This is what I saw. Frametimes were 40% less than DX11 and min fps was 50-100% higher. The patch after the Mantle patch fixed the hitching. It's insane how smooth it is compared to DX11.
Mantle was amazing for BF4 in my usage.
Low settings, maximum FPS. I received like a 40-60fps boost to my minimums if I remember correctly. I would go from dips into the 50's and 60's in dx11 to constantly 120 fps on demanding maps. With an ancient 2500k and a r9 290. I posted pictures back in the mantle thread when it came out. I've always been chasing high fps for competitive advantages. This is where Mantle shines.
Unfortunately, the random hitches in Mantle became far too annoying for me to deal with.
This is what I saw. Frametimes were 40% less than DX11 and min fps was 50-100% higher. The patch after the Mantle patch fixed the hitching. It's insane how smooth it is compared to DX11.
I recently picked up Dragon Age Inquisition and have been running it on the system in my sig (only the Q9550 is at stock settings thanks to being stuck on a g41 back-up board that I never got around to replacing). With the DX 11 renderer the game is perfectly playable at high settings but a little choppy in heavily populated areas.
I decided to give the mantle renderer a shot, and WOW, the game runs smooth as butter everywhere at high settings. Its like night and day. DA:I was the first game I've played that made me think it was time for that core upgrade I've been putting off (Married with children, don't game as much as I used to). Mantle may have just saved me a couple hundred bucks.
Reviews never really did mantle much justice as they're usually run on beefcake processors that don't bottleneck a card like my Q9550 does my 7950. I'm a true believer now, I've dipped into the punchbowl and drank the coolaide.
If this is the future Vulcan and DX12 promise, sign me up. Thanks AMD!
Someone might have an older PC which can support a newer graphics card, but has an older CPU. Mantle is a boon to them, reducing the need to spend hundreds of dollars on things like a new motherboard, CPU, CPU cooler, and RAM. A CPU upgrade can be much more extensive and expensive than a GPU upgrade.
As I can tell, the use case scenarios for Mantle (and also DirectX 12 and Vulkan) are these:
--Multiplayer games and strategy games with high CPU loads that can tax even mid and high end processors (Battlefield 4 is an example of this, and the Oxide game engine is targeting this).
--Getting the most out of multi-core processors thanks to superior multithreading. This especially helps AMD processors.
--Bringing down the CPU requirements in games in general to help older systems run games without requiring an upgrade. IE my brother's Q6600 system.
That's bullshit. GCN 1.2 supports Mantle. And no need to do more. Of course it don't able to use some specific features, but GCN 1.2 will be as good as GCN 1.1.The 285 and Fiji both are no longer getting updates for Mantle so some games may not work great with it. It'll work just fine on DX12 and Vulkan though once those hit.
That's bullshit. GCN 1.2 supports Mantle. And no need to do more. Of course it don't able to use some specific features, but GCN 1.2 will be as good as GCN 1.1.
With these explicit APIs the driver is not really more than a shader complier. This means all the things that was in the kernel driver now must be implemented in the application code. This will eliminate the kernel driver and all of it's server threads permanently, which helps to use the CPU more efficiently. As a drawback upgrading the driver won't do give any advantage, because the application has explicit control over the memory management, and the implemented optimizations might not works on the newer architectures as efficient as on the older ones. That's why Battlefield 4 doesn't works well on GCN 1.2, because it never get patched to support that architecture efficiently. But the never games are optimized for GCN 1.2 - out of box or with a patch.
This is exactly the same model what D3D12 and Vulkan will introduce, and it's really sad that nobody say anything about it's drawbacks. Don't get me wrong, I think these APIs will allow to create more optimized PC ports, but sure getting a new hardware that doesn't really supported on the actual games may result equal or worse performance that it's predecessor, even if the new hardware is faster in theory. The developers have to make a patch with some specific optimizations. There are several interesting strategies to make this less problematic, but on the PC it will need some bigger engine changes, so this is an ongoing research and it won't have results in the short term.
That's bullshit. GCN 1.2 supports Mantle. And no need to do more. Of course it don't able to use some specific features, but GCN 1.2 will be as good as GCN 1.1.
With these explicit APIs the driver is not really more than a shader complier. This means all the things that was in the kernel driver now must be implemented in the application code. This will eliminate the kernel driver and all of it's server threads permanently, which helps to use the CPU more efficiently. As a drawback upgrading the driver won't do give any advantage, because the application has explicit control over the memory management, and the implemented optimizations might not works on the newer architectures as efficient as on the older ones. That's why Battlefield 4 doesn't works well on GCN 1.2, because it never get patched to support that architecture efficiently. But the never games are optimized for GCN 1.2 - out of box or with a patch.
This is exactly the same model what D3D12 and Vulkan will introduce, and it's really sad that nobody say anything about it's drawbacks. Don't get me wrong, I think these APIs will allow to create more optimized PC ports, but sure getting a new hardware that doesn't really supported on the actual games may result equal or worse performance that it's predecessor, even if the new hardware is faster in theory. The developers have to make a patch with some specific optimizations. There are several interesting strategies to make this less problematic, but on the PC it will need some bigger engine changes, so this is an ongoing research and it won't have results in the short term.
Does this mean a developer can optimize for the different fermi iterations, keplers, maxwell 1, maxwell 2, gcn 1, 1.1 and 1.2 as well as the intel chips?
I guess most games will stick with dx11 a while longer then.
Kidding me, AMD's CPUs were getting creamed. Mantle and soon DX12 is gonna give that failed uarch a breath of life. Not sure if it will boost sales but it might at least plug some holes from people waiting to jump ship long enough for Zen to manifest.
It's actually the other way around. The Intel compiler used to develop the vast majority of Windows software sandbags AMD processors. As many of us have talked about it in the CPU forums -- an FX-8350 will rival (and actually outrun in some benchmarks) an i7 3770k under Linux.
http://www.phoronix.com/scan.php?page=article&item=amd_fx8350_visherabdver2&num=1
Windows 10 appears to be leveling out the software -- as Linux has always done.
Does this mean a developer can optimize for the different fermi iterations, keplers, maxwell 1, maxwell 2, gcn 1, 1.1 and 1.2 as well as the intel chips?
I guess most games will stick with dx11 a while longer then.
Not even remotely. Unless you call Ryan Smith vouching for AMD's own word
http://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x-review/12
@zlatan From a high level pov,
Designing a game engine with multiple ways of doing x
-like shadows, or hair simulation and rendering
Doing pre install benchmark of the possible routes
- is the compute shader faster or x other shader stage.
Choose a mix of the fastest\lowest latency routes.
-tressfx hair, pcss, smaa
Or
-hair works, chss, mfaa
Sorry for the probably naive presentation of such complex systems
He asked the wrong person, but he didn't specified who was it. Probably a marketing guy. In the 15.7 Catalyst driver Mantle is updated with some functions and optimization. In fact there will be more than half-a-dozen games that uses Mantle in some way in the future. Most VR games will use Mantle with the LiquidVR SDK.
For GCN 1.1 or 1.2 GPUs though? What Ryan has quoted is that GCN 1.2 isn't receiving future Mantle specific updates beyond what just dropped. My advice is to not count on there being any updates to fix broken things on the Mantle side for GCN 1.2 going forward from AMD. E.g. don't expect a patch from AMD to fix 285/380/Fiji inoperability in Civilization Beyond Earth. Maybe a patch from Firaxis but not AMD. That's where we're at according to the official record as of now. Unofficially I wouldn't be surprised if a few things get patched anyways. So my original post is definitively not "bullshit".
According to some murmurs I've heard, Intel's compiler still spits out the most performant binaries even without proper optimizations. ICC outperforms gcc and llvm.
Exactly. And it's intels compiler. Why should they optimize for AMD? AMD could build their own compiler optimizing for their CPUs.