Graphics AMD’s DirectX 12 Advantage Explained – GCN Architecture More Friendly To Parallelism

http://www.dsogaming.com/news/amds-...re-more-friendly-to-parallelism-than-maxwell/

 

Interesting. I did find this article difficult to stomach as a 980Ti user.

I have faith that Nvidia will work it out somehow, at least for Maxwell cards. They must have known about DX12 during the Maxwell design process!

 

First DX12 game is still a decent bit of time away. We shall see what the data looks like when we have an actual release game to test with.

Ashes and 3d mark are at best test software. Ashes will eventually launch but still lacks a launch date at last check.

By which stage you would assume Nvidia will work out there issues in DX12 games.

 

Its a hardware issue though isn't it. Its about how good GCN is at parallel loads. Its not something that can be 'worked out' without a significant change in architecture design. Maybe Pascal will bring that, maybe not.

 

Everything I've read about Pascal seems to indicate that it will be changing to processing in parallel rather than Maxwell's serial nature, that's unless I'm confusing the ability of NVLink to work in parallel with other GPUs and that's not going to carry over to CPUs.

 

I'm going to wait and see some more DX12 workloads. DX12 very much relies on the engine developer to perform architecture-specific optimisation (rather than the driver developer as with DX11). Stardock have proclaimed "both GPU developers have access to the source code!", but for DX12 this is practically meaningless.
AotS started life as a Mantle demo, so the chances are high that it was optimised from the start to take full advantage of GCN. As with Hairworks and it's technically-to-the-spec use of tessellation (just in vast quantities) working well for Maxwell but pounding other architectures, it's not outside the realms of possibility that there is some architecture-targeting going on here. With Nvidia taking a hit from DX11 to DX12, that seems even more likely.

 

The strapline in the article says:

Yeah, the 290x achieves almost the same frame rate at 1080p as the 980Ti manages at 2160p.

I'm all for AMD clawing back some ground in the performance stakes, but... please!

 

The problem, as I see it, is that AMD have been pushing those features since GCN hit the market and there just wasn't the software to take advantage of it, that's why they were forced to develop Mantle.

If DX12 had been released at the same time as AMD released their first CGN card we would probably have a very different situation than what we currently have, as it is by the time we start seeing more than half a dozen DX12 games Nvidia will probably have Pascal on the market making the fact that GCN handles parallel workloads better a fairly mute point.

Personally I'm holding out hope that Kronos can build on the Mantle code base to offer a viable alternative to DX12, it is (imho) a better API even just for the fact that it's cross platform.

 

Not my kind of gaming, but it appears that the first DX12 game is tomorrow - http://steamcommunity.com/app/346110/discussions/0/594820656447032287/

 

Wonder if we will get some benches.

 

Wow! Amazing news for Ark fans.

 

Not entirely clear if this is just using the DX12 path (reducing driver load) or actual DX12 API calls. I was under the impression the firmer was automatic, but maybe not.

 

I would say, ignore benchmarks: Nvidia will drop game optimized drivers in time. They do this very consistently and very well.

Any DX12 game will have a fallback to DX11. There will be no single renderpath only.

 

The onus of optimisation for DX12 is NOT the driver developer, but the engine developer.

 

really? when did you start working for NVidia!

 

Whatever it is your are trying to say here, that is a childish way to express yourself.

Its all a bit up in the air at the moment, so there's very little reason for what he's saying to be impossible.

 

http://wccftech.com/oxide-games-dev-replies-ashes-singularity-controversy/

reply from oxide - they do make a huge statement , one that NVidia is VERY quiet about;

when using Async , as exposed in the driver on Maxwell - the card is slower as if its actually doing it in software and not native hardware.

basically Maxwell is not a full tier 3 DX12 card

 

Relevant quote:

 

with AMD hardware in the consoles and the money in consoles far far more han pc`s - they will code for what gives the best performance for the masses - look at the debacle of ubisoft and you can see the idea.

we will see far more Async useage if it means you get prettier games at a constant frame rate @1080p on xbox one / PS4