AMD Mantle API [updating]

If DX12 is supposed to be so "Mantle like", AMD should have DX12 working drivers.

 

The question is more, do they want to send the driver "today" to developpers. I dont know if they have conduct all the test they want for it.

DX12 features, specification, technical aspect are still "work in progress".

I believe for Nvidia ( and looking how much they have speak about Mantle during GDC ), sending "pre alpha drivers DX12" to some developpers was more like part of publicity ( i say developpers, i should say the lucky few ones who have got a pre alpha sdk from MS ( i dont even now if someone have it for be honest outside MS team today, at this minute ).

Nvidia have do a good work at marketing during GDC lol. Most jurnalist point in their articles that there's a lot of marketing behind this type of annonces and it is obvious.

 

Without the manpower needed for working Mantle drivers and titles today, there probably wouldn't even be a Mantle-like DX12.
AMD has over a year and a half to get DX12 drivers. Having them today is useless except for those going into defence for not having Mantle and needing to present some pretty charts to the audience (nVidia).

 

A year and a half before having DX12 drivers available on a development basis would be too late for applications being developed so that they can release soon after the API or new Windows.

 

Actually, Turn 10 / MS's presentation referenced the 4 man months for porting Forza from DX11.x to DX12; we estimate that porting Mantle to DX12 will take even less time, so that's not really the case. What's more important is understanding the benefits that these low-overhead API's can bring to your title and then tuning the game to make the most of that (i.e. at the very least significantly increasing the number of objects that you've been limited to previously) and that's where developing on Mantle right now will get you a leg up with an environment that in use by a number of developers and shipping in some instances; those principles will apply across consoles and DX12 when it becomes available.

 

Waiting 18 months for drivers gives developers approximately 0 man months, man seconds, or dayfly milliseconds to iterate over anything.
DX11 has been out for years, and there's still the age-old song and dance with custom work done pre-release to create beta drivers that support newly released games or applications, then hotfixes and incremental update rounds for weeks or months afterward.
BF4's Mantle rollout showed no change to this, which is something that I thought would have been curtailed since so much should have been on DICE.

Waiting 18 months to get something available for developers means they release a game without the necessary drivers, so something like the RAGE launch, only instead of just not delivering the driver ID worked on, AMD would be asking a developer to start final development work on the AMD driver path as it's being sold and played by everybody else.

The claims being contested involve AMD exceeding a very low bar with an alpha-quality driver capable of surviving a limited demonstration, which a tech media report asserts it has not managed yet.

 

Not sure why you are talking about waiting 18 months, nobody said that. But the simple fact is that Mantle is available now and having titles released on it and those principles that hold on Mantle right now will hold on an even wider set of platforms in the future.

 

I responding to a post saying drivers were not needed for a year and half.

 

Except it didn't say that.

 

I guess our interpretations differ, but it would be unproductive to hash out our reading of someone else's post.

 

Once middleware like say Unreal Engine ships with DirectX 12 support perhaps it will kick off significantly.

However, UE3 had DX10 support early on but it went unused for various reasons like XP and D3D9 GPU market share. I imagine that DX12 will be Windows 9-only and many people will have Windows 7/8 + D3D11 cards for years. I also wonder how long D3D9 will be supported by games. D3D9 might finally be losing its grip these days.

 

Link me the download!

 

Runtime is available in the latest Catalyst downloads!

 

The GPU market share difference should be the kicker for DX12. DX10 had very little early on whereas virtually all GPU's used for PC gaming will support DX12 by late 2015. Even if it is limited to Windows 9, developers should have a lot more motivation to support DX12 right out of the gate than they ever had with DX10. Expecially considering the commonalities with the XB1 API and the draw call requirements of current generation console games.

 

An in-depth review of Mantle in both Thief and BF4:
http://www.techspot.com/review/793-thief-battlefield-4-mantle-performance/page4.html

Huge gains for AMD's FX 8350 (especially in Thief), not so much for Intel.

 

Do they compare CPU use on those non-CPU limited setups, though?

 

they do a bit in battlefield and it looks like if your gpu limited in dx11 you'll gain 5%(ish)

 

Thief numbers are severely impacted because of the activation of SSAA, which is a part of the Very High preset used by the game. It's an AMD trend, whereas other games decouple SSAA from the general quality presets, AMD often tries to impose it in, especially if the game is light on graphics and not performance intensive. This happened in Sleeping Dogs and now Thief. Tomb Raider had SSAA too but it was decoupled, probably because TressFX was taxing enough at the Ultimate quality preset.

Anyways, SSAA does benefit the game visually, and the game runs at very good performance levels with it. Never mind the official benchmark as it is just a worst case scenario, 90% of the game is spent on closed spaces where the frames soar. Mantle's benefit should be more pronounced when testing with FXAA or with lower settings (as done by other sites), which is a moot point probably considering high-end GPUs are aimed for max quality.

 

There's no comparison to be made efficiency-wise between Mantle and DX. One blows the other out of the water. The difference is not small, and it is not 5%, at least in multiplayer.

http://www.pcgameshardware.de/Battl...tlefield-4-Second-Assault-Benchmarks-1109970/

They did a follow-up to see what it took to get Nvidia cards to par with AMD cards (Windows 8.1 and a 4770k @ 4.6Ghz):

http://www.pcgameshardware.de/Battl...ield-4-Second-Assault-Geforce-Radeon-1110584/

As you can see, they're still behind, and still probably CPU limited-- you can see no change between the 670 and the 770, or between the 560Ti and the 570. The differences between these two pairs of GPUs are so great that I doubt the unpredictable nature of multiplayer testing will cover it up, and they probably ran through several times on each GPU anyway. On the AMD side, there is a clear progression in performance between the slowest and fastest GPU tested.

Even if their average matches up though, averages can hide really poor behavior. There's probably a point where you get no returns and you're purely "GPU limited" but if it doesn't happen with Haswell @ 4.6GHz then it is a point that is mostly academic. And I'll assert, though I'm too lazy (and poor) to try and prove, that Mantle will have a much tighter spread of frame rate no matter what CPU you use.

My experience with Mantle is largely positive and reflects these results, on a 4770K both at stock and overclocked to 4.1 Ghz and a stock R9 290.

...5% . Even if there were no graphs that prove otherwise, I would still laugh at that number. I could feel the difference straight away. And my 4770K... well, it wasn't cheap.

Sorry, I had to make this account because I was so shocked at those results.

 

interesting article: http://www.anandtech.com/show/7889/...w-level-graphics-programming-comes-to-directx