Chiggs
Gold Member
A minimum increase from 19 to 31 is pretty hot
Indeed. That's pretty fantastic.
A minimum increase from 19 to 31 is pretty hot
10 fps less with fury x in 1440p
With the latest AMD driver (16.3) installed, I'm getting slight improvements on an R9 Fury (1080p):
DX11 Overall: 57.91
Mountain Peak: 60.00
Syria: 58.48
Geothermal Valley: 55.34
DX12 Overall: 59.27
Mountain Peak: 58.09
Syria: 59.87
Geothermal Valley: 59.95
The Geothermal Valley definitely felt quite a bit smoother on DX12 (at least in the benchmark, haven't tried it in gameplay yet).
With the previous driver, I did get about 8FPS less on average with DX12, though, so I guess those who're getting worse performance on AMD GPUs with DX12 are still on the old driver.
Your DX11 bench shows a minimum of 9/10 FPS while DX12 shows a minimum of 36/24 FPS. That's an improvement.
Yes there might be some headroom for the upper limits of the framerate but damn again, that increase in min FPS is insane and should remove a lot of visible stutter. Still shows a lot of promise
Typical Nvidia Tech. Twice the price, same product.
It wasn't functional, they just released another patch to fix it.
Something is really off about these benchmarks. How do you get such huge minimum frame rate improvements and yet have drops in the average frame rate. If you have such a large improvements in frame times you would think the average and maximums would rise as well.
I'll take a stab at it and suggest that the DX12 path in the game isn't very optimized vs DX11 driver optimizations. However, the DX12 the baseline frame rate when there are a lot more draw calls or when the CPU is more loaded is a lot higher than passing all that stuff through a thick dx11 driver layer.
But since the benchmark doesn't provide a lot of information all that is speculation.
guys you should temper your expectations wrt dx12 performance improvements when gpu limited on nvidia cards. other than algorithms that make direct use of the 11.3/12_1 hardware features, nvidias architecture currently cant benefit from multi engine rendering on the gpu side so general performance will at best be equal to dx11.
Ok, but AMD performance is also worse.
but we have instances that show amd can benefit from dx12 on the gpu side
Only on games specifically designed to target AMD's advantage in asynchronous compute/shaders
multi engine gpu rendering was kind of the only fundamental change made on the gpu side of dx12.
This has nothing to do with what I said
it just seems odd to write off the fundamental improvement dx12 brings to the gpu side of rendering as an amd advantage that games must specifically target
it just seems odd to write off the fundamental improvement dx12 brings to the gpu side of rendering as an amd advantage that games must specifically target
Well are the games which are currently benchmarking better on AMD specifically tethering heavy use of asynchronous compute? Yes or no?
You're acting like game engines are all going to use this a lot now just because it's available. And that seems unlikely because most games will still target DX11 for compatibility reasons.
Also you're acting like Nvidia won't ever implement a hardware scheduler, what will happen if Pascal has one? I can't wait to see how the AMD fanboys behave when this one vaunted feature of DX12 becomes available on both sides.
lol. not an amd fanboy. good try tho. id imagine most if not all of the high profile games that support dx12 will absolutely use multi engine gpu rendering. its already starting to gain heavy traction in the console space as its the best way to keep gpu utilization as high as possible at all times. i dont see why they wouldnt use it. future nvidia gpu architectures that may possibly support multi engine gpu rendering have no relevance today. i hope they do, but i have my doubts that it will come with pascal.
and why wouldnt dx12 games make heavy use of async compute? im trying hard to think of reasons why it would be in a developers interest not to use something they probably already are on the consoles but im coming up blank.
It might be because I dunno a console is a fixed hardware and PCs aren't? I'm not a rocket scientist but even I know that the total number of gaming PCs out there with the hardware which greatly benefits from async compute, specifically a very weak CPU with a strong GPU which is made by AMD, is a small percentage of the market. Async compute is amazing for helping out the netbook-class CPUs in the PS4 and Bone. It's not quite as useful when your gaming PC has an i7. Even the vaunted Ashes of the Singularity, the poster child for async compute, has Nvidia DX11 performing similarly to AMD DX12.
Async compute is amazing for helping out the netbook-class CPUs in the PS4 and Bone.
damn again, that increase in min FPS is insane and should remove a lot of visible stutter.
the game runs a few frames lower and its stuttering all the time now with DX12 on.
How can they "rise up to the challenge" when right now the big problem is that PC ports are RUSHED? If DX12 requires MORE time and expertize how can we expect it an improvement if *time* was the problematic variable in the first place?is one major hurdle for new developers, particularly those who don’t have a firm grasp on the hardware. The low-level nature of DX12 means that more control over optimizations will be in the hands of developers – and they will need to rise up to the challenge for best results – as opposed to video card drivers.
As Baker notes, since the PC is such a large and varied platform “You can never perfectly optimize for every platform because it's too much work” as compared to the highly regulated consoles, so instead the name of the game is making generic optimizations and try to be as even-handed as possible. At the same time the company has also been atypically transparent with its code, sharing it with all of the GPU vendors so that they can see what’s going on under the hood and give feedback as necessary.
https://twitter.com/FioraAeterna/status/703294520369160196true fact: hardware devs are sometimes nervous about helping you optimize arch N, because it means their arch N+1 looks worse by comparison
https://twitter.com/rygorous/status/653041674218672128considering how much some gamers freaked out at the "3.5GB" thing, revealing how broken GPUs are might make their heads actually explode
Everybody wants the level of control but nobody wants the responsibility that goes along with it.
if DX12/vulkan are magic bullets, their ability is to auto-aim at the programmer's foot
blah blah blah blah
If anything, expect your framerates to drop even lower.
Async Compute has nothing to do with the CPU. A fairly apt analogy is that it is SMT for the GPU.
Of course it has to do with the CPU. Where do you think many of those compute tasks were being done before?
Of course it has to do with the CPU. Where do you think many of those compute tasks were being done before?
On the GPU. Just not concurrently.
remember that DirectX 12 is about making the API more efficient so it can take better advantage of multi-core CPUs. It's not really about graphics cards. It's about exploiting more performance from CPUs so they don't bottleneck the GPU.
Of course it has to do with the CPU. Where do you think many of those compute tasks were being done before?
On the GPU. Just not concurrently.
It does make sense to some extent. Assume that VXAO is a rather fixed load regardless of the scene complexity. In that case, in complex scenes, the relative performance impact will be significantly lower than in simple scenes.Oh yeah, it makes a pretty big difference actually. So much of a difference, in fact, that I've restarted the campaign, and will play through the entire thing with VXAO now that's available. The interesting thing about performance I've noticed so far is that early areas are now considerably more demanding with VXAO, but the previously most demanding section the game, Geothermal Valley, is only slightly more demanding with the new AO. Does that even make any sense at all? I would've expected the GPU load with VXAO to go up proportionally in all areas.
So can I have a quick rundown of how VXAO works? Is it still based on a depth buffer or does it actually know where objects are and how close they are to other things?
This may be of interest:
http://on-demand.gputechconf.com/gtc/2015/presentation/S5670-Alexey-Panteleev.pdf
But as a quick answer to your question: No, it doesn't use the depth buffer and it is operating in world space as opposed to screen space AO like SSAO and HBAO.
It does make sense to some extent. Assume that VXAO is a rather fixed load regardless of the scene complexity. In that case, in complex scenes, the relative performance impact will be significantly lower than in simple scenes.