2 FUCKING TERABYTES This will be incredible for content creators

>>61667604
It'd be neat to use as a RAMdisk too

>inb4 favela monkey shitting in another thread

>>61667604
This is some shit for the Pixar rendering farm, isn't it.

File: 1471093004705.png (437KB, 2586x1567px) Image search: [Google]

437KB, 2586x1567px

I wanna see this babe in action

>implying you'll get your cheetoh-dust coated fingers on one before the cryptoniggers do

>>61667604
That can't be 2TB of ram. Does that say 16gb of ram with 2tb solid state memory for swap?

>>61667709
Is this the new /r/edd.it team + boogieman? A few months ago it was "img xxx".

>>61667713

Yeah its for super high end post production houses doing animation rendering on farms of these or things like color grading on RAW 8k footage. It costs at least $7k.

>>61667747
I think a shit ton of cores or whatever the hell are better for mining than VRAM

24tflops of fp16????!!!

Is it wrong that I just want this to test mining and running folding@home all day?

>>61667837

It'd be a monster for mining but again the price would be prohibitive to make it actually worth it.

>>61667604
No shit.
AMD basically won the Hollywood with it.

>>61667753

16GB of HBM2 + 2TB SSD. Ideally, you're gonna put the best fucking SSD you can find into it, not some generic turd TLC nVME drive.

In a nutshell its: 16GB HBM2 = L3 >> 2TB SSD = L4 >> GOD FORBID YOU NEED MORE >> HBCC = main RAM access @ L5 cache.

Man the way Vega works with memory is fucking unheard of for GPUs.
Why didn't nvidia try that before?

>>61668064
It's using Samsung's top crop drives.
Samsung and AMD are basically buttbuddies now.

>>61668078
Too busy coming up with astounding achievements like really, really big dies.

>>61667604
why do we use gpus and cpus if gpus are so powerful

>>61668165
GPUs are only good at number crunching.
Sure, a lot of tasks can be solved by pure number crunching, but not every one.

>>61668103

They became butt buddies the day GloFo got access to IBM's fabs and IBM/Samsung black magic 0.5 gen optical to 1.0 gen EUV 7nm fab process.

>>61668199
GloFo was part of Common Platform for a long time.
Anyway, Samsung shills EBYN and Radeons, AMD shills awesome Samsung memory.
It's a win-win for both.

>>61668106
and really, really low yields

Also Samsung is ramping 8-hi HBM2 stacks production, now you know precisely why.

>>61668048
>hollywood now can pump out more cgi shit even faster
wtf i hate amd now

>>61668048

>To render Avatar, Weta used a 10,000 sq ft (930 m2) server farm making use of 4,000 Hewlett-Packard servers with 35,000 processor cores with 104 terabytes of RAM and three petabytes of network area storage
>Creating the Na'vi characters and the virtual world of Pandora required over a petabyte of digital storage,[110] and each minute of the final footage for Avatar occupies 17.28 gigabytes of storage.
>2009

In essence, this Radeon Pro SSG is able to take a significant % of that compute and RAM/cache requirement and condense it into a card. So instead of needing 35,000 CPU cores + 104TB of RAM; now you'll need only like a few thousand CPU cores + few hundred of the Radeon PRO SSGs w/ 2TB Samsung nVMEs and bam. You've got a render farm.

You're going from 10,000sqft server farm to like 20 rack towers and calling it a day. This has zero benefit to gaming, but high end content creation is going to be absolutely stellar. Hell, this BTFO's Nvidia completely in this particular use case.

>>61668309
>10,000sqft server farm to like 20 rack towers
Did AMD just unlesh an unspeakable evil?

File: fc4bd6a4bd6fdc89bc59c5117a37eb65e072281e_hq.jpg (37KB, 1024x576px) Image search: [Google]

37KB, 1024x576px

>>61667861
But it would be absolutely devilish to mine coins out with it so the gpu snatchers don't get them.

>>61668165
You can have a million different finely tuned op codes or you can do a handful with incredible speed.
We pick both, put them in separate modules, and call it a day.

>>61668346
That's basically what they did with Epyc, too.

File: [HorribleSubs] Gabriel DropOut - 08 [720p].mkv_snapshot_10.53_[2017.02.27_14.17.01].jpg (168KB, 1280x720px) Image search: [Google]

168KB, 1280x720px

>>61668365
That is definitely a Satania tier plan.

>>61668365
Dumb Satanichia poster.

>>61667713
>>61667785
Sadly most likely not big companies like pixar or dreamworks
I worked IT at dreamworks for 5 years, all of their shit is CUDA based. Their rendering farm is entirely Nvidia basically. Hearing from a friend who still works there though that they are switching to Threadripper or epyc, I don't remember which. Can't say anything on Pixar though

>>61668421
The offering that is SSG is too nice to pass no matter how much did you invested in competitor's ecosystem.
It's 1S EPYCs of GPU world.

cool will order 5000 of these for my eth mining rig

>>61668495
>$7k a pop
Good luck.

File: 1470045403614.png (267KB, 758x789px) Image search: [Google]

267KB, 758x789px

>>61668250
HBM on CPU package when?

>>61668459
Dreamworks gets incredible deals on their hardware
People throw quadros around there like they are nothing. We had a stack in the peripheral room. I don't think they have that anymore

>>61667774
isn't IMG_XXXX just due to FB/iOS posting or something?

>>61668520
High latency so never

>>61668520
Whenever AMD wants.
>>61668526
It's not about deals.
Nothing can compete with SSG, simple as that.
It's 1S EPYC of GPU world.

>>61668529
iOS posts are always named image.extension

>>61668529
It literally just means you either
>Downloaded an ios photo
or
>posted on a phone

>>61668537
I'm just saying the facts
I doubt they will ditch their current rendering farms for SSG since the majority of what they do is CUDA based

>>61668589
HIP makes porting from CUDA relatively trivial.
Honestly it's up to them and everyone else to endorse AMD products.

>>61667837
There are other vega cards that are much cheaper for that. This thing costs 7k.

>>61668421

They will probably consider switching over to PRO SSG. I get the value of CUDA, I really do. But that HBCC is black fucking magic, and having access to a 2TB addressable L4 on card is absolutely insane.

But that's just the tip of the iceberg. HBCC basically makes any Vega Pro SSG capable of 512TB virtual address space aka cards can handle up to a theoretical 512TB data set and do just fine with the 484GB/s mem bandwidth on tap.

You factor in Vulkan's explicit multi-adapter, and you rewrite your render farm software with Vulkan + Radeon PRO SSGs with validated drivers and the possibilities are mind boggling.

Let's go with a hypothetical for a moment. Let's say each rack can do 6 Radeon PRO SSGs w/ 2TB SSDs for L4 Cache. Each Tower can do 16 racks; and you have 20 towers on tap. That's a total of 1920 Radeon PRO SSGs w/ a total addressable space of 3.84PB as an L4 cache. It also means you have 23.5968PetaFLOPS of compute performance at FP32 and 47.1936PetaFLOPS @ FP16.

FP16 render benefits: http://www.geeks3d.com/20100916/fp16-demotion-a-trick-used-by-ati-to-boost-benchmark-score-says-nvidia/

Nvidia's shitposting aside, when you look at the two image samples provided; there's an obvious loss in quality to an extent--but its not statistically significant that you'd throw the entire frame out. So, factor in CGi rendering. Let's say smoke effects, explosions, basically all around heavy alpha effects that are compute heavy but also don't necessary need absolute precision. Here FP16 would be beneficial. Or if you're rendering big fucking skyboxes and massive landscape shots where the macro is more important for immersion to the audience, than the absolute minute details--especially if you don't intend to spend longer than 5-10 seconds on the shot. Here, again, FP16 would be valuable.


Vulkan + this setup + explicit mGPU is probably another 5-8 years out to really take off, but even without it, the PRO SSGs bring a fuckton of value that Nvidia can't.

>>61668684
>AMD casts black magic on enterprise markets...
>...again
Hilarious.

You tellin me I can finally run Skyrim with all my texture mods? Using the best CPU, a 2500k of course.

>>61668724
Nothing, nothing can fix the eternal Bethesda pajeet.

>>61668064
Wait you put an ssd inside it?

Or sata's or m2'd to the mobo

>>61668571
or
>Trolling is a art

Why didn't AMD shill it like there's no tomorrow before?

File: cad_point_maya.jpg (19KB, 500x500px) Image search: [Google]

19KB, 500x500px

>>61667604
Looks like I picked the right time to get into CGI animation. :)

Because if jacketman had something like this we would've listened about it for 5 keynotes in a row.

>>61668747

Enterprise grade NVMe SSDs, yeah into the card itself, not mobo.

>>61668714

If you go back to the specsheet for Vega and look at what hardware improvements/additions they did to Vega and consider it in the context of the Radeon PRO SSG; they all make perfect sense. Just, they have very little value for gaming. But for massive data sets, rendering, and CGi render farms, the PRO SSG cards officially are in a position to completely BTFO Nvidia & CUDA. AMD's open compute platform lacks the library/efficiency capacity of CUDA, but it now has something that Nvidia completely lacks; cache viability.

A big die quaddro could come up with 24 and 32GB of GDDR5X or HBM2, but those dies are gigantic and very very expensive and once you cap out for the 24 or 32GB of G5X or HBM2 you're done. PRO SSG hits 16GB HBM2 and goes "let me just take a few steps to the right and I have 2TB to play with."

Yes its not as fast as G5X or HBM2, yes you're sacrificing some degree of latency, but 2 fucking terabytes on a GPU man; up to 2GB/s reads and 1.25-1.5GB/s writes; add in the Draw Stream Binning Rasterizer that Vega introduces and voila, black fucking magic.

Consider this:

DSBR takes target frame and breaks it down into 32 frame shards >> Vega renders frame shard & sends to SSD DONE space >> works on next shard >> DONE space >> works through all shards >> DONE space >> recalls all frame data >> switches to FP16 and begins post-process workloads >> masters frame >> sends to EXPORT space on SSD >> goes back to new framedata from WAIT space on SSD >> reads into HBM2 >> REPEATS

Meanwhile, application takes data in export space and sends it to the SAN. Then a video engineer points Vega Pro to the SAN and it imports the gorrillion frames, which are then compiled into a film. Factor in after effects doing the intros/outros, and any kind of specialized typestting; which are then added to Vegas Pro and bam you're done with a film.

>>61668965
Yes, Vega looks like another explicitly enterprise focused uarch from AMD, just like Zen.

>>61668789

That thing is going to cost in the 4 digits

>>61667604
How is it at buttcoins

>>61669060
They can charge $20k for it and it'll STILL sell like hotcakes.

>>61669022
Pretty much.

Raja and AMD did a good job here, but /g/amerrs won't realize or understand what exactly they did here.

>>61669147
Honestly they need to enable DSBR in gayms and it'll sell just fine.
They have 2 weeks left before launch.

Wait, Tim "I suck JHH's cock" Sweeney on AMD keynote?
You gotta be shitting me.

File: IF_VEGA.png (1MB, 2000x1125px) Image search: [Google]

1MB, 2000x1125px

>>61669022

RTG and AMD, till Vega, didn't have enough money to face gaming and enterprise simultaneously. Vega will be just enough for gaming, and pure carnage in enterprise space.

AMD & RTG are both are probably hoping to claw out 10-20% space in Enterprise from Intel & Nvidia. Doing so, will net them $1-2Bn in additional annual revenue. This will give them some breathing room to be able to focus more on drivers and better architectural optimizations. On paper, Vega should for all intents and purposes slaughter the 1080 and give some competition to a 1080Ti by either matching it or coming within 10-15% of its performance. But it does neither; and that has a lot to do with drivers and also architectural optimizations, both of which need time and money. Which AMD & RTG have neither off at this moment.

But give it a year or so, when they claw a significant portion of the Enterprise market for themselves with Zen/Threadripper/EPYC + Radeon Instinct & PRO SSG and they'll have enough money to do a clean die reset with the next Radeon GPU core like they did with Zen. Take the best of Vega and factor it into Navi.

Also, everybody under the sun seems to have forgotten PIC RELATED.

>>61669431
Yes, Vega layers IF protocol on top of on-chip crossbar.
Anyway, drivers fixed Tahiti.and Hawaii, drivers will also fix RX Vega.

>>61669431
>45MB of SRAM
HOW
WHERE

File: slides-38.jpg (138KB, 2000x1125px) Image search: [Google]

138KB, 2000x1125px

>>61669479

>mfw each compute unit has 703.125KB of L2 SRAM

>>61668965
So realistically how long would it take leather jacket man to make something similar to this?

nah

>>61669147
Gaymers understand very well what AMD is doing with Vega, but they are understandably upset because they feel like they've been thrown under the bus for enterprise profits after being strung along for over a year.

>>61667604
Can it run Crysis on max, though?

>>61669746
It can render Crysis: The Movie.

File: slides-09.jpg (185KB, 2000x1125px) Image search: [Google]

185KB, 2000x1125px

>>61669683

Without violating any patents? Two to three generations. HBCC won't work without Infinity Fabric.

Nvidia's NvLink is a good starting point for their own Infinity Fabric-like implementation. Once they figure that out, then they can begin work on the cache controller. But the thing is, the HBCC is a 30/70 shared responsibility piece of architecture w/ respect to driver/hardware level data management. The issue I see Nvidia running into is that their entire GPU stack is designed for the driver to handle 70% and hardware does the last 30% with unparalleled power and efficiency.

It will take them at least a full generation to do an Infinity Fabric like implementation. Furthermore, they will have to try it out in the consumer market, because there's too high a risk in their enterprise markets for this kind of tech fucking up performance and precision; and if they fuck it up, their performance will go straight into the toilet. Furthermore, they will have to start shifting their GPU stack from driver > hardware to driver <> and then eventually hardware > driver.

Which in turns means admitting that AMD is right in both software and architecture choices for the future, a pill they have a very difficult time swallowing; given who vehemently they refuse to move to DX12/Vulkan--pure hardware scheduler with minimal driver overhead approach to things.

Their command list implementation is masterful, and credit is where credit is due. Its async-lite, with the driver effectively acting like the hardware scheduler. Pic related paints the direction in which Nvidia needs to shift to get an IF implementation to then get an HBCC. And most importantly, they must be willing to accept the risk that there's a significant potential to fuck up performance and release a super hot product that performs barely better than last generation, and even worse than their direct competition.

Finally, they have no CPU division to prototype it. ARM doesn't count. At all.

File: 1469265359818.jpg (93KB, 804x743px) Image search: [Google]

93KB, 804x743px

>>61667604
>content creators

File: 1475211353488.jpg (108KB, 782x782px) Image search: [Google]

108KB, 782x782px

>>61667604
DELETE THIS RIGHT NOW

THATS TOO MUCH MEMORY WE CANT AFFORD IT

ISRAEL WILL GO BANKRUPT

SHREEEEEEEEEEEEEEE

>>61669819

Also known as ENTERPRISE MARKET.

>>61669794
Interesting, thanks for the explanation.

>>61667802
Not in the case of ethereum

>>61669819
Dumb frogposter.

>>61667604


only $2000 dollars when you need about 10 in one server rack and about 20 server racks of the same thing to render one expensive scene which takes a whole day.

>>61667802
It's only 16GB VRAM mate... the 2TB part is an SSD.

>>61669845
>>61669872
Whilest probably a dumbfrogposter, I'm going to give them the benefit of the doubt.

Hyping this card here is fucking stupid, absolutely no one here is going to be using it in a professional capacity here unless they're cancer youtube shits.
Yes, it has a market - but the professionals who work that market don't come to /g/ to discuss their new renderfarm.

>>61668589
You also just finished saying they don't give a fuck about the gpu's and toss them around like toys

File: next gen graphix.jpg (2MB, 1788x2082px) Image search: [Google]

2MB, 1788x2082px

>FP32 is 0.02 Tflops lower than it should be

R U I N E D

>>61669952
>"content creation doesn't exist outside youtube"

File: slides-12.jpg (256KB, 2000x1125px) Image search: [Google]

256KB, 2000x1125px

>>61669952

>they don't come here to discuss their new render farm

Wow, really? Next you'll tell me that the sky changes color depending on the time of day.

>>61668309
>60s of Avatar final footage 17.28GB of data
>Vega can handle up to 27GB of assets and 500M triangles on the fly

That's cool as fuck, and people like talking about it. What's your problem with it? Hype? No one's hyping this up for any buyers. People on in HPC/RenderFarm market will do what they will.

>>61669060
Seems like a fair price point for a professional animator.

File: slides-18.jpg (104KB, 2000x1125px) Image search: [Google]

104KB, 2000x1125px

>>61670044

Just piggybacking.

>quad sum of absolute differences
>The sum of absolute differences may be used for a variety of purposes, such as object recognition, the generation of disparity maps for stereo images, and motion estimation for video compression.

>video compression
>dealing with 8K video in realtime
>being able to use GPU for encode/decode
>object recognition which an HPC function
>disparity maps aka determining object parallax aka better understanding depth information of specific or global geometry within the frame
>primary benefit being rendering, virtual/augmented reality solutions, potentially physics calculations, lighting/shadowing, etc

Also, a quad sum of absolute differences also measures the similarity between image blocks. This would be absolutely valuable for newer types of anti-aliasing solutions, as well as checkerboard-type upscaling methods, effectively needing to render only half the frame and then using the data from each quad to approximate the missing framedata.

>PS4 Pro and XB1 implement checkerboard rendering to hit 4K render targets as best as their hardware allows
>QSAD would allow future consoles to either do native 4K rendering or close to 4K render, use this new instruciton set to better approximate new framedata from adjacent quads and fill in the gaps resulting in a clearer picture overall with negligible performance losses
>then use the remainder of the processing power for improved post-process or additional onscreen objects

Lots of cool architectural bits in Vega, that won't see their actual strengths play out for several years. It is nonetheless unfortunate, that AMD threw gamers under the bus for this.

>>61668684
But will it play Crysis?

>>61670174

Yes, at 1440p60 maxed out.

>>61670212
Dude, that's insane, think it'll go to 2xAA!?

>>61669060
the article I read said $9999

>>61670247

I think it might be able to pull of 4xMSAA!

>>61670395
nah the SSG is $2000 USD.
its the quardo 6000 is $7000 USD

the SSG is avaiable in ~14th september

t.GTX1070 user
also thanks for shitting on me /g/tards

hypothetically

if i had large amounts of money that i could spend however i wanted and value is not a concern

could i buy this and not worry about getting a new GPU for at least a decade?

>>61670776
Yep.
rich fags get all the nice shit.

>>61667604
They put a SSD on it

>>61670776
If you had large amounts of money, why would you even worry about the monetary value of these things?

>>61670130
>>61670030
>>61669794
>>61669431
>>61668965
>>61668684
I don't really understand everything, but this is interesting to read. We need more anons like you instead of shitposters.

>>61670850
This. The posts are so informative and fun as fuck to read for me, and yet i get to read this kind of stuff near to nevere...

>>61669733
I can't blame AMD for prioritizing Gaymers over Enterprise folks. Even when AMD delivered products better and cheaper than Nvidia, people still overwhelmingly buy Nvidia.

>>61668421
most of the programs are giving support to the ssg so having cuda or not is moot unless they run some in house app that uses cuda which this isnt the case

>>61667604
Good for mining maybe? And no I will not fuck off to /biz/ thank you :^)

>>61671254
Read the thread dipshit.

File: F_K0fNr2WL43u_mVh2FyAjPXWthIjeepbWxB22A1vGI.jpg (75KB, 505x767px) Image search: [Google]

75KB, 505x767px

>>61668309
you mean like this?
https://www.youtube.com/watch?v=EHDY5uAC9bE

>>61671254
Theoretically it should be the best mining card ever made. But it costs $9,999. Might as well buy 20+ 580s.

>>61667604
>16GB Vram
>i need more vram
>"we convert ssd into vram with slower speed"
>technically video card has up to 2tb ""vram""
at least it works, slower of course

>>61669794
>Which in turns means admitting that AMD is right in both software and architecture choices for the future, a pill they have a very difficult time swallowing; given who vehemently they refuse to move to DX12/Vulkan--pure hardware scheduler with minimal driver overhead approach to things.
Nvidia hates vulkan because they can't play dirty with driver implementation because it's so strict. It doesn't have anything to do with amd creating it, more that they can't out perform competitors

http://blog.mecheye.net/2015/12/why-im-excited-for-vulkan/

File: 691c0ce32d038795e96062d95329921d5e9205ee.jpg (18KB, 396x222px) Image search: [Google]

18KB, 396x222px

>>61667604
>But y tho
>Srs
>I into applied math phd
>can into parallelized linear algebra for reasons at times
>can only afford Steam during sales
What the fuck do you actually do with one of these things and how much would an informed enterprise user pay? My senile professors still talk about doing distributed work on Playstation consoles. What does all of this (exponential) GPU scaling actually do?

>>61667604
IT'S OVER
NVIDIA IS FINISHED

>>61671308
>they can't play dirty with driver implementation because it's so strict
>more that they can't out perform competitors
Fucking hilarious if true, i love it.

>>61669431
nor do they have it now...vega is literally the most gcn compute monster to date they clearly went full enterprise this time around
given that they gave the prosumers sriov also on a price that totally destroys everything on the market
its a no brainer i was shitting this place saying that amd is sandbagging harder than zen and i wsa right but i guess only in enterprise tho LOL

>>61669794
for a better tl dr hbcc is literally a more advanced hardware sc an I/O and storage hardware sc to be exact
how they manage to have 2 sc's into the damn card and not having a significant bump on the powr draw it amazes me i expected hbcc to be responsible for yet another big bump on the power similiar to the bump we saw from VLIW to GCN

>>61669952
we are not hyping the card we are hyping that amd might possibly at some point offer the same tech in the prosumer segment or even for us the lowers the plebs...

imagine a fucking rx vega with a 200gb ssd effectively cutting out all the fucking middleman on a system that causes lag and dropouts this way all the damn cgn cards will be filled 100% and not having half of their rops sitting back and drinking tea while doing nothing..imagine the perfomance on games using vsr wont even be a problem cause the card will handle it like a champ
imagine all the fucking possibilities of it..

>>61671309
see
>>61668309

One card costs about $7000.

>>61671334
which units have sr-iov support?

>>61671421
for now confirmed the firepros from low to up
amd said that vega FE also had it so i might dare to say the high end rx vega might also have it

>>61671437
I'm buying vega 64 for mesa/wine-nine but if 64 doesn't have SR-IOV and FE does I'm going to have to do some serious thinking.....

>>61671451
mid august we shall know (apparently 15 is the day that they will hard launch )

File: 1474058606838.jpg (37KB, 550x368px) Image search: [Google]

37KB, 550x368px

>>61668421
they might be switching to petaflop rack
cloud workstation demo was kind of neat

File: 1490104602434.jpg (25KB, 641x530px) Image search: [Google]

25KB, 641x530px

>>61671308
I feel shameful to even consider waiting for volta

>>61671522
with the rumors of volta having a hardware sc and having a 300-350watts of tdp? lol i wonder what excuse they gonna find for that

>>61670752
the SSG is $6999, raja said it on siggraph

>>61671602
right then my bad.

>>61671602
if ssg was on 2000 amd would have to create an insane stockpile to feed the miners...

>>61671639
Fuck the miners

>>61671550
>rumors of volta having a hardware sc
nope, scrapped. it was supposed to have it year ago not anymore, my bet they simply couldn't make TDP in reasonable frame

>>61671680
the GV isnt the true volta the volta we know its coming will have a hardware sc its either have it or get left behind since most of the games on 18 and beyond will be based on fp16 anc compute shaders in general..we saw that trend since star citizen..

>>61671711
>we saw that trend since star citizen..
meh, more importantly AMD is long term buddies with bethesda, that's much bigger
will see how it works in wolfenstein first

>>61671728
nah id is famous for doing games that run perfectly well on everything even on a potato

im more interested on far cry 5

>>61671728
>>61671744
>Wonder what the fuck you 2 niggers are talking about
>See this in my feed http://www.dsogaming.com/news/wolfenstein-2-far-cry-5-will-support-amds-vega-exclusive-feature-rapid-packed-math/

Welp, guess that explains that mystery. Had my thoughts that Wolfestein did some shit, like how Rage was big on using CUDA for it's texture bullshit, but had no idea about FarCry.

File: 31175613121l.png (748KB, 1200x675px) Image search: [Google]

748KB, 1200x675px

>>61671786
>Wolfenstein is using ID Tech again
>DX12
Why the fuck would they support that mess when they got Vulkan?

>>61671786
i always found it funny that rage was using so much power and was heavy on every card and yet no one every did a power draw test on nvidia while playing it..

now i know why LOL

>>61671786
nvidia will go way of 3dfx if they keep up same shit 3dfx did and market finally stomped them into the grave

Also weird how nobody talks about "AMD Studio" I though it was the most giant announcement for gamers, it means the devs will have free resources that already optimized for hardware

>>61671823
I don't think it matters much, also different team altogether
doom vulkan runs splendidly on everything because doom was really really well optimized game that could run even on consoles 60fps even before vulkan, it would run this way on dx12 as well

>>61671823
bewcause dx12 is nice on nvidia vulkan on the other hand is very strict on using abstracts
nvidia is trying to bring their gameworks shit on vulkan with a high level layer https://www.custompcreview.com/news/nvidia-releases-high-level-abstraction-layer-vulkan/36450/

but i doubt it will work this time

>>61671823
wait, it says dx12/vulkan there

>>61671851
>AMD Studio

you mean the gpu open? cause i cant find anything with this name

>>61671876
watch siggraph stream
https://youtu.be/nuFGUVg0U9U?t=1h16m19s
from 1h16 or so

also this thing is kind of big
https://www.youtube.com/watch?v=9FnqjXjIbW8&t

>>61671912
oh the gpu profiler
yeah i know its literally their console profiler but a very basic one considering their console profilier has access to literally everything

>>61669431
>Vega will be just enough for gaming, and pure carnage in enterprise space.
It really won't be. OpenCL/Vulkan just doesn't have the software ecosystem anywhere near that of CUDA's. That ALONE is enough to stop a lot of potential customers from switching over. AMD had outright better compute performance from the launch of the first GCN, up until the launch of Pascal, but saw hardly any uptake in enterprise sales. Secondly Vega just about competes with GP100. However now VP100 is on the horizon and completely outclasses Vega, especially when it comes to machine learning which is an absolutely huge source of revenue for nVidia. At best SSG will find a solid niche, but at this point AMD should just cede the GPGPU market to nVidia and focus on gaming. In fact that's what they should have done all along. Recent sales of high end GPUs show there's plenty of money to be made gaming.

Stop shilling and just accepted that Vega is going to be pile of shit on all fronts.

>>61671943
no. watch the stream recap
they are basically making in house software studio to work on game engine stuff that used in everything nowadays

>>61671956
stop posting pastas, make an effort.

>>61671957
no i found out lol
amd studio is literally a content studio on hollywood
http://www.hollywoodreporter.com/behind-screen/siggraph-amd-studios-opens-hollywood-1025318
haha oh man this is getting funny in one year lisa literally open all of her torpedo tubes and is ready to fire

>>61667730
The video is dead and I can't find a mirror, but they showed a 1TB model on SIGGRAPH last year that could do realtime 8K rendering at 90Hz.

>>61672051
there is a video here
https://www facebook com/jarred.land/posts/10154878409465415?comment_id=10154878470535415&notif_t=like&notif_id=1501276190497990
the guy works at RED ssg literally murders titan xp

>>61671968
It's not a pasta, and I'm not wrong. nVidia have the GPGPU (can you even call it GPGPU at this point? Considering nVidia have started using different silicon for enterprise GP100 vs GP102) on lockdown. They have better hardware, and they have a better software ecosystem.

I'm probably still going to buy a Vega, to upgrade my 290, because fuck g-sync.

>>61672020
>>content studio
Like a farm for Avatar like movies? Platform developers will be able to rent?

>>61672071
>the guy works at RED
He the ceo of red

>>61672089
no, open the video >>61671912

>>61667785
Stop spewing shit
http://www.techradar.com/news/amd-radeon-rx-vega-could-match-gtx-1080s-speed-but-maybe-not-its-price
>On that score, as eTeknik reports, a Swedish site (Nordic Hardware) claims to have seen the retail price list for Radeon RX Vega, which will allegedly be around 7,000 Swedish kroner plus VAT, which would equate to about $850 plus tax, or £650 plus VAT (AU$1,070).
>select helicopters.

>>61672073
I'll say it again: 3dfx had 90% of the market and died overnight doing same shit. Market can take proprietary monopoly only for so long.
Just look how amd taking back gaming market game by game, and only a year ago it was gimpworks in every title.
Things will normalize to 50/50 like it should be in next 2-3 years.

>>61672134
>what is reading conprehension
The coment chain is about the Radeon Pro SSG, not RX Vega.

>>61672153
oh sheeet

>>61671126
All the more reason to hold onto the few fans they still have.

>>61668529

My old Canon DSLR names images like that, as does my older Samsung tablet.

Facebook image names are like

10665378_576729962473309_1460754340001904308_n.jpg

>>61667713

Pixar uses RenderMan, which is a CPU based renderer and has an optional Denoiser they use that is CUDA based.

>>61668421

No company on that scale with active projects that has a GPU based render pipeline is going to swap to a CPU one. Bullshit harder.

>>61669060

So does Maya every year so he's already rich enough.

File: Screen Shot 2017-08-01 at 12.52.21 AM.png (81KB, 684x279px) Image search: [Google]

81KB, 684x279px

Did AMD just crack the code?

>>61672281

>Every AMD shill under the sun is now suddenly an anonymous source at a world class VFX studio

What a coincidence.

>>61672299
>What a coincidence.
someone that just stated the truth without even shilling is an amd shill
/g/ 2017

>>61672316

>Stated the truth
>Claims rendering is done on CPU because of memory loading the scene as the first reason

Well at least now I know you're posting images while knowing nothing about the subject source.

File: 1493483841007.png (59KB, 265x265px) Image search: [Google]

59KB, 265x265px

>>61672299
Anon that was posted last year by Larry Gritz - Software Architect at Sony Pictures Imageworks.
https://stackoverflow.com/users/3832/larry-gritz
Also from last year.
>One of Pixar's software engineers, Jeremy Cowles, said that the industry tells NVIDIA, as they question them every year, that they need more memory to actually use them on a large scale.
Not only have AMD swooped in before Nvidea they're going to steal the render market from Intel.

>>61672330
>>Claims rendering is done on CPU because of memory loading the scene as the first reason

a system with 2tb of ram is better than a gpu with 12-24 gb of ram
a gpu with 2tb of ram is better than both
who would have knew!

>>61671272
Jesus.

This thread is comfy

They talked a lot about this in SIGGRAPH 2016, they even had a working fiji-based beta version: https://youtu.be/vyKgT5QUU2Q?t=1h05m55s
This one should be much better, considering that the Vega architecture has a lot of new stuff specially made for this, unlike that "glued-together" fiji version.

>>61672559
Really, uh, impressive stuff, you know, but man, you know, uh, Raja is, uh, you know, not a good, uh, presenter, you know.

>>61669683
Leather jacket man fell hard for the machine learning meme.

Is it possible that PS6 or something will use SSD that gives direct access to CPU and GPU? Something like hybrid version between standard and what we have here, for faster gaymes frames rendering.

>>61672757
pretty sure its not the storage that gives shit framerates.....
its their stupid decision to have a beefed gpu smacked with a cpu that performs slower than a p4

>>61667785
So cheaper than the RED card and this one doesn't require another card to work.

>>61671272
What's the price for one of those?

>>61668309
Content also increases in resolution and post processing. Think 8K HDR for next Avatar.

I may have missed, but does this thing use power of Infinity Fabric™?

>>61672909
It does, it layers it on top of on-chip crossbar.

File: 1482685379641.jpg (362KB, 1106x1012px) Image search: [Google]

362KB, 1106x1012px

>gpu with ssd glued on

>>61673057
Yes, it's also on of a kind.

>>61668365
>>61668386
>>61672341

why are these guys using anime avatars when posting? somebody please explain

File: 1467904866360.png (373KB, 656x685px) Image search: [Google]

373KB, 656x685px

>>61673192
Anime website.

>>61673192
you are not strong enough for smugness philosophy course

File: 1501163544894.png (31KB, 720x644px) Image search: [Google]

31KB, 720x644px

>>61673192
they're autistic and usually gay, please don't make fun of them.

>>61667604
"content" "creators"

>>61672757
My predictions for next gen consoles, dual IF connected 4 zen+ cores paired with 2048 navi cores APU dies on an interposer.

>>61673256
Hilariously complex and costly for console hardware.
It's gonna be a single chip.

>>61673267
considering how cheap its for amd everyone will go for mcm

>>61673285
Silicon interposers are not cheap.

>>61673312
yes they are................????????????????????????????

>>61673217
Recovered passwords are very much content.

>>61673392
It's relative. Nobody anywhere wants to pay for a 20-layer ultra-fine-pitch pc board, but when you gotta, you gotta.

File: 1485428103191.gif (3MB, 400x225px) Image search: [Google]

3MB, 400x225px

>>61669825

>>61668048
>More CGI thanks to faster rendering
Woopey more CGI fest movies, like we needed a new superhero movie every day instead of a week

>>61674396
Luddite please.

>>61673312
The estimated cost for fiji's interposer was like 3 bucks

File: 1500482359678.gif (607KB, 800x792px) Image search: [Google]

607KB, 800x792px

>>61669825
keked so hard
Good on 'ya anon.

>>61674550
Before I forget.
>inb4 dumb frog poster

>>61668230
>Samsung Memory
>AMD GPU
Can it be liquid cooled? I want that card to heat up my house and supply hot water :)
(for reals now, if AMD can collaborate with other workstation brands say for example Dell and HP they will pretty much win the whole workstation market)

>>61674583
>Dell
Certainly.
>HP
Maybe.

>>61674396
you know that CGI is used even in melodramas? CGI is in every movie now.

>>61669900
20k is literally nothing.
Some time ago I read an article from hacker news which said that if some companies even have to pay 6 figures it will make profit after 1-2 weeks usually. IT shit is cheap as fuck for companies. Only normies cry usually

>>61670030
>pic
still looks like shit without proper texturing

>>61674608
Forgot one
>Apple
101% possible since they shill AMD Graphics (source: the new Mac Pro)
A lot of people use software that is exclusive for mac for example final cut pro, and currently with the new Mac pro they will just offer "UP TO 3x MOAR PERFORMANCE THAN ?" With this card, just don't expect that it will cost less than $8000 for that upgrade :(

>>61674860
>less than $8000
I expect $11k minimum

>>61674860
The card would cost less if nVidia could compete.
But the only thing nVidia makes right now is biggur dies.

>>61674629
What is a melodrama?
I never heard of it, Google dindu nuffin

>>61674860
Apple, amd and samsung are buddies now? WTF.

>>61674902
Wake up, Raja and Papermaster worked at Apple.

>>61674902
They are on a holy Crusade called "screw Intel", I guess

>>61674902
I told you that this was the real master plan. Amd was getting friends while both Intel and nvidia cornered themselves with their monopoly tactics and MUH PATENTS, until they can't do anything but making bigger dies.

>>61675162
IBM, Apple, AMD and Samsung holy crusade.

So when will the SSG support those 15 TB ssds?

>>61675392
>slow 15tb ssds vs ultra fast 2tb ssds
Why?

>>61675392
>15TB
It's usually TLC shit.

>>61668787
>Why didn't AMD shill it like there's no tomorrow before?
Corporate new product marketing strategy is to say as little as possible about upcoming new products except to lower expectations before release.

>>61672930
So, so far it allowed AMD:
CPU+CPU
CPU+GPU
GPU+GPU (whenever those are going to be done)
GPU+SSD...
What other black magic tricks can we expect thanks to it?

>>61667709
Sorry anon, I'm still white european mustard race

>>61675440
Vega20 will have DDR4 PHYs on-die to allow it to work with VERY large compute datasets.

>>61675440
>GPU+GPU (whenever those are going to be done)
navi. in 2 years.

>>61675402
HBM2 ssd.

>>61675468
>navi
I hope they get the wattage down. I'm not looking forward to hearing 800w housefire. That and it's beyont the point of opening my door to let the a/c in.

>>61667604

(((content creators))) are already rubbing their hands

>>61675496
dont worry, nvidia's solution will also be a "housefire"

>>61675496
You simply clock ALUs lower when you have a LOT of them.

>>61675496
>I hope they get the wattage down
they will, shader count>clock speed.
think like this, an undervolted rx 480 pulls about 90 watts.
connect 4 of them through infinity fabric. now you have 144 cu behemoth, only pulling 360 watts with near perfect scaling.

File: 1501492630726.png (81KB, 769x670px) Image search: [Google]

81KB, 769x670px

>>61673203
(you)

>>61675532
You need a physical layer fast and wide enough to do that though.

>>61675563
raja says it is possible with infinity fabric, it is fast enough.
and navi will be 7nm. decrease in node size will help immensely also.

File: [HorribleSubs] Gabriel DropOut - 05 [720p].mkv_snapshot_02.48_[2017.02.06_14.00.42].jpg (114KB, 1280x720px) Image search: [Google]

114KB, 1280x720px

>>61675562
Anime website.

Btw 7nm LP is so hilariosly good Qualcomm will be fabbing *something* on it.

File: 1501528373829.jpg (157KB, 452x404px) Image search: [Google]

157KB, 452x404px

>>61675601
keep repeating it shutin, maybe you'll eventually convince your'reself

File: 1494872130773.jpg (141KB, 1280x720px) Image search: [Google]

141KB, 1280x720px

>>61675627
Anime website.

>>61675639
hold on, i'll try!
website without retards

File: 1453052666861.png (248KB, 500x495px) Image search: [Google]

248KB, 500x495px

>>61675654
Anime website.

>>61675664
i guess it didn't work, go figure

File: ahagon.gif (573KB, 400x225px) Image search: [Google]

573KB, 400x225px

>>61675673
Anime website.

>>61675687
not true

A newfag retard and a dweeb faggot ruining yet another thread wow !!!!!!!!!!

>>61675989
in my defense the thread was mostly dead by the time i started shitting it up

sorry though

>>61676050
It's a porn website, by the way.

>>61668684
This is fucking cold in machine learning and you actually have as big datasets as you possible can get.

I have an EC(sml) derivate from the campus I went to for instance that has been tuned over and over and over again to utilize the gpu, other shit too ofc but this one is kinda noteworthy as so many PhD students have worked on it to optimize it. You ofc code shit your self too

>>61672105
>He is the CEO
So they don't work?

>>61667604
Have fun in fucking debt

>>61677701
Are you retarded? Nobody but businesses are even going to think about buying one of these.

>>61675496
navi will be an mcm with IF literally a zen gpu
but the IF used will be the one that they currently use on epyc systems

the speeds of IF depends on the lanes since its pretty much a pcie 3.0 on bandwidth and scales linear...
also we know that there is a vega refresh coming before navi so i doubt we will see it in 2 years..

>>61678217
GMI is too narrow for GPU.
Navi will use custom, very wide low latency links, defintely not GMI.

>>61675562
>>61673192
The one bringing up anime wasn't even the one who posted with an anime picture.

>>61677788
He means AMD is in debt.

>>61678217
>also we know that there is a vega refresh coming before navi

They'll have a 32gb version with 4 HBM2 stacks(Vega 20), and two cut-down versions to replace Polaris (Vega 11, 12).

There won't be a refresh with fixed silicon per se. Not unless the extra features they put in Vega are so broken that they are impossible to use and require a new chip.

>>61680031

Welcome to the last 10 years of AMD's trading but hey, AMD bankrupt this time next year for reals right?

>>61668187
Example?

>>61680491

He has none. At a fundamental level all a cpu or gpu does is add, multiply, subtract and divide numbers.

>>61668421
>Their rendering farm is entirely Nvidia
not buying it that building would have burnt down

>>61682566
Fermi was long time ago, anon. Welcome to 2017.

>>61680763
True, but not even wrong.

>>61667713
But can it run crysis max specs in 4k at 120fps?

File: 1474398393917.jpg (56KB, 750x935px) Image search: [Google]

56KB, 750x935px

>>61671272
>>61668309

>pull a random number out of my ass for how many racks per tower for Radeon Pro SSG/Instinct GPUs
>20
>watch SIGGRAPH unveiling of Project 47 the following evening
>there's exactly 20 racks in that tower

mfw

>>61670850
>I don't really understand everything, but this is interesting to read
I used to read entire threads on /lit/ about philosophy books that I knew shit about, had such a great time.

>>61672816
just a few drinks

>>61675687
faggot