Professional Gpu / Content creation Thread

We tried, can't start discussion because it wasn't made by Nvidia.
Lets try again in 2 years.

>load immense work files into the GPU one time
>it can then access them instantly from that point on
>enormous videos can be edited in real time
>tremendous 3D cad files opened in seconds

Pretty goddamn big for a lot of people. It'll be impressive to see how 36CU Polaris 10 fairs in DP throughput unrestrained.

>>55768607
Pretty impressive desu

>>55768607
Is this only for video editing and rendering? Someone explain this shit

>>55768714
No i know people who use them to render huge 3d mapped spaces. The real time feed rendering is used in live scenarios too.

>>55768714
it's pretty good for 3d modeling as well, but it's pretty dumb to buy this for gaming.

File: squid.gif (617KB, 500x375px) Image search: [Google]

617KB, 500x375px

GO AMD GO

>>55768714
The first demo AMD is said to be showing is a use case for movie editing and cleanup on the GPU. What is the issue here you may ask, this is old hat and has been done on the CPU for years. Some GPUs can even assist it without slowing things down in the process, so what does SSG add? How about 8K movie streaming and cleanup in realtime. At 96FPS. Sure you can do this with traditional methods but the best of them will run the same task at 17FPS.

AMD is happy to point out this is a 5.6x speedup or so for the cost of two consumer SSDs. Before SSG, possible but slow. After SSG, fast enough for most users. The impossible, realtime 8K cleanup, is now possible. Hugely complex and highly detailed CAD models that took the better part of an hour to load up and decompress will still take the better part of an hour to load up and decompress on an SSG GPU based system. Why bother? Because it takes the better part of an hour to load and decompress the first time, then it can stay resident on the GPU’s flash storage. The second time it should take seconds. If you look at the cost of a modern automotive or aerospace engineer’s time, SSG is a no-brainer, any CTO would be foolish not to deploy this tech ASAP, the ROI would be measured in weeks.

>>55768607
>Why is no one talking about this?
Because about 95% of the people posting on /g/ only understand using GPUs for playing games and this isn't intended for that.

God watching the Nvidia and AMD fanboys bickering and being all round retarded is annoying.

>>55768607
>able to render the feed at approximately 17 frames per second, which is pretty respectable for 8K video

hnnnnng, how much though?

>>55768869
$10,000 for developer kits

>>55768869
>able to render the feed at approximately 17 frames per second, which is pretty respectable for 8K video
That's the current tech, AMD claim this can do it at 90

>>55768931
oh shit

>>55768922
OH SHIT

The percentage of people who would even make use of this is fucking miniscule .

>>55769070
And?

File: 05_i_001_53_c0_4551_1_b.jpg (19KB, 341x341px) Image search: [Google]

19KB, 341x341px

>>55769070
>It doesnt support muh videogames so it sucks!!!
Are Nvidiatards the most angry, autistic manchildren on the planet?

>>55769107
but can it render at SONIC SPEEDS?

>>55769070
its for film preduction you tard, not 4k 60fps gaymen at max settings and shit, its for work, not play

>>55768607
How the fuck are they allowed to call this thing solid state when it has a fan?

>>55768697

Do we know that its polaris based? Anandtech was merely guessing.

>>55768788

y halo thar charlie

Wasn't Intel doing something similar with crosspoint? What the fuck is happening with that btw?

>>55769207
Yes, it is.

>>55769185
if you put a fan on a ssd does that make the ssd mechanical ?

>>55769185

Why do you think its blue retard? To keep the computational niggawatts frozen so it doesn't heat up and cause a loss of rotational velocodensity.

Fuck sake /g/ is retarded at times.

Nobody is talking about it because it's not something 99.9% of us could make use of.

I do a lot of 3d modeling as an engineer, but even for me that card would be beyond overkill. My large and complex assemblies are many GB sure, but nowhere near 1TB. It's really geared towards people like Pixar animators so they can store insane amounts of file size on vram.

>>55769268
More like oil and gas companies where datasets are flipping enormous.

how much FPS in CSGO

>>55769235
It can't be solid state if it has any moving parts.

Just read about this.
A pretty nice leap in this tech.

I hope this eventually makes it's way to the consumer side.
For example if we can throw stroage into the same package with the GPU, we're not going to be needing those M.2 slots on the motherboard, or PCIe SSD cards.
Just buy the SSG with couple of TB of storage and you're set.
No need to go SSD shopping after that.

With this the whole system would get even more compact.
Not by much, but still these small steps are important.

>AMD
>Pro
Choose one

>>55769447
This is pointless for desktop users because the primary job of this technology is to free up congestion of the bus relieving the bottleneck.
Utterly useless for 99% people here.

>>55769447

Riddle me this, just how much cpu overhead would a system using these firepros need? Just thinking in a bit of a vaccuum one could use a very low power cpu and then go balls out on the gpu to save both cost of hardware and (potentially) power draw.

>>55769447
This SSD is not for regular storage, it is not a PCIe SSD like we have now

>>55769659

We don't really know that yet.
I mean this is just a couple of regular 950 Pros they're pairing with this via an M.2 slot, I don't really see why it couldn't function as a storage like any other PCIe SSD.

>>55769841
Because the M.2 slots are wired to the GPU directly and not the PCIe

>>55769441
god you must actually be retarded
how do you breathe

How many frames does it get in Ashes of the Singularity?

>>55769922
probably breathe the tears of stupid people he trolls on /g

>>55769945

All of them.

File: 1445297301531.jpg (78KB, 816x840px) Image search: [Google]

78KB, 816x840px

>>55768607
>AMD workstation GPUs have been historically garbage
>/g/ 2016
>It'll be different this time guise

>>55768950
You say that as though $10k is a lot for a high end workstation card.

>>55770116
It's a devkit holy shit not the retail card.

>>55770114
only because developers never fully optimize for OpenCL, they're too busy sucking CUDA dick. (as are you c;)

>>55770114
>>AMD workstation GPUs have been historically garbage
They've never been garbage, it's just all a matter of what the software is better optimised for, or the workload you're handling.
AMD have always had the edge in number crunching thanks to the way GCN is designed, whereas nVidia have the edge in tessellation (Kek?) due to CUDA. Both see wide adaptation of it in workloads or projects where such matters.

File: 1445294603347.jpg (38KB, 450x495px) Image search: [Google]

38KB, 450x495px

>>55770153
>>55770158
>T-They just never optimized for AMD those meanies >:(
Oh boy

>>55770153
>>55770158
Because instead of following the industry AMD makes their own convoluted shit that developers don't bother with. That's the main issue with them. It's like having a perfectly good highway and then building a shitty sideroad and getting upset that nobody uses it over the highway.

>>55770198
>someone states strengths of both brands as to compare them in a diverse environment

>SHILL, HHAHAHAHAHAHA, FUCKING POORFAG SHILL

Would you consider going back to your Battlestation thread? Or is there some thread back on /v/ that you would feel more comfortable in?

>>55769228
nvidia is content to let technology stagnate while they increase efficiency +3% year over year

File: 1458682473494.gif (721KB, 227x165px) Image search: [Google]

721KB, 227x165px

>>55768788
>The second time it should take seconds.
this would be some bomb ass shit

>>55770227
all i have to say about that is

>CUDA
>wasn't convoluted shit when it came out

>>55768607
>professional
>/g/

you can only choose one.

>>55770261
It wasn't, but you were born after it's release so you wouldn't know that.

>>55770241
he feels right at home here, shitposting anti-amd sentiment while ironically trying to seem superior. It works every time because AMD fanboys get so defensive

>>55770292
Implying fanboys.

Something like this SSG coming along makes a big difference to everyone, and is something new for the industry to move with.

The only fanboys here are the ones shitposting in an attempt to derail an otherwise rarely decent thread on /g/, where people actually discuss the technology and the possible implications as opposed to just calling everyone niggers, and sending a thread into a spiral of shitposts and traps.

>>55770324
shutup amdnigger

>>55770292
Stay on funnyjunk kiddo

>>55770153
They've been garbage long before any kind of gpgpu existed. The ati has no drivers is not just a meme.

>>55770287
But it was because it only works on Nvidia hardware, meanwhile OpenCL, which AMD pushes works on everything.

It seems the thread has gotten popular enough to where all the shit-posters are coming in. It was a good thread while it lasted guys.

:O Yeah but what video games does it support? :/ ._.

>>55770398
>Hey this thread was pretty nice before people came in here with different ideas
>Being this new
Lurk more before you post newfag. Also listen to this anon while you're at it >>55770366

>>55769396
Probably the same as a 480, if not less

>>55768607
Pajeet, blue is INTEL'S color
Lawsuit when?

Also, kek for Pajeet using regular SATA-based SSDs for their rendering storage and cache and not the BASED PCIe3 8x SSDs

>>55770552
>M.2 connected
>"SATA"
Please go back to /v/ and stay there.

>>55770552
>using regular SATA-based SSDs for their rendering storage and cache and not the BASED PCIe3 8x SSDs
Both are limited by the speed of the connection they use, the SSD is plenty fast enough in the SSG to feed it sufficiently, as demonstrated by that monstrous 90+fps in 8k

>>55770585
>http://www.tomshardware.com/news/amd-radeon-ssg-1tb-gpu-memory,32325.html
There are no such things as superfast 1TB M.2 SSDs yet. Quit lying, Pajeet

>>55768697
Yo, how the fuck do you pre-render a scene in AE again? Like, how do you get it to pre-load a clip so that you can scrub through it freely without it just loading whatever part you're on at the moment?

I know it's literally one button but I can't find the damn shortcut anywhere.

>>55768607
So should I get this instead of rx 480? Is this new pro series answer to 1080?

>>55770633
No, these are for workstation. Not bideogames you manchild.

hahaha but can I use for bideogames like the Nveedia??? :DDDD?

No?? Fugg AMD : D D

Normally, to communicate with a PCIe SSD, the GPU would have to go over the PCIe bus and through the chipset, which would add a ton of latency. Depending on how they wire it, putting it right on the GPU board skips all of that. So the GPU gets a giant, slow but not prohibitively slow pool of memory to work with for things like massive 8K video frames.

>>55770287
It was, it was also really early tech, at the time people used the other propietary niche API's and completely ignored ATI Stream and Nvidia CUDA
Howhever Nvidia did commit to CUDA, Stream was abandoned after two or three iterations, the other API's ported their shit to CUDA, then AMD started to commit to OpenCL, quite late in the game by then

>>55770681
>>55770407
>>55770337
>>55769396
>>55769226
>>55768769
Why are you doing this, Fag?

>>55770593
>http://www.tomshardware.com/news/samsung-sm961-1tb-nvme-m-2-ssd,32117.html
>http://www.thessdreview.com/our-reviews/ngff-m-2/toshiba-xg3-m-2-nvme-ssd-review-1tb/
>http://www.anandtech.com/show/10328/the-toshiba-ocz-rd400-pcie-ssd-review

Nice try mr pattel, but no dice...

POst the video where this thing does that 90hz rendering

>>55769841
Maybe it could, but its not meant to, just buy a PCIe SSD card for that, you'd be misusing the product

>>55770746
https://twitter.com/RadeonPro/status/757782561313566720?s=08

this can work for games, actually.

if a game supported it you would "install" the game to the gpu, it would take a long time copying all the textures to the gpu but then texture loading wouldn't cause bus traffic or involve the CPU (much).

>>55770738
All you've done here is prove the point that there is no need to fit a PCIe SSD to this thing, in that even at the speeds that M.2 and SATA SSDs reach, it is enough to feed this GPU with the speed it needs for 8k streaming and clean up.

>>55770799
It would be, like, a hard drive full of games.

amirit? :D :D :D :D EBIN

>>55770614
I think you didn't read anything. It runs 8k video at 96fps. A normal set up would suffer to do that in 1080p.

>>55770820
LATENCY
A
T
E
N
C
Y

speed has nothing to do with it. its all about clearing up the latency.

>>55770725

>>55770836
a hard drive full of games where its graphics would load with near zero latency and less cpu usage.

I'd be curious to know what sort of man hours this might save a company. I would guess it wouldn't have to be many all told to recover the initial cost of the hardware itself.

https://www.youtube.com/watch?v=vyKgT5QUU2Q

>>55770872
I was sort of trying to refer to that image of a facebook thread, where someone posted an image of his gpu, and, you know what, nevermind.

>>55770783
nigga thats really great and all, but those 5k fps when steeping back and forth in the progress bar made me laugh
>counting skipped frames

>>55770872
Think they might put out 256gb ssd versions of this ?

can't wait for the new mac pros to come shipped with these puppies

>>55770820
The point was pretty conclusively made in the demo AMD put up at SIGGRAPH 2016.

It's not about the speed of the SSD, but how fast the GPU can access it. Normally the GPU has no direct access to the SSD, instead the host (CPU) application has to first figure out that the GPU needs data off the SSD, read it from the SSD to RAM, then from RAM to the GPU and then let the GPU know that the data is available. With this the GPU can access an SSD directly, cutting out the need for the host application to do the time consuming lookup and copying to GPU memory, massively increasing performance in jobs where memory has to be constantly swapped in and out of GPU memory.

>>55770920
Would be interesting for smaller businesses or those who feel like they don't need quite THAT much room to play with.

Really they would be silly not too, it would open it up to a lot more customers.


As of right now, I imagine the producers at Pixar all have diamondillium hard dicks at the thought of upgrading their render farms to these things.

>>55770920
if it ever gets popular and makes it way into games.

its kinda like creatives "x-ram." it needs per application support. if the application doesn't have support for it, it cannot use it.

really for games it will probably never take off. latency isn't an issue in that arena.

>>55770263

>/g/aymen benchmarks
>what x product do I buy
>which Linux build is best for doing nothing productive and watching anime
>what do these numpad buttons do, I've never used them before

That's /g/ in a nutshell. I wouldn't be surprised if some people are wondering where all the Firestrike benchmarks for this card are.

>>55770968
Especially when developers have been extremely lazy with fucking SLI, something that directly benefits gaming. This has no foreseeable future in gaming until we see extremely resolutions become the norm, but other things would have also progressed by then to probably make this irrelevant

>>55771123

SLI is totally fucked because DX is setup in a way that multi-gpu requires throwing darts at a board and praying you hit bullseye to get anything beyond negative scaling. Plus Nvidia's jewry on bandwdith isn't helping.

How much VRAM does it come with normally? From what I have read it seems like the 1TB is the max you can add to it.

>>55770967
They probably have the first 100 or so preordered.

>>55768607
/g/ is just a bunch of /v/ fags in disguise. They don't care about technology, only about shilling.

>>55770477
ayy look, you proved my point, manchild

File: DeepinScreenshot20160726192902.png (9KB, 376x159px) Image search: [Google]

9KB, 376x159px

>>55770737
???

>>55770967
Pixar uses Nvidia.

>>55771674
pixar probably uses what's best, which right now is Nvidia

>>55769228
No matter how fast disk/ram gets, it still has to go through the pcie, then to the cpu, then through the pcie to get to the gpu. And that's only one way. It has to go through all that again if the gpu wants to change something. SSD right next to the gpu is a lot faster even if m2 speed aren't all that fast.

>>55771674
not for long :^)

>>55771180
it's a lot better under vulkan, but it seems like nvidia is phasing out SLI

at least AMD still supports crossfire

>>55768697
but transferring these files from an SSD to the card will still take ages

>>55771840
Yes but the point is that once they're there you have basically no latency to accessing them a second time.

>>55771123
dx12 and vulkan help remedy a lot of the complaints that developers had about sli / crossfire.

they have far more control this time around.

microsoft just released updates and new tool sets for developers to help implement multi-card support with dx12 too. microsoft new tools apparently make it easier to add in support than dx11 offered. it adds in another abstraction layer to the EMA that provides multi-gpu support with very minimal coding that will provide performance similar to decent dx11 scaling. if they wish to optimize it and gain the benefits of the low level api they still can and access even more performance that put dx11 multi-card to shame.

but of course, it comes down to developers not shitting on users.

ema is really cool. allows for both nvidia and amd gpus to work together, doesn't just have to be two of the same. keep that 290 and pair it up with a 1080.

>>55771807
nvidia isn't phasing out sli.
there is one thing you have to remember, the $200 market has been a bane to nvidia's existence. THEY HATE the $200 range.

as linus said himself with cheap & budget cards:
https://www.youtube.com/watch?v=sph6cjJeRdI
>they're not fun. they're not exciting.

nvidia doesn't care, nor want to care about the $200 price bracket. they want to push people up into the higher tiers. nvidia doesn't make much profit off their $250 segment in comparison to their $650 1080. don't get me wrong, they even hate the $350 market segment but they at least profit more off of it than they do with the $250 segment. if they were able to gut the $250 market the $250 market would be gutted next.

they simply don't want people buying $250 cards. they ditched sli on them not because they want to phase it out or because its "not worth it." actually, with the way the 1060 performs in in dx11, two would actually be cheaper and faster than a single 1080 and not use much more power and heat over it thanks to its 120 watt tdp. but that's the problem, there isn't that many negatives against utilizing two 1060's. with decent scaling it would achieve 1080 performance, in poor scaling slightly higher than a 1070. either way, a win win. that would chip away at their 1080 sales.

the 960's were absolute garbage in sli performance. actually, they were flat out garbage to begin with. barely faster than the 760 it replaced. two 960's were not even faster than single 980 in most games. it was barely faster than a 970. and for nvidia, that was perfect. they didn't have to worry about it chipping away at their higher tiers and it actually helped push people up.

970's where not worth sli'ing after the 980 ti came out due to similar performance. 980's where in a price bracket that didn't made them a terrible price+performance ratio, even before the 980 ti came out. but the 980 ti was a great price + performance and came with a price tag nvidia had orgasm's over.

>>55772642
>that didn't made them a terrible price+performance
that made them a terrible*

>>55771789
They just upgraded to Pascal.

>>55772642
>there isn't that many negatives against utilizing two 1060's.

cept for microstuttering

it's no surprise that problem was never really fixed nor that we haven't seen a beefy dual gpu card in a long time

>>55772642
its the whole reason why nvidia came out with the "x80 ti" and titan series.

if you think about it, and look back at history, the "980 ti" would of been the 980, the 980 would of been the 970, and the 970 would of been the 960. the "titan" would of been either the dual gpu on a single pcb or the "285 / 8800 ultra" type card.

the 1000 series would of been:
titan = dual gpu / x95 / x ultra
1080 ti = 1080
1080 = 1070
1060 = 1070

but they pushed everything up for higher profit margins.

if you want a worth while upgrade over that 290x or 780 ti you need to buy a 1080 for the price of $700. before, unless the generation was a refresh, the $350 would of been a worth while upgrade. we get this feeling of small upgrades. i saw a lot of people actually disappointed about the 1080 because it was only, on average, 25% faster than the 980 ti. but then we realize, actually we're not getting small upgrades, the bigger upgrades are just now on a higher tier that cost more.

and now nvidia tossed in founders editions to help keep prices inflated.

thanks nvidia. shows you really care about us consumers.

>>55772730
Shouldnt a gsync/adaptive sync monitor help mitigate microstutter? Especially under dx12, assuming it really does manage to itilize both cards better, considering the microstutter mostly stems from one gpu waiting for the next on occasional frames.

>>55772730
dx12 / vulkan fixes it.


> Multi-Threaded Performance: Compared to DirectX 11, which used single-threaded and multithreaded performance, Ashes of the Singularity will see huge gains in complex workloads. This will allow for significantly improved balance of workload between CPU cores.
> (Coming Soon) Explicit Multi-GPU Support: With DirectX 12 comes the ability to use multiple GPUs. This allows for better control of multi-GPU rendering by developer, thus controlling the frame. Without this ability, the integrated GPU would remain idle and unused as an untapped resource to increase rendering and speed up the frame.
> Asynchronous Shaders: This allows the schedule of work for the GPU that will be performed. Traditionally, the GPU’s command queue would have had stalls, and DirectX 12 essentially provides more work done for free.
> Explicit Frame Management: DirectX 12 will see a reduction in latency, which will provide a more responsive game experience. This will also allow tracking of specific information, such as whether something is GPU or CPU bound.
> Updated Memory Management Design: A radical change in the memory management design allows developers to remove traditional performance issues such as micro-stutters that have plagued D3D11.
DirectX 12 has taken great strides to prevent “screen tearing” by defaulting to having images rendered at the screen’s refresh rate (which would result in frame rate limiting). However, for benchmark purposes, we have taken steps to bypass this to give a more accurate representation of potential performance.
> MSAA is implemented differently on DirectX 12 than DirectX 11. Because it is so new, it has not been optimized yet by us or by the graphics vendors. During benchmarking, we recommend disabling MSAA until we (Oxide/Nidia/AMD/Microsoft) have had more time to assess best use cases.

>>55772797
see
>>55772833

>Updated Memory Management Design: A radical change in the memory management design allows developers to remove traditional performance issues such as micro-stutters that have plagued D3D11.

File: AMD_Radeon_ProDuo_CM.png (974KB, 1310x740px) Image search: [Google]

974KB, 1310x740px

>>55772730
Still king of the hill

>>55768607
>Why is no one talking about this?
/v/

>>55769441
The solidstate terminology was coined when electronics started moving from tube based to semi-conductors. We call solid state drives solid state not because they dont move, but because their memory is stored in silicon based ic's as apposed to magnetically.

>>55772867
>two meme fury gpu in a PCB