DirectX 12/Vulkan Thread

File: 100l[1].jpg (153KB, 1457x817px) Image search: [Google]

153KB, 1457x817px

GPUs on old APIs

File: 100m[1].jpg (135KB, 1465x819px) Image search: [Google]

135KB, 1465x819px

GPUs on new APIs

>>52583516
>>52583525
>>52583568
Guess we'll just wait and see.

Great time for gaming is comeing

>>52584112
>wait and see

Spoken like a true AMD fanboy

>>52583516
>Is Microsoft developing any features for DirectX 12 that will NOT be available in Vulkan?
I don't think you realize what Vulkan/DX12 is.

File: macaroni2.png (434KB, 1228x1062px) Image search: [Google]

434KB, 1228x1062px

>>52585008

File: macaroni3.png (468KB, 1228x1062px) Image search: [Google]

468KB, 1228x1062px

>>52586495

File: images[1].jpg (5KB, 223x226px) Image search: [Google]

5KB, 223x226px

>>52586940
>Jim keller is NOT stellar

>>52583516
I find it weird too that MS doesn't choose to offload "Present to user" to a different core, considering developers will be lazy and dump the majority on the first pipe by habit. Could they change it to the 3rd or 7th core instead, or does it rely on being on the first pipe?

>>52588603
"Present to user" in this case would be the image upload to the monitor via the framebuffer swap, plus a few other nuances.

>>52583516
devs are retarded and won't use dx12 features when they could, they especially won't use vulkan at all unless someone pays them

File: Medusa_s_first_time_with_Vulkan.jpg (128KB, 736x981px) Image search: [Google]

128KB, 736x981px

Soon

>>52590120
>they especially won't use vulkan at all unless someone pays them
wat. Pajeeeeeeet no. Is there any place you will not shill?

>>52591946
Simply ebin(a)

File: 123123123123123.jpg (3KB, 160x160px) Image search: [Google]

3KB, 160x160px

>>52585008
>Waiting on how new technology we have limited knowledge about will actually turn out in the future makes you a fanboy

Here's your reply

>>52583516
>Will 8-core FX chips be revitalized with DX12's release?
no because the current AMD cpu with 8 cores are simply 4 cores with hyperthreading.

Call me when modern games are any good to give a shit about this.

>All of this circlejerk about muh optimization when all you will get with your 3000 dollar gaymen rig will be half assed console ports with tech from 10 years ago.
:v)

>>52595276
Rendering APIs are used everywhere, not just in modern games. Explicit control over memory/transfers/caches/resolves will enable improved memory efficiency and lower energy consumption everywhere, for example in browsers. This is important everywhere but especially important on mobile.

>>52595276
Go back to your containment thread, hipster.

>>52595250
Tech-illiterate or Intel shill.

>>52583516
>Is Microsoft developing any features for DirectX 12 that will NOT be available in Vulkan?

DirectX 12 is already out

File: 8c1.jpg (7KB, 200x263px) Image search: [Google]

7KB, 200x263px

>>52583516
>windows 10

>Is Microsoft developing any features for DirectX 12 that will NOT be available in Vulkan?

They are different APIs, you can't compare them like that.

>Will 8-core FX chips be revitalized with DX12's release?

DX12 has been out since July so you tell me.

>Does Vulkan stand a chance of defeating DX12 in the Windows space?

No.

>Will the industry quickly adopt these new APIs?

Yes, Unity and UE4 will/already do support DX12 and have said they will support Vulkan.

>Will this finally kill bad PC port performance?

lol

>>52583516
>>52583525
>>52583568
all that microsoft shilling

>LOOK GUYS HOW AWESOME DX12 IS LOOK AT THIS
>OH YEAH DON'T FORGET TO UPGRADE TO WINBLOW 10 BECAUSE IT ONLY RUNS ON THAT FOR REASONS. FUCK YOU

>vulkan
lol

>>52595711
You forgot to mention:
>vulkan
>gets 40% performance compared to dx12

>>52595711
>only the latest windows version supports the latest DX
gee, you were probably also upset that Widows XP didn't get DX10

>>52595775
>how to spot a m$hill

>>52595770
nobody will use vulkan, it's gonna be 99% dx12

>>52595815
>I can't contest his point
>better call him a shill

>>52595830
Dx12 costs money to license. There are going to be a ton of people using Vulkan, even if it's performance is terrible compared to dx12.

>>52595835
There is no reason that all these features have to be rolled out together. On a proper OS (i.e. Debian), you can update individual components without needing to totally reinstall the system. Updating the graphics API does not require changing your DE, reverting your settings, and installing bloat. Updating the graphics API does not change anything except the graphics API.

Windows does not allow that. This may be because Windows is a pile of shit and literally cannot work like that, or it may be because M$ is evil. Either way, on a Linux system parts can be updated individually while on a Windows system you have to regularly install big monolithic upgrades.

>>52595884
>Dx12 costs money to license
drop in a bucket compared to the cost of making a game.

look how well shit like opengl went.

>>52595922
That's cause OpenGL is fucking garbage. Vulkan will probably not be fucking garbage.

And people DO use OpenGL. Small studios use it all the time, as do hobbyists. All AAA games use DirectX, but that doesn't mean OpenGL is dead.

>>52595954
>That's cause OpenGL is fucking garbage
This guy is posting in another thread how opencl is better than cuda right now

AMD cocksuckers are a waste of air.

>>52595922
There is a massive indy game market. For example, Factorio runs on an open source engine which uses OpenGL and has made $1,700,000 in sales over the last 2 years without even being on Steam.

Vulkan will vastly improve the capabilities of indy games, and we could see a huge leap in their quality 4 or 5 years after Vulkan is released and it's used in open source game engines.

File: amd shill.png (729KB, 1100x1002px) Image search: [Google]

729KB, 1100x1002px

>>52586940

>>52596001
>Vulkan will vastly improve the capabilities of indy games, and we could see a huge leap in their quality


a bad workman always blames his tools

>>52595557
http://arstechnica.com/gadgets/2015/11/amd-sued-over-allegedly-misleading-bulldozer-core-count/
tell it to the judge

>>52595977
>This guy is posting in another thread how opencl is better than cuda right now
Are you retarded? OpenCL is not the same as OpenGL. They have no relation except the name.

And OpenCL IS better than CUDA. OpenCL can work with many products from many vendors. It can work with GPUs, CPUs, DSPs, and even FPGAs. This means you can use multiple processors in one program more easily, while with CUDA the GPU and CPU components must be clearly divided. OpenCL offers a more versatile command queue, more low-level control of the device, and run-time code generation. And last but certainly not least, OpenCL is an open spec while CUDA is not.

>>52596067
>better tools in a person's hand doesn't improve his product

Ah yes, let's ignored 10s of thousands of years of human history so you can win an Internet argument.

>>52596105
Was thrown out in court because as expected it was just some inbred trying to stir shit up.

>>52596107
An OpenCL program on an AMD GPU is also consistently faster than an equivalent CUDA program on an nvidia GPU. the OpenCL program on the nvidia gpu performs like dogshit, meanwhile (i.e. nvidia gimp their OpenCL support on purpose).

>>52595954
Virtually all scientific applications use opengl, too. So do most engineering applications (but not all).

File: L24cy.gif (1MB, 290x189px) Image search: [Google]

1MB, 290x189px

>>52596173
>An OpenCL program on an AMD GPU is also consistently faster than an equivalent CUDA program on an nvidia GPU

>>52596150
AMD babby pls go

>>52583516
>Will 8-core FX chips be revitalized with DX12's release?
No, it'll take years before this is used propperly.
>Does Vulkan stand a chance of defeating DX12 in the Windows space?
Of course not.
>Will the industry quickly adopt these new APIs?
of course not. dx9 is still a thing.
>Will this finally kill bad PC port performance?
No, it might help a bit tho.

>>52596029
>at least he gets paid :^)

>>52596105
Someone filed a frivolous lawsuit. Oh, boy. You sure showed me!

It isn't hyperthreading. Each thread has dedicated ALUs, while with hyperthreading they are shared. Each thread runs independently and has dedicated hardware. Each module has two integer cores. This meaning of "core" has been industry-standard since 1996, when DEC introduced the Alpha 21264. Go sue DEC if it's really that misleading rather than suing AMD for following the industry-standard terminology.

The cores DO share an FPU, but that can't possibly be part of the definition of a core. After all, in the past many CPUs had no FPU at all. Many CPUs STILL have no FPU in the embedded world. Are those all ZERO-core?

And if sharing resources is forbidden, what about all those shared resources on Intel chips? Are all Intel's CPUs single-core because all the cores share the hardware RNG and cryptocoprocessor?

And this is rich:
>The suit claims that Bulldozer's design means its cores cannot work independently, and as a result, cannot perform eight instructions simultaneously and independently
Even a REAL four-core chip can execute eight instructions simultaneously. ALL modern desktop CPUs are superscalar.

>>52596245
>using the smiley with a carat nose

>>52596173
That's not true. OpenCL on AMD is competitive with CUDA on Nvidia.

Maybe OpenCL on AMD beats OpenCL on Nvidia, but I haven't seen any information on that.

>>52596183
OpenGL? Why? Just for visualization, or do they actually use it for processing the data?

>>52596281
It does, Nvidia support for OpenCL is abysmal.

>>52596184
Found the shill. Case in point: shitcoin mining.

>>52595954
But I dont want to play indie games

Guess I'm going to get Windows 10 when I have a new PC in a week or 2

>>52596281
You would never use opengl over opencl to process data (and nobody uses opencl over cuda to process data anyway because nvidia gives GPUs to everyone for free and amd has little to no support for any of their shit). For visualization and graphical simulations (typically, the simulation is done in cuda/opencl and the result is rendered with opengl).

>>52596253
it works the same as hyperthreading, and take your shilling elsewhere.
bulldozer/piledriver functions in the same league as intel 4-core cpus with the exception of the 25% decreased performance due to shitty architecture.
If compared to an actual 8 core processor we see that the intel cpus have more than double the performance over the AMD cpus.
And if we look at the amount of watts the AMD cpus eat it's almost three times as much as their intel ""equivalent""

>>52596362
>cucking yourself willingly
OK kid

>>52596355
>OpenCL
>faster than CUDA
keep dreaming m8
there is reason nvidia doesn't give a fuck about OCL support on their cards

>>52596405
The shilling is real.

File: foxgrapes.jpg (234KB, 688x1434px) Image search: [Google]

234KB, 688x1434px

>>52596433
>AMD is all I can afford with my autismbux so it's best xDDD

>>52596423
That reason is that nobody would be using cuda if they did. After all, the same OpenCL program can run on CPUs, GPUs, FPGAs, APUs; phones, tablets, workstations, desktops, servers; and it's all opensource.

>>52596461
>it's all opensource.
so it's shit huh

>>52596405
Are you actually trying to win an argument where your position is objectively wrong, out are you just shitposting? I'd really like to know why people like you post at all.

>>52596454
Where do I sign up and what's the starting salary? I want to be like you senpai!

>>52596433
If I really were to shill like that AMD guy I would format my text.

>>52596473
>being this ass frustrated
:^)

>>52596461
>open source
No anon; open standard. Very different

>>52596469
Seems like I actually won the arguement here.
>MUH SHILL
>y-y-you a-a-are w-w-wrong

>>52596362
No reason to do that. Basically no games are using DX12 yet. A few years down the line you'll NEED DX12 support, but for now 11 is good for the vast majority of gaming. Hell, you can get by on 9 most of the time.

>>52596405
>it works the same as hyperthreading
>dedicated ALUs are the same as shared ALUs
OK, kid.

>>52596405
>in the same league as intel 4-core cpus
Who cares? There are 8-core ARM chips that are slower than dual-core Intel ones. That doesn't mean they don't have 8 cores.

They're slow, but they do have as many cores as advertised. I'm not defending Bulldozer, but the terminology is correct.

>>52596572
If you care at all about how a game looks and performs, you want windows 10, since it scores better in benchmarks.

>>52596478
I ain't no shill. I don't own a single AMD product. I just have a clue what I'm talking about, while you obviously do not.

>>52596604
>since it scores better in benchmarks.
SLIGHTLY better. 10 is better for gaming, but it is not so much better that it's mandatory.

>>52596547
No, you didn't. See >>52596253
Hyperthreading is a duplicated register file that can be loaded by the memory fetch while another thread is executing is time slice. When that time slice is over, or a cache miss occurs, the CPU can dunno the pipeline and switch to the second register file.

A hyperthread cannot perform a second "operation" simultaneously (ignoring superscalar) at once. A module core can.

Intel being 50% faster than AMD per clock is not relevant to the argument.

>>52596487
Please tell me, I've always dreamed of having an easy job I could do from home.

>>52596643
s/dunno/dump

File: whatareyoudoing.jpg (18KB, 366x380px) Image search: [Google]

18KB, 366x380px

>>52596405
>This guy
>MFW

Hyper Threading (or SMT as known elsewhere):
>Two threads, one core
>In essence a single hardware core with two threads
>Performance gain comes from having that one core share it's time intelligently between the two threads (i.e when one thread stalls due to memory access or something the other can use the otherwise wasted time by running)
>No real duplication of compute resources

In contrast, what AMD did with the bulldozer cores:
>Cores come in blocks of two
>Takes in a single thread
>Dispatches this single thread to these two cores
>Cores however share certain resources like the FPU
>With newer silicon there's fewer shared resources

In short: /v/ pls go...

>>52596604
Worse*
ftfy
3% worse than win8, which itself is 2% worse than win7.
On top of that, tons of games on win10 have input delay and stutters.

>>52596664
Your module explanation is wrong. It is 2 cores. The fpu is shared for 256 bit fmac operations, but split for 128 bit (32 bit vs 64 bit simd)

>>52596670
Not with any benchmarks. Witcher 3 specifically could get 10 fps more in Windows 10 than Windows 7. Battlefield 4 ran better on 8 than it did with 7.

Windows 7 is now 2 DirectX versions behind.

AMD CPUs are much faster than Intel CPUs at the same cost.

Let's look at AMDs high end:
http://pcpartpicker.com/p/LcC9D3
It only costs $200 and can easily OCed to more than 4+ GHz.

For the same price even by taking only the cheapest H170 board and 8GB DDR4 RAM you can't even get a Pentium G4500.
http://pcpartpicker.com/p/26HkVn

>>52596701
Well seeing how 256 bit and bigger FMAC's are the norm by now the individual 128 bit ones are hardly relevant. Sure, you can do some minor floating point math, but any even slightly bigger number crunching is going to be done on a 256 bit or larger FMAC.

>>52596788
You are right. It was a bad gamble. Doesn't change the fact though.

>>52596788
>Well seeing how 256 bit and bigger FMAC's are the norm by now the individual 128 bit ones are hardly relevant.
Correct. Bulldozer sucked. But it wasn't false advertising.

>>52583516

>Does Vulkan stand a chance of defeating DX12 in the Windows space?
Yes
>Will the industry quickly adopt these new APIs?
Yes
>Will this finally kill bad PC port performance?
No

>>52596760
[shilling and reality denial intensifies]

>>52595884
That doesn't matter when Microsoft will be sponsoring it.

>>52596253
>The cores DO share an FPU, but that can't possibly be part of the definition of a core.
They don't even do that, each core has its own 128-bit FPU which can be co-opted by the other core in the module to do 256-bit operations.

If anyone wants to argue that because each integer core doesn't have its own 256-bit FPU that they aren't cores, then you're trying to change the definition of decades of CPU design and you're opening the door to billions of dollars in lawsuits over both intel and AMD advertising their chips as dual or quad or even single core chips when they wouldn't be by that definition.

>>52596184
But he is right, AMD is way better at computing than Nvidia, don't you remember the flood of autist mining memecoins and raising the prices of amd gpus?

File: RK5bf35.jpg (167KB, 1000x1000px) Image search: [Google]

167KB, 1000x1000px

>>52586940

>>52592459
Here's why I think 2016 will be the year OSX gets no games.

>industry has mandated a switch to Vulkan/DX12
>old APIs are usable, but so horrible and inefficient in comparison
>developers have just finished Linux/Windows releases of all their games, maybe even an Android release (85% of the market)
>They decide to go for OSX and get that last 5%
>tfw they realize they have to re-write their entire game, de-optimize everything and have it run like shit, and hack it for months to make it work almost as good as it did with Vulkan on sensible PCs
>they give up because OSX has no game development options anymore

>>52595659
>They are different APIs
They are literally the exact same thing, with different branches.

Both started as Mantle and have nearly identical functions, manuals, and capabilities. AMD donated to both of them simultaneously in secret.

>>52598449
>both of them
>in secret
C U C K S

>>52595711
>accuses everyone of microsoft shilling
>laughs at vulkan

>>52596067
So does someone with broken tools

>>52595916
Note that opengl runs on older OSs and has exactly the same features as DX11; it was never an OS issue.

>>52598449
>in secret
No, it was pretty fucking obvious

>>52598424
You mean the year they adopt vulkan like everyone else?

You forget that they are at a top in the Khronos group.

DX12 will not take off. It will only be available on Windows 10 which has about 3% adoption rate.

>>52583516
Vulkan will lose to DX12 simply because of all the crossplatform support. Some older OS'es are just not efficient and will hold it back.
>>52592459
>Windows XP
kek. Into the trash it goes.

>>52600046
Found the M$ pajee- I mean, employee!

File: 1453510916997.jpg (270KB, 836x750px) Image search: [Google]

270KB, 836x750px

>>52586940
>Jim Keller is not stellar