-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
DirectX 12 GPU exclusive features to be shown/announced at GDC
- Thread starter Nirolak
- Start date
To be honest, the most Im expecting, are framerate improvements, in open world games like assansins creed and GTA5, if AC:unity was developed on DX12 it might of been a locked 30fps.
We are honestly bringing up D-GPU? Why are you trying to turn this into a strawman argument?
Move engine and double read/write ESRAM is not fabled it is actually part of the real hardware.
They aren't being coy, Phil Spencer said it does support DX12 and I don't consider that being coy.
The Shogun
Banned
Yeah I'm not sure exactly how all that works. I'm a PC gamer and know a decent amount about the tech itself but really nothing at all in terms of how this helps devs.
To anyone privy on this stuff do you think MS claims of big performance increases has any weight with DX12? Not just on Xbone but in the PC market?
So why hid this future hardware features from developers even in your own SDK documentation?
On the PC market... yes.
On Xbox One Phil Spencer specifically said it will not show a "massive" improvement.
When I say knew all along I mean when they initially sat down to figure out their plans for DX12, Nvidia without a doubt knows now what is needed as they are currently designing hardware for it.
DX12 isn't finished yet and isn't even in beta yet.
They did announce them for NVIDIA in Maxwell presentation back in september.
becuase that SDK didnt support DX12 yet?
Dictator93
Member
1. Microsoft talked about it having something like 200GB/s right? Actual real world answers and logic turned that in to like 120 or so... and only for stuff going through ESRAM.
2. Support DX12 does not mean having DX12 hardware features. If no AMD cards developed and designed, before, during, or after the xb1 have dx12 hardware support. Why on earth would the xb1? You are completely overestimating the designing of the GPU internals that occurs at a place like MS. They "shop around" for parts and then try and put tiny inputs into the things development (like the small stuff cerny did). They do not design these GPUs, you know how massive a task it is to design a modern GPU? Especially entirely new feature sets for a GPU?
Everythign we know about the xb1 design is that it was not about power, not about features, not about performance, but about cost cutting and manufacture control.
They already have a hardware specific version of DX on the xb1.
UnrealEck
Member
I'm wondering if Xbox One has full DirectX 12 feature support and AMD's 200 series didn't, why didn't it? If Xbox One was in development with DX12 support being included, why would AMD's 200 cards not have had it? Or nVidia's Maxwell. I guess maybe a deal could have been done between AMD and Microsoft, leaving nVidia out for their Kepler and Maxwell cards but that'd be a bit of a stretch I think.
Angelus Errare
Banned
I was being sarcastic. I expect nothing to change in 2016 except MisterX's story.
Many other things are shown as "future update feature".
In everything there are theoretical maximums.
A cars engine dyno is the theoretical maximum for car performance but real world performance completely varies. Same goes with this technology, just because theoretical data doesn't match real performance doesn't make it mythical.
You completely missed every article and breakdown on the Xbox One SoC design. They put heavy design into the Xbox One, they also started building their own hardware chips at the same time.(All the chips and hardware inside the Kinect are designed in house by Microsoft.)
Xbox One already uses DX12 features.
If it was built for the idea of cost cutting and manufacturing control it wouldn't be the largest SoC ever mass-produced.
Microsoft designed the Xbox One SoC based on AMD parts.
Dictator93
Member
It is the largest SOC ever built because ESRAM takes up ridiculous amounts of diespace. Its size is directly as a result of them being cheap / wanting to control its production. Its die space/ performance dividend is hilariously low. It (the ESRAM taking up massive die space) is only there because they cheaped out (DDR3) and wanted to control shrink production in the future (no EDRAM daughter die!).
The original figure given was mythical and basically impossible for all intents and purposes in game development. It was just thrown out there to obfuscate the power differential between PS4 and xb1. Same with the whole "5 billion transistors number."
EDIT: look at the god damn space all that "oh shit we fucked up" ESRAM takes up:
So if I understand this correctly, my GTX 970 will still get the performance improvements from DX 12, but it is possible I will miss out on some other new features in DX 12? The performance improvements is the thing that interests me the most, so I am ok if that is the case (plus, developers will not utilize those specific DX 12 features a lot if only a handful of pc gamers have capable cards).
You mean like every mythical theoretical maximum talked about in every field of technology?
I don't understand why we got here from talking about DX12 support but I'm think it has something to do with trying to derail the thread by saying Microsoft is too cheap to even support their own software on the hardware they put heavy design into......
Or trying to imply that some DX12 features somehow equal a mythical performance boost.
You will get performance improvement but it is up in the air if you will get every single feature that DX12 will include.
themadcowtipper
Smells faintly of rancid stilton.
Have fun waiting if you are trying to be on the cutting edge of pc tech. A gimped 970 runs pretty much every game better then current home consoles.
Eclectic Jabber Monkey
Member
I've experienced this type of marketing double speak by Nvidia in the past and been burned by it, so until they say flat out "We support the new hardware features in DX12" I wouldn't believe them, or other hardware maker for that matter.
I was looking into building a new PC in 2007 and Nvidia told enthusiast websites that the 680 series would support the new upcoming Intel manufacturing process, 45nm I believe. I ended up buying a 680 series mobo and dual core 65nm chip, expecting to upgrade to a 45nm quad core a few years down the road. Low and behold, when the 45nm chips come out only Core 2 Duo dual core processors are supported, which they unsurprisingly failed to mention to enthusiast websites. That chipset also had a BIOS update that would stealth down-clock one of the buses as a result of Nvidia losing a patent lawsuit.
Wasn't sad to see them get out of the chipset business.
dontbelievetheirlies.jpg
I thought it was well known that the gpu in xb1 was GCN 1.1 or maybe 1.0. The ps4 definitely seems 1.1. I've seen no mention of the XB1 supporting things like the volatile bit.
dr_rus
Member
Unlikely as 970/980 doesn't support even DX11.2 and that doesn't stop anyone from buying them.
Dictator93
Member
We got here because I mentioned the similarity between the idea that "the xb1 has hidden hardware GPU features unlocked by dx12" reminds me of "dgpu in the powerbrick" and other crazy stuff that I have heard regarding the xb1.
You argue that they were deving dx12 whilst deving the xb1 and therefore xb1 probably has GPU hardware features in dx12. They will then surprise devs, the public, and everyone by unveiling and unlocking this hardware through a dx12 update.
To me, this sounds like a DGPU being unlocked. It really does.
I am pretty sure Bonaire is GCN 1.1.
I would imagine the new Maxwell cards would allow one to get the full benefits of DX12. I mean, NVIDIA and Microsoft talk to one another, no? I have a hard time believing that the engineers at Microsoft that have been working on DX12 for the past ~2 years never once communicated their intentions or shared their software with NVIDIA.
If not...then would we have to wait for Maxwell's successor to see cards that take full advantage of DX12, or could the eventual 14nm FinFET Maxwell cards accomplish this?
DX12 is not finalized and requires input from all hardware vendors. However Hardware requires far more dev time and validation than software. This means things have to be finalized far earlier than software, so it is highly possible it will not support all dx12 feature levels (which is why they exist).
.. really?
It supports full dx 11.3 and so on. (It is Nvidias Tock in Intels roadmap.)
Only question is how low level DX12 will go, will it support everything each new GPU will allow or is there similar feature levels to DX11 except for each vendors.
Certainly would love to see things like interpolator access in pixel shader to be available in most/all cards.
Hopefully this:
Fullly supports all DX12 / Direct3D12 features.
As well as AMD's equivalent (R9-3XX).
GCN 1.1 (Ish?), The problem is that AMD never defines what they consider the various GCN versions. The number are attributed by the press in order to sort of describe the hardware iteration.
For example: Bonaire is 1.1, but the 'volatile bit' feature was put in the ps4 at Sony's request (it is implied that they even drew up the specification). The R290X, another gpu considered to be GCN 1.1 is known to contain this feature.
So the question remains, what features are necessary for GCN versions to be defined. The XB1 does not contain Trueaudio which is another feature often considered to be part of 1.1.
On the other hand Bonaire was released in March 2013 and it could fit so presumably was developed in parallel to the ps4 gpu (despite having seperate teams).
It should, like all current Maxwell2 GPUs.Hopefully this:
Fullly supports all DX12 / Direct3D12 features.
As well as AMD's equivalent (R9-3XX).
Nvidia hasn't really added features between really big upgrades and the cycle is years.. (It seems like Fermi was their previous big upgrade in terms of features.)
So this Techreport article from March last year was apparently right all along.
There was never any dispute of this.
We can deduce there are at least 3 levels of DX 12 support.
- Level 1: GCN 1.X, Fermi, Kepler and 1st gen Maxwell. (feature level 11.0-1-2?)
- Level 2: 2nd gen Maxwell (and GCN 2.0)? (feature level 11.3?)
- Level 3: Pascal and GCN 3.0? (feature level 12.0?)
I suppose level 3 will be full support and level 2 is Maxwell partial support.
- Level 1: GCN 1.X, Fermi, Kepler and 1st gen Maxwell. (feature level 11.0-1-2?)
- Level 2: 2nd gen Maxwell (and GCN 2.0)? (feature level 11.3?)
- Level 3: Pascal and GCN 3.0? (feature level 12.0?)
I suppose level 3 will be full support and level 2 is Maxwell partial support.
Dictator93
Member
First gen maxwell has at least 4 DX12 features (the OIT stuff, the conservative rasterization, etc...)
dr_rus
Member
Really.
11.3 is DX12 features on DX11 "thick" API. Until now there is no indication that support for 11.3 will mean support for 11.2. Some features in 11.2 are optional.
Also - if it supports "full dx 11.3" then it supports full DX12 as well since these are same things feature wise.
Of course there will be feature levels, how else would all NV's DX11+ chips support D3D12?
Dictator93
Member
oh hah, forgot about that bugger.
Is MS going to define a Feature level for a single card. Correct me if I'm wrong but the first version of Maxwell only appeared in the 750ti. Unless of course DX12 starts doing it the OpenGL way.
Agreed.
DX11.3 and DX11.2 thing would be interesting indeed.Really.
11.3 is DX12 features on DX11 "thick" API. Until now there is no indication that support for 11.3 will mean support for 11.2. Some features in 11.2 are optional.
Also - if it supports "full dx 11.3" then it supports full DX12 as well since these are same things feature wise.
Also DX12 will most likely be expanded in similar fashion in future.
Of course there are feature levels, but will there be one for each vendor?
IE. Nvidia, AMD and Intel each having their own path.
dr_rus
Member
That will completely defeat the purpose of a common API and force developers to code for each GPU architecture separately.
A feature level is a level of support which isn't tied to any vendor and is actually open to be supported for any vendor at any time.
If I have to guess I'd say that we're looking at four feature levels in DX12: d3d11, d3d11_1, d3d11_2 and d3d11_3.
Fermi, Kepler and Maxwell1 are d3d11.
GCN1 is d3d11_1.
GCN11 and GCN12 are d3d11_2.
Maxwell2 is d3d11_3.
I think misterX said somewhere that 2016 will be the year when the hidden magic juices of the Xbox One that are steeping right now will finally be unleashed.
It's a not-so-clever little gotcha question, trying to make it seem like you're spreading misterX's BS around.
UnrealEck
Member
I can't think of any game that doesn't.
Why doesn't Xbox One have full support for these DX12 features now? Spencer says they're coming, but I'm wondering why they weren't already in. If they're coming because they weren't finished until recently, post Xbox One hardware finalisation, wouldn't that mean these additions are software based? If so, why wouldn't GPUs like Maxwell have support for the same features?
It would make it harder, yes.
DX11 already was similar as DX11.2 was only supported by AMD and Intel.
Developers had to write code for each vendor if they wanted to take advantage of fetures and/or avoid pitfalls.
DX12 is a 'low level' API and when going low enough there will be more differences between GPUs.
Agreed, this is likely.
Cannot wait to see what next GPUs from AMD and Intel have for us.
oh well, i am kind disapointed by all this, i was reading this article(http://blogs.nvidia.com/blog/2015/01/21/windows-10-nvidia-dx12/) last night, and came across this "We’re more than ready. GPUs built on our Maxwell GPU architecture – such as our recently released GeForce GTX 970 and GeForce GTX 980 – fully support DX12." only to find out now that its only partial support, i am sad now.
Dictator93
Member
It is market talk as I said before. "fully support" vs "support full"
It is still BS though obviously. I would expect GM200 or GM210 (whenever that exists) to be fullllllllll dx12 though.
