DirectX® 12 for Enthusiasts: Explicit Multiadapter

Sounds pretty great, if some games I play end up supporting DX12 + Explicit multi-adapter, I may (probably will) go SLI. With the integrated graphics doing physics/lightinig/etc. I wonder if the lesser CPUs like quad cores with integrated graphics will do better than the 6 and 8 core rigs with the same graphics card..
 
I just hope that there will be a lot of developers who implement this. It will be nice to finally put that dormant iGPU and bored PhysX card to use.
 
mem.png


Fury CF has already been doing very well, will be interesting to see how well it scales in DX 12.
 
This article is more of a best-case scenario that what we're likely to actually see. For example, if you split the screen in half both cards still need a lot of the same textures.

Yes, DirectX 12 is much better than DirectX 11. But unlike what AMD wants you to think it's not magical.
 
DracoNB said:
mem.png


Fury CF has already been doing very well, will be interesting to see how well it scales in DX 12.
Click to expand...

The pic on the right basically reinforces that this will get close to zero adoption. Anyone think most developers are going to create content for niche users that require different cached content on each GPU? Hell no. AMD has trouble even getting Crossfire support, and that has been around for many generations of GPU hardware.
 
ocellaris said:
The pic on the right basically reinforces that this will get close to zero adoption. Anyone think most developers are going to create content for niche users that require different cached content on each GPU? Hell no.
Click to expand...

If they want to get in on the VR game, yes, they have to.

Also, why the fuck not? That comment was borderline retarded.
 
DracoNB said:
mem.png


Fury CF has already been doing very well, will be interesting to see how well it scales in DX 12.
Click to expand...

I have little faith in this for how much it's being hyped. AFR and SFR are out the window as soon as you start considering this.

I imagine it will be exceedingly difficult to get very good, broad reaching performance gains. What happens when you get to a part of the scene where one card is holding most of those resources? Congrats your second card is now sitting there doing nothing. Now you get to either try and play fix up, or weather the storm.

If you load balance it somehow and then walk right out the door you came in, you're back at square one. Getting perfect memory scaling is probably super, super, unrealistically optimistic and I expect disappointment if people go in thinking it's going to work that awesomely.
 
Last edited:
Remon said:
If they want to get in on the VR game, yes, they have to.

Also, why the fuck not? That comment was borderline retarded.
Click to expand...

VR needs extremely similar content loaded for both eyes for common applications. In situations where content is different, managing multiple sets of resources in different locations is still a pain in the ass.

This is another "feature" that isn't going to go anywhere. AMD should focus on improving their overall user experience instead of making up more tech buzzwords that won't get supported.
 
ocellaris said:
VR needs extremely similar content loaded for both eyes for common applications. In situations where content is different, managing multiple sets of resources in different locations is still a pain in the ass.

This is another "feature" that isn't going to go anywhere. AMD should focus on improving their overall user experience instead of making up more tech buzzwords that won't get supported.
Click to expand...

You do know we're talking about dx12, right?
 
As far as I can tell, any way you slice it, each gpu needs to hold approximately the same textures as the other one. The only way 4+4=8 is if each gpu could have high bandwidth, low latency access to the other gpu's vram. Then it could be treated as a single large pool. I understand that theoretically that both gpus could be given two sets of different textures, but when is that going to actually happen, realistically.
 
So each GPU will render a tile of a frame, making the queuing non existent.

I still don't get how the whole 4+4=8? If each GPU renders part of a frame, don't they have to load all of the data and then figure out which GPU will render which part of the frame? Or the GPUs render predefined parts of the frame and only partial textures need to be loaded (this does not make sense)?

The only way 4+4=8 makes sense is through the increased draw calls between the CPU and GPU with DX12.
 
Maybe someone will figure out some genius implementation, but I can't see how it'll work


xorbe said:
The only way 4+4=8 is if each gpu could have high bandwidth, low latency access to the other gpu's vram.
Click to expand...

For now we have the PCI-E bus - which means trying to do that is a bad idea.

xorbe said:
I understand that theoretically that both gpus could be given two sets of different textures, but when is that going to actually happen, realistically.
Click to expand...

If you want to make this work, this is what you'll be doing. You'll have to guarantee that one GPU will not need the other's resources - as in one GPU will _only_ render certain things, and the other GPU will _only_ render the others.

Find a part of the scene where one GPU doesn't really have what it needs to render any significant bits and down the drain with performance scaling it goes.

KGBinUSA said:
So each GPU will render a tile of a frame, making the queuing non existent.

I still don't get how the whole 4+4=8? If each GPU renders part of a frame, don't they have to load all of the data and then figure out which GPU will render which part of the frame? Or the GPUs render predefined parts of the frame and only partial textures need to be loaded (this does not make sense)?
Click to expand...

You'll need to partition the scene so each GPU never renders something that the other will, and never needs resources from its partner.
 
This hole thing will be very interesting indeed. Also by the time this kicks off it will be the make or break moment for AMD. I suspect they will start to rebound by this time next year and hopefully well have a clear picture of just how well this technology plays out with games and the hardware built on the TSMC’s 16nm FinFET.
 
ocellaris said:
The pic on the right basically reinforces that this will get close to zero adoption. Anyone think most developers are going to create content for niche users that require different cached content on each GPU? Hell no. AMD has trouble even getting Crossfire support, and that has been around for many generations of GPU hardware.
Click to expand...

This is why Gameworks is so successful. Game developers won't go the extra mile to enable features like this. However if it's provided to them in a nice usable package they will use it.
 
PRIME1 said:
This is why Gameworks is so successful. Game developers won't go the extra mile to enable features like this. However if it's provided to them in a nice usable package they will use it.
Click to expand...

The fistfull of cash that gets handed to studio executives for integration doesn't hurt either!
 
KazeoHin said:
The fistfull of cash that gets handed to studio executives for integration doesn't hurt either!
Click to expand...

You're right, it doesn't. Because software development isn't cheap and heaven knows consumers sure aren't willing to pay for it.
 
jwcalla said:
You're right, it doesn't. Because software development isn't cheap and heaven knows consumers sure aren't willing to pay for it.
Click to expand...

awww... game designing is so hard... poor little multi-billion-dollar corporations...

woody-harrelson-zombieland-money-cry-gif.gif
 
socK said:
CIV implements SFR.

The slide literally says this isn't SFR.
Click to expand...

Its a more advanced version of SFR, but will utilize the same techniques. It will let you split out memory though unlike SFR.

I'm sure DICE will be using it in their next games, they are very willing and wanting to push graphics.
 
Quix said:
Yes, DirectX 12 is much better than DirectX 11. But unlike what AMD wants you to think it's not magical.
Click to expand...

In a way it is magical ;) very magical even Mantle got people moving in the right direction where gamers can expect rather big improvements. That we can't have them all on day 1 of release does not matter the changes that are coming are rather huge.

Several decades gaming on PC has been stifled by MS lack of innovation and leadership. This DX12 API allows developers to make use of all of the components in a PC regardless of vendor, leaving developers in charge to program features which they need to make their game better. There is no limiting factors any more for this if you want your game catered for 8 cores and SLI/CF you can now without being held back. And even better there is no special driver any more where you would always "need one" for SLI/CF.
 
DirectX 12 is out, Vulkan is not. Seems odd to say MS is stifling gaming on PCs with that in mind.
 
Pieter3dnow said:
In a way it is magical ;) very magical even Mantle got people moving in the right direction where gamers can expect rather big improvements. That we can't have them all on day 1 of release does not matter the changes that are coming are rather huge.

Several decades gaming on PC has been stifled by MS lack of innovation and leadership. This DX12 API allows developers to make use of all of the components in a PC regardless of vendor, leaving developers in charge to program features which they need to make their game better. There is no limiting factors any more for this if you want your game catered for 8 cores and SLI/CF you can now without being held back. And even better there is no special driver any more where you would always "need one" for SLI/CF.
Click to expand...

So your claim is that AFTER AMD announced "Mantle" and relased in alpha/beta-format (limited API, needs "handcrafting" to each GCN SKU) suddenly Microsoft did a nice "magic trick" and buildt a multivendor API from scatch and got it from Alpha to Beta to Gold faster than AMD....yeah....those magical pixie-fairies are beyond belief.

Your claim sounds like comming from someone who has never dealt with a API in his whole life.

A less far-fecthed senario is that Microsoft has been working on DX12 for years (known by NVIDIA, Intel...and AMD) and someone at AMD decided to make some PR FUD, try and steal the spotlight, goobled up by hungry fanboys and spread like gospel on forums.

Try and figure out why Skylake's IGP is DX12_1 feature level.
How long do you think Skylake was in the pipeline?
If you say "about a year", you will only confirm my suspicion that you have a very shallow knowlegde of the topic.

So somehow both NVIDIA and Intel "magically" guessed the specifications of DX12_1 feature level (independant of each other) way in advanced of Microsoft doing any work on DX12...you know how ridiculous that sounds?

You got any more joke like that you want to share? :D:D:D:D
 
DX12 may have been "talked about" for years, according to Nvidia, however Ms has been dragging their feet with reagrds to gaming support.

Just look at Games for Windows live if you have any doubt.

AMD just g ot fed up and did it themselves to light a fire under MS's ass.

They said this from the very beginning before mantle was even released that they didnt care if Mantle survived, they just wanted to push the tech forward and hoped someone would take it and run with it (Vulkan obviously did)

MS finally got their head out of their proverbial backside, and got DX12 up and running. Mantle is no longer necessary once those games start to release or current games updated ( i believe CDPR said they would patch in DX12 to witcher 3)

It did its job however.

The half a dozen or so games currently with mantle run extraordinarily well (the buggy mess that is BF4 aside)

Even on my somewhat high end rig, i routinely saw 10fps+ uplifts, and they were way more smooth than their DX counterparts.
 
Last edited:
Dragging their feet? MS could not put it into Windows 8 because it requires WDDM 2, which means too many kernel change for an update to that OS. MS probably always intended to put it into Windows 10. And what does Games for Windows live have to do with anything, that's completely optional and unrelated.
 
it relates to their general apathy towards gaming.

yes, dragging their feet.

DX12 doesnt "need" wddm 2.0, Mantle and Vulkan pretty much prove that.

They have chosen to tie it to WDDM 2.0 as a selling feature.
 
yes, dragging their feet, they got DX12 out before Vulkan, I guess Vulkan wants in on those xbox sales, so they're dragging their feet too? Again harping on about GfWL is just trying to score cheap points because it's not popular, it is not something that affects the core of games or anything close to it. It's like saying MS is holding back audio by not updating Direct Music, it's unnecessary and superfluous.

Oh yea, MS is absolutely posting news paper ads about WDDM 2. News flash, hardly anybody has even heard of it. Their version of this requires WDDM 2, maybe for better features, ease of development or additional features. Whatever it is, it is still required for their version of this.
 
Last edited:
i never said they were holding it back, i was talking about their general attitude.

The disaster that is GFWL is just an example.

Theve been apathetic until recently, surprisingly until right after mantle was announced.

This isnt new, people have been harping about it for years.
 
Actually,
DX12 may have been "talked about" for years, according to Nvidia, however Ms has been dragging their feet with reagrds to gaming support.

Just look at Games for Windows live if you have any doubt.
Click to expand...
is what you said. Kinda the same thing as saying it's holding back gaming.

They didn't release anything about DX12 because they don't announce features for new OSes far in advance of the OS. You're assuming things.
 
The Mac said:
DX12 doesnt "need" wddm 2.0
Click to expand...
It actually does. Major new features in DX12 are dependent on structural changes to the driver model itself. It's true that MS could support a SM version bump and offer more efficient APIs on earlier WDDM versions, but that's only part of what DX12 is.

I think there's a better argument of whether DX11.3 needs WDDM 2.0 if it's ultimately not available on Win8.x.
 
The Mac said:
DX12 may have been "talked about" for years, according to Nvidia, however Ms has been dragging their feet with reagrds to gaming support.

Just look at Games for Windows live if you have any doubt.

AMD just g ot fed up and did it themselves to light a fire under MS's ass.

They said this from the very beginning before mantle was even released that they didnt care if Mantle survived, they just wanted to push the tech forward and hoped someone would take it and run with it (Vulkan obviously did)

MS finally got their head out of their proverbial backside, and got DX12 up and running. Mantle is no longer necessary once those games start to release or current games updated ( i believe CDPR said they would patch in DX12 to witcher 3)

It did its job however.

The half a dozen or so games currently with mantle run extraordinarily well (the buggy mess that is BF4 aside)

Even on my somewhat high end rig, i routinely saw 10fps+ uplifts, and they were way more smooth than their DX counterparts.
Click to expand...

You didn't adress any of my points, you just repeated the PR FUD I mentioned.

Again, explain that Skylake and Maxwell is DX12_1 Feature Level...reality dosn't agree with your claims.

Funny, how Microsoft supposedly "dragged their feet"...when they beat both Mantle and Vulcan in going "Gold" ...your claim makes AMD look even more incompetent....good job.
 
Factum said:
So your claim is that AFTER AMD announced "Mantle" and relased in alpha/beta-format (limited API, needs "handcrafting" to each GCN SKU) suddenly Microsoft did a nice "magic trick" and buildt a multivendor API from scatch and got it from Alpha to Beta to Gold faster than AMD....yeah....those magical pixie-fairies are beyond belief.

Your claim sounds like comming from someone who has never dealt with a API in his whole life.

A less far-fecthed senario is that Microsoft has been working on DX12 for years (known by NVIDIA, Intel...and AMD) and someone at AMD decided to make some PR FUD, try and steal the spotlight, goobled up by hungry fanboys and spread like gospel on forums.

Try and figure out why Skylake's IGP is DX12_1 feature level.
How long do you think Skylake was in the pipeline?
If you say "about a year", you will only confirm my suspicion that you have a very shallow knowlegde of the topic.

So somehow both NVIDIA and Intel "magically" guessed the specifications of DX12_1 feature level (independant of each other) way in advanced of Microsoft doing any work on DX12...you know how ridiculous that sounds?

You got any more joke like that you want to share? :D:D:D:D
Click to expand...


Please don't quote him I have him on ignore for this reason, same exact conversation that went on for pages about this and low level programming......
 
razor1 said:
Please don't quote him I have him on ignore for this reason, same exact conversation that went on for pages in another thread
Click to expand...

I understand your feelings, but posts based on ignorance needs to be countered before the FUD spreads.
But I will try and avoid quoting him in the future, people should be able to figure out who I am countering.
 
Factum said:
I understand your feelings, but posts based on ignorance needs to be countered before the FUD spreads.
But I will try and avoid quoting him in the future, people should be able to figure out who I am countering.
Click to expand...


true,
 
Multi-gpu left up to the game Devs? Meaning this goes nowhere outside of a few outlier examples.

What ever happened to Lucid Hydra anyways? Well besides sucking...
 
Factum said:
Your claim sounds like comming from someone who has never dealt with a API in his whole life.

A less far-fecthed senario is that Microsoft has been working on DX12 for years (known by NVIDIA, Intel...and AMD) and someone at AMD decided to make some PR FUD, try and steal the spotlight, goobled up by hungry fanboys and spread like gospel on forums.

Try and figure out why Skylake's IGP is DX12_1 feature level.
How long do you think Skylake was in the pipeline?
If you say "about a year", you will only confirm my suspicion that you have a very shallow knowlegde of the topic.

So somehow both NVIDIA and Intel "magically" guessed the specifications of DX12_1 feature level (independant of each other) way in advanced of Microsoft doing any work on DX12...you know how ridiculous that sounds?

You got any more joke like that you want to share? :D:D:D:D
Click to expand...
When multiple developers working with Mantle joke about the DX12 SDK looking suspiciously like Mantle on Twitter this is FUD?

You pretend that MS had things ready on DX12 which is not the case in interviews with AMD in the BSN youtube showing Mantle developers talk AMD rep stated he clearly asked devs from DICE and Oxide never heard of a low level API from MS at the time AMD approached them for what is now known as Mantle .

When you are talking out of your ass about skylake IGP and DX12 support is just that Nvidia and AMD both have older part supporting DX12. "any" card or IGP can have DX12 drivers the features or support is all up to the vendor supporting this (hardware or software).

AMD told Nvidia they could use Mantle so the multi vendor API is not that surprising.

Factum said:
You didn't adress any of my points, you just repeated the PR FUD I mentioned.

Again, explain that Skylake and Maxwell is DX12_1 Feature Level...reality dosn't agree with your claims.

Funny, how Microsoft supposedly "dragged their feet"...when they beat both Mantle and Vulcan in going "Gold" ...your claim makes AMD look even more incompetent....good job.
Click to expand...


FUD more FUD from FuDUM...
 
A lot of time "threading the needle" ends up with quite similar looking concepts concepts.
The major difference under the hood is going from a primarily singlethreaded API to a much more multithreaded API.

And if it is so "similar" how come Mantle only works on a select number of GCN's SKU's, while DirectX 12 run on any DX11+ NVIDA SKU, Any Intel Haswll, Broadwell and Skylake SKU and on any AMD GCN SKU.
DirectX 12 run on more AMD SKU's than Mantle does.
You should take a look at the driver level again, might be an interesting experience to you.

And once again:
"So your claim is that AFTER AMD announced "Mantle" and relased in alpha/beta-format (limited API, needs "handcrafting" to each GCN SKU) suddenly Microsoft did a nice "magic trick" and buildt a multivendor API from scatch and got it from Alpha to Beta to Gold faster than AMD....yeah....those magical pixie-fairies are beyond belief..."

You also might want to look at what DirectX 12_1 Feature Level entails.
There are some good "gems" here:
https://software.intel.com/sites/de...75/Deep Shading Buffers on Commodity GPUs.pdf

(Hint: Stuff dosn't just appear "out of the blue")

And let us not forget this:
http://www.tomshardware.com/news/intel-amd-mantle,27121.html

Care to tell me when Intel got access to Mantle? :)
 
You must log in or register to reply here.
Back
Top