I thought it's only for frontiers
confused

>>60778521
frontier comes in late june

>>60778498
Launch on SIGGRAPH with worldwide availability in August. I sincerely hope all cryptocucks would kill themselves by then.
>>60778511
I think nVidia should merge Gameworks with GPUopen since Vega will process Gameworks effects roughly as good.
>>60778521
No. Vega FE launches late June. RX Vega (and it's flavours) launch at SIGGRAPH 2017.

>>60778452
/v/irgin here, this thread is fascinating. I really hope Vega will actually deliver, though.

>>60778540
It will, just like RV700 did. Pajeet usually delivers.

>>60778533
I don't mind minerfags it's good if you're invested in AMD. What AMD should do is make minor exclusive cards, they increase margins if they did and also meet demand for the gaymers.

>>60778590
Miner exclusive card means Instinct-lite with no video outputs, which means no resell value and cryptocucks will continue to buy consumer cards.

Also fuck nVidia and
FUCK GAYFORCE EXPIRIENCE 3.0 WHY DO I HAVE TO REGISTER TO USE FUCKING HARDWARE VIDEO CAPTURE?
t. 780ti cuck

>>60778636
Fuck GeFEX, just install your drivers manually.

>>60778677
Which i did, but the lack of Shadowplay is not fine.
Why the fuck ragtag team of Pajeet can deliver better driver suite than fucking nvidia is besides me.

>>60778379
we don't know what clocks the game cards will come.

The only thing we know is AMD quoting 4096 shaders and 25TFLOPs (I think, or was it just 22?), and people are making assumptions on the clocks based on that.

>>60778707
25 tflops at half precision.

>>60778707
It's 13tflop FP32 (which the games use) with 4096shaders clocked at ~1600mhz. The main improvements in Vega are in the front-end. That's why you can't properly guess the performance.

>>60777026
>>60777055
>Undervolted Polaris is king of perf/watt.
>polaris works amazingly efficient at lower wattage

You are so cuteeee, it's like to say pascal works amazingly efficient at lower voltage...Polaris is less efficient than Pascal, deal with it

>>60778771
Undervolted Polaris is the king of perf/watt, anon.
>it's like to say pascal works amazingly efficient at lower voltage
While delivering no compute performance. GCN is inherently more efficient when it comes to tflop/watt while operating at optimal clocks, deal with it.

>>60778771
yeah, pascal doesn't have hardware scheduler.

>>60778788
It also uses tiling of some kind (probably rasterization).

>>60773231
what'd he mean by this

AMD stocks skyrocketed again. Why though?

>>60778958
apple. please read the fucking thread.

>>60778782
>Undervolted Polaris is the king of perf/watt, anon.
Citation

>>60778688
There are a lot of good alternatives to Shadowplay. As for drivers, it's almost like nVidia wants to fuck over their customers so much it's impacting their ability to develop good software.

>>60776474
how? is it already 2018?

>>60778707
Flounders Edition(Workstation) is at 1600MHz, and that's the aircooled version.
Raja said that the gaming card will be faster and clock higher.
So it's 1600+

File: AMD%20Embedded%20GPUs%20Launch%20Presentation-06_575px.png (907KB, 955x537px) Image search: [Google]

907KB, 955x537px

>>60778986
See pic.

And the 65W TDP 580(full 2300 shaders) clocked at 1200MHz ASUS is releasing in in the Ryzen laptop.

AMD has been selling scrap to gamers for a while, while relegating good bins to embedded and partners.

Vega is the first gaming-oriented AMD release since forever. Vega10/11 have no 1/2 FP64.

>>60777104
>What "version" exactly? even rx 460 is 75w
pro and oem/embedded get the pick of the bins while gaymers are left with the scraps that need more voltage to run well

eg: https://www.techpowerup.com/gpudb/2882/radeon-e9550-mxm
full-fat polaris 10 at 95w tdp

>>60779172
>And the 65W TDP 580
citation

>>60779299
read the goddamn thread>>60777136

>>60779299
>AMD’s cards have been infamous over the years for running fairly hot, but the Radeon RX 580 should be OK on this front. It features the full desktop complement of 2304 cores and 1200MHz base clock speed, all on very low TDP of 65W.

Why would I make this shit up? Did you even follow computex?

http://www.game-debate.com/news/23001/first-amd-ryzen-laptop-surfaces-as-computex-asus-rog-strix-g702zc

>>60779321
>ANUS sells some quality workstation hardware as gaymen laptop
Why.

>>60779347
A 580 and 1700 aren't workstation hardware.
The 1700 could be used as one, 580 isn't even the Pro version, no ECC, no pro drivers.

>>60779331
>>60779347
>impying pascal is not more efficient

>>60779412
[Citation needed]

>>60779412
Cute, 1060 mobile is 90 or 80W, I can't remember.

>>60779412
the 1080 is, lower cards are less impressive

I wonder if Volta will be any faster than Pascal sans bigger dies.

>>60779534
Seems like Kepler 2 to me.

>>60779434
>>60779436
>amdfags now say polaris is not a housefire

wew lad

File: 58b65eee57839f73000d602926415c8c8a29a7d2762c7354e68ebaa569be33e7.gif (4MB, 480x270px) Image search: [Google]

4MB, 480x270px

>>60779434
>>60779436
>Amdrone thinks polaris is more efficient than Pascal

>>60779592
>>60779595
Sshhh, no tears, only dicks now.

>>60779534
It would be pretty weird if there were no improvements at all.

>>60779595
>>60779592
Shh /v/, go away. 8 Geometry units. 8 Geometry units.

>>60779581
Honestly it sounds EXACTLY like kekler2:
>wew we're bringing FP64 hardware back
>only for one enterprise card
Swap for Volta and bolted ASIC aka Tensor Cores.
>>60779592
>>60779595
Polaris is more efficient than Pascal, given binned dies and optimal clocks.
>>60779608
Higher base clocks due to better "12"nm process (which in reality is rebranded 16nm FFN+). And bigger dies.

File: (you).jpg (2MB, 8624x3896px) Image search: [Google]

2MB, 8624x3896px

>>60779632
>Polaris is more efficient than Pascal

>>60779652
It is. No, overvolted OOTB consumer cards do not count.

Can you imagine how buttblasted he is? Where's that 1060 at 65W, anon?

File: 2017-06-06-13-45-49.png (169KB, 1033x1339px) Image search: [Google]

169KB, 1033x1339px

>>60779632
>Polaris is more efficient than Pascal

kys

>>60779707
>overvolted consumer trash
Dear /v/edditor, someone in this thread already explained you that AMD never gives good bins to consumer.
Where is 65w 1060?

>>60779632
>Polaris is more efficient than Pascal

Please stop

>>60779652
99% chance that you are just pretending to be retarded but let me lay it out for you. Polaris is weird arch that has massive exponential relation between voltage and power draw. Plus there is a massive difference between the bins. My own rx480 can run with just 90w of power if I underclock and undervold it to 1220mhz. When I over clock it to 1450mhz it draws 200w of power. The relationships is even more skewed for the better binned embedded or mobile chips. Nvidia chips have binning and voltage-power relationships but nothing this extreme. All the desktop polaris cards are with bottom of the barrel barely functioning chips.

I can actually see him foaming.

>>60779707
>poopowerup
>/v/irgin gaymer
>nvidiot
remove yourself from the gene pool

>>60779753
He's having a stroke.

>>60779769
The summer heat doesn't help his case at all.

File: 1491954309271.jpg (29KB, 640x361px) Image search: [Google]

29KB, 640x361px

>>60779737
>>60779739
>it has 4chan pass

>>60779766
>Polaris is more efficient than Pascal

>>60779724
>you can't undervolt pascal

>>60779739
>>60779737
>>60779707
>This critter visits /g/ for some reason

>>60779785
Undervolting Pascal does fuckall, anon. It's intended to work at higher clocks due to longer bibeline since Maxwell.

>>60779785
[citation needed]

>inb4 technaziup gaymes

>>60779800
No, it helps, but not nearly as much as AMD's chips, both CPU and GPU, which have been overvolted since I can remember.

>>60779785
Have you actually undervolted a nvidia card since maxwell? If you had then you wouldn't be saying such shit.

Dude's literally having a stroke.

>>60779812
Even the 1070 draw less power than the 480...

N17E-G1-A1
TDP 80W

What a housefire

>nvidiot gaymers resorting to argumentum ad hominem
jej

File: RX-480-ABC-91.jpg (122KB, 614x404px) Image search: [Google]

122KB, 614x404px

>>60779846
hahahaha, oh wait you are serious, let me laugh even higher HAHAHAHAHA

Good thing he came in when the thread went into autosage.
Did he ragequit already? Hard numbers must hurt.

>>60779879
I'm not a Nvidia fan but the faggot who said Polaris is more efficient than Pascal is a massive idiot tho

>>60779924
Find me a 65w 1060.

>>60779919
>peak watts

>>60779929
I looked for it. only find 80 watts.

>>60779929
Find me a 65w 480

>/v/ermin is completely clueless about voltage binning and efficiency sweet spots

Color me surprised.

File: zen fmax.png (55KB, 1509x905px) Image search: [Google]

55KB, 1509x905px

ITT anon doesn't understand how gpu architectures scale (or not as the case maybe) with voltage in relation to clocks. Here we can observe the trend that ryzen follows to given an indication of its effeciency (to wit: ryzen 7 and 5 are massively over-volted to hit the clocks they do).

GCN of all flavours is equally massively overvolted to hit clocks over 1ghz (naturally some revisions are more effecient than others). Polaris is pushed way beyond its peak effeciency point to maintain the performance it does. Polaris is a fantastic undervolter because of this.

>>60779963
read the fucking thread you moron.

>>60779973
You talk about the Asus laptop?

File: 1436475125825-0.jpg (262KB, 1000x750px) Image search: [Google]

262KB, 1000x750px

>>60779940
>muh idle

File: Screenshot from 06-06-2017 20:06:30.png (2MB, 1774x1059px) Image search: [Google]

2MB, 1774x1059px

>>60779963

Want a 580?

>>60779331

You're welcome

File: 82412.png (32KB, 650x450px) Image search: [Google]

32KB, 650x450px

>>60779940
>idle watts
stay mad nvidiot

File: untitled-1.png (62KB, 684x851px) Image search: [Google]

62KB, 684x851px

>>60779984
peak watts =/= average

>>60779996
They're lying

>>60779972
Whilst we're on this subject. What is the best way to undervolt gpus? Currently only have my 290 trix at 22% below power target in ccc (any lower and it throttles in fur mark).

>>60780023
>voltage binning and undervolting are lies

File: Screenshot from 2015-03-06 20:08:35.png (74KB, 242x269px) Image search: [Google]

74KB, 242x269px

>>60780023
The salt in your tears certainly isn't lying.

>>60780005
The 1070 is not faster? So it is more efficient, it is?

File: wheresmymic.gif (2MB, 320x240px) Image search: [Google]

2MB, 320x240px

>>60780023
1.7 woodscrews porkshoulders housefires fuck off Huang.

>>60780026
>ccc
update your drivers. use wattman

>>60780020
>calculated
>average
just stop posting and go back to /v/

File: 1443505113252.jpg (41KB, 600x479px) Image search: [Google]

41KB, 600x479px

>>60780023

This thread was so comfy. People were actually talking about the hardware and its technologies. And now we're back to shitposting.

Why you gotta ruin our comfy, anons?

>>60780026

Inch down the voltage, run a benchmark and if it doesn't crash repeat until it does. You can lower the power target in conjunction with undervolting as well but finding that magic point where the gpu will simply not go lower is harder. Naturally (and relevant for hawaii) dropping clocks to 900mhz or just below can give you enormous leeway when undervolting - hawaii was never really meant to be pushed from the 800-900mhz envelope old GCN loves most.

>>60780050
lern engrish frist

>>60780079
It's just one shitposter. Thread's already hit bump limit anyway. Report and ignore.

>>60780079
You need to mention Intel/nVidia once for monkeys to arrive and start flinging shit at everyone.

>>60780026
Easiest way to be to lower the power target as far you can.
That's the quick and dirty way, otherwise the same thing as CPU overclocking, lower voltage until you hit stability issues at a certain clock, then lower clock.

>>60780079
Thread's on autosage anyway.

>>60780039
>>60780042
>>60780051
>>60780063
>you need undervolting you gpu because he is so hot/draw to much power

And you think it will perform like the desktop version? No.

So if Vega is designed to be scaled up and down more easily, like Ryzen is, does that mean that after a few months we may see some slashed down Vega units replacing Polaris 10's current position in the lineup? Much like R7 to R5, and presumably R3 coming down the pipe.

I'm asking because, from reading the thread, Vega is a significant change to the architecture and not merely "big Polaris".

>>60780184
Nobody's stopping you from putting a 150W GPU in a laptop, but you'd be retarded for it.

>>60780212
Both companies can undervolt their GPUs...This is not an argument

>>60780244
Vega 11 seems to be Q4 and a replacement for Polaris.

>>60780244
RX 580 and lower won't be replaced until next year.

>>60780244
>So if Vega is designed to be scaled up and down more easily
No, that's Navi. Vega is about making front end that does not suck horse cocks. But there will be a Vega11 to retire Polaris lineup.
>>60780254
Undervolting nVidia since Maxwell does fuckall due to long pipeline.

>>60780254
Sure? Now show me a 65W 1060.

>>60780277
>Undervolting nVidia since Maxwell does fuckall due to long pipeline.
b-but Nvidia is supposed to be superior at everything ;_;

I think I figured out how they got from 95W in Polaris 10 embedded to 65W in Polaris 20, the updated LPP process is more than likely more efficient at lower voltages than the first LPP iteration, exactly what happens with Zen

>>60780263
>>60780265
That sounds about the timeframe I'd figured. Vega probably won't hit mass availability until the end of the year, and any cut down or Vega 11 would be months after.

>>60780277
I failed to read up on Navi's position/purpose. Thanks for the clarification.

>>60780318
>Vega probably won't hit mass availability until the end of the year
Why? .

>>60780279
Nvidia doesn't need undervolt because their current GPU are more efficient than Polaris. The 480@65w will not perform like the desktop 480.

>>60780331
Sniped by shitcoiners.

>>60780331
HBM

>>60780333
nVidia CAN'T undervolt, longer pipeline means higher optimal clocks, anon. It's like underclocking Presshots.
>>60780344
Nah. Costs too much. They'll still buy every Polaris card on the market.

>>60780333
How do you think they got their 1060 at 80W besides undervolting you moron? It runs at lower clocks too.

Wow, who would have imagined.

>>60780353
Considering they are using fucking UNANNOUNCED 8-hi stacks, which means they have no problem with HBM yields/TSV stacking.