DirectX 12 destroys competing API’s with up to 1600% improvement: AMD/Nvidia.

[Boomstick777]

Recently the folks over at Eurogamer did a few benchmarks using AMD and Nvidia hardware on the latest build of the DirectX 12 API on Windows 10. Microsoft recently released the free upgrade to millions with the aim of getting 1 Billion devices upgraded to Windows 10 within 3 years. They also announced that they managed to rake in 14 Million upgrades within 24 hours, which is just under 160 upgrades per second.

http://www.ticgn.com/directx-12-destroys-competing-apis-with-up-to-1600-improvement/

 

[fs123]

So basically they just tested with the 3dMark API overhead test.
For a second I thought they had access to some new DX12 demo.

 

[Mauller]

The Directx 12 benchmark for 'Ashes of the singularity' is out on the 13th. so people will have to be patient till then to see the difference.

 

[drunkenmaster]

What people will probably entirely overlook is minimums when it comes to Dx12 games. Benchmarks have and will continue to do absolutely nothing for me, including those game benchmarks which don't actually reflect real in game performance at all but just use the same engine with higher/lower settings than the real game.

With Mantle in almost all the games the minimums was the biggest difference and caused the biggest improvement in gameplay and smoothness. Without remembering the exact numbers you would see in Thief maybe a 10% raise in average fps but minimums going from 30 to 50fps. You might only have 1% of all frames being that low and as such the average doesn't go up drastically but it's the drops from say 70fps to 30fps that you notice. The smaller the change in frame rate the less obvious the drop and the better the perceived smoothness.

People may(or may not) be disappointed in max frame rate improvement but overlook minimums improvement and general improvement in feel of games.

This is talking about games which basically port or add in DX12 purely for a speed boost. Other games will enable many more effects using way more draw calls under DX12 and then you just have to see how well performance copes with higher IQ. Dx12 will be capable of a hell of a lot more in higher settings and more effects usage. I'd expect the former type of game to appear basically straight away and most games maybe xmas onwards to be DX12 supporting. Games that really go big for DX12 and can effectively not run under DX11 will be rare for a while but should still appear much faster and in much greater quantity than any other new graphics API is launched.

 

[Mauller]

In terms of stuff running on DX that just can't on DX11. Ashes already does that. just look at their lighting engine. they have thousands of point lights and real light sources in game with no detriment.

We could see real global illumination sooner than later with DX12. bye bye to raster shadows/lighting and HBAO finally.

 

[mmj_uk]

We heard all of this before when AMD announced Mantle and look at how that turned out, in actual gaming users will probably get a small boost on high end hardware and it'll take years before DX12 starts pushing beyond what DX11 can do.

Most games today are just console ports and they have tragically slow CPU's so I wouldn't expect games to suddenly have 10 times more draw calls, the best we can probably hope for over the next year or two is the odd PC exclusive where the developer (or NVidia with GameWorks) took the time to add an enhanced DX12 mode.

 

[r7slayer]

Wasn't there a certain game dev which ported a certain racing game and from DX11 to DX12 they saw about 40% extra performance on the same hardware??? Sounds impressive to me! Sounds like DX12 is pushing well beyond DX11 as it is even with draw call performance and reducing CPU overheads. Not sure why people are hating...

Also mmj your post looks a lot like its making out mantle to be similar to DX12 where a lot of nvidia users tell us differently. lol make your minds up.

 

[Kaapstad]

So do games run any faster

Probably not

It is a bit like running minesweeper on quad TXs, I get 30 fps the same as everyone else.

 

[r7slayer]

So do games run any faster

Probably not

It is a bit like running minesweeper on quad TXs, I get 30 fps the same as everyone else.

Wut?

 

[Boomstick777]

We heard all of this before when AMD announced Mantle and look at how that turned out, in actual gaming users will probably get a small boost on high end hardware and it'll take years before DX12 starts pushing beyond what DX11 can do.

Most games today are just console ports and they have tragically slow CPU's so I wouldn't expect games to suddenly have 10 times more draw calls, the best we can probably hope for over the next year or two is the odd PC exclusive where the developer (or NVidia with GameWorks) took the time to add an enhanced DX12 mode.

Agreed, I would be happy even if it was just a couple games that really took advantage of PC hardware. Might get me back into gaming on PC.

 

[D.P.]

Competing APIs, what, like DX11??


Anyway, next person who posts a draw-call benchmark as if it is an actual test of graphics should get a 1 week suspension. 3DMark warns against such things.



We all know you can send far more empty draw-calls to the GPU with DX12, that has a pretty minimal bearing on performance in itself. What really counts is how long it take to render a draw-call that actually has real content to render.

 

[Silent_Scone]

So do games run any faster

Probably not

It is a bit like running minesweeper on quad TXs, I get 30 fps the same as everyone else.

Coming from someone who actually owns quad TX's this is both defeatist and rather stupid lol.

You should look forward to what's to come. I agree technology demos don't serve a real purpose when it comes to performance, though

 

[Mauller]

We all know you can send far more empty draw-calls to the GPU with DX12, that has a pretty minimal bearing on performance in itself. What really counts is how long it take to render a draw-call that actually has real content to render.

The 3dmark api draw calls are not empty, they are rendering simple textured cubes in the test, without any visual culling. But i know what you are trying to say.

Which as i mentioned. we should get the first DX12 benchmark on the 13th. it should also have asynchronus shaders as a feature.

http://forums.ashesofthesingularity.com/469403

 

[gpuerrilla]

Sounds impressive to me! Sounds like DX12 is pushing well beyond DX11 as it is even with draw call performance and reducing CPU overheads. Not sure why people are hating...

Also mmj your post looks a lot like its making out mantle to be similar to DX12 where a lot of nvidia users tell us differently. lol make your minds up.

I think we have all observed the shift in green team noise as mantle is conveniently in the past (to them). I guess there are people no matter how good it is for all PC gamers still brandishing their loyalty without realising.

 

[D.P.]

The 3dmark api draw calls are not empty, they are rendering simple textured cubes in the test, without any visual culling. But i know what you are trying to say.

Which as i mentioned. we should get the first DX12 benchmark on the 13th. it should also have asynchronus shaders as a feature.

http://forums.ashesofthesingularity.com/469403

I know, my point was there is no advanced pixel shaders, tessellation, vertex or geometry shaders going on. So the number of drawcalls is fairly irrelevant.

We know DX12 has a far lower over head. We know what that does thanks to mantle, when the driver is optimized there is a small boost and some CPU thrashing is resolved, when there is no driver optimizations it performs 20% worse than DX11 in BF4 on Fiji. Driver optimizations will still dominate DX12 performance.

 

[melmac]

People are too focused on the highest FPS. Where Dx12 will shine is in raising minimums, reducing massive frame rate drops and speeding up frametimes. We already know from Mantle roughly what the performance of Dx12 will be.

And lastly, if Dx12 really is a low level api, then it will be up to the game developer to get the best out of the hardware. You shouldn't be seeing driver level improvements as much any more.

 

[Mauller]

I know, my point was there is no advanced pixel shaders, tessellation, vertex or geometry shaders going on. So the number of drawcalls is fairly irrelevant.

We know DX12 has a far lower over head. We know what that does thanks to mantle, when the driver is optimized there is a small boost and some CPU thrashing is resolved, when there is no driver optimizations it performs 20% worse than DX11 in BF4 on Fiji. Driver optimizations will still dominate DX12 performance.

I know what you meant, is why i mentioned it.

There will still be optimisations in the Api translation layers etc. but we wont need to see constant driver updates for every game that drops, unless a major bug is picked up etc. or some further optimisation has occurred.


BF4 is a good and bad example overall, it has always had problems. but the issues with low abstraction api's can hopefully be resolved when using WDDM2.0.

Reading some of Johan Andersons tweets kind of pointed the problem at the way WDDM1.1-1.3 got in the way of memory management. But that issue is cleared up with WDDM2.0 in win 10. He even mentioned that mantle would perform far better if coded with win10 and WDDM2.0 as the baseline.

https://twitter.com/repi/status/585556554667163648?lang=en

So in regards to the above, with DX12 and WDDM2.0 we should hopefully not see the issues that BF4 had with mantle.

I am thinking the performance issue with the initial fiji boards (since the 285 only had 2GB of ram) and the fury go back to BF4's memory management problems. Since thief and CiV:BE have no problems in mantle with the fury, since it has enough ram compared to the 285.

 

[Kaapstad]

Coming from someone who actually owns quad TX's this is both defeatist and rather stupid lol.

You should look forward to what's to come. I agree technology demos don't serve a real purpose when it comes to performance, though

Where I am a very old fart I have seen these claims so many times for so many things and been disappointed so many times.

What was the last one I saw for a particular GPU, an overclockers dream !!!!

 

[D.P.]

I know what you meant, is why i mentioned it.

There will still be optimisations in the Api translation layers etc. but we wont need to see constant driver updates for every game that drops, unless a major bug is picked up etc. or some further optimisation has occurred.


BF4 is a good and bad example overall, it has always had problems. but the issues with low abstraction api's can hopefully be resolved when using WDDM2.0.

Reading some of Johan Andersons tweets kind of pointed the problem at the way WDDM1.1-1.3 got in the way of memory management. But that issue is cleared up with WDDM2.0 in win 10. He even mentioned that mantle would perform far better if coded with win10 and WDDM2.0 as the baseline.

https://twitter.com/repi/status/585556554667163648?lang=en

So in regards to the above, with DX12 and WDDM2.0 we should hopefully not see the issues that BF4 had with mantle.

I am thinking the performance issue with the initial fiji boards (since the 285 only had 2GB of ram) and the fury go back to BF4's memory management problems. Since thief and CiV:BE have no problems in mantle with the fury, since it has enough ram compared to the 285.

While that is true, it was a very difficult for developers to optimize through the stack of API abstraction, windows management, driver voodoo etc., DX12 won't stop the need for driver optimizations. One of the major issues is that there are different GPU vendors with different architectures and even the different vendors have different chips. A game developer cannot optimize for all the different GPUs. Moreover, even if they wanted to optimize things the developers don't know details of the hardware. and then there is the issue of new hardware released after the game launch, e.g. a recently released game wouldn't have had optimization support for Fury cards.

So AMD/nvidia will still have to out a lot of effort into optimizing games, but game developers can do some optimizations themselves with more predictable results without driver voodoo and API abstraction interfering to such a degree. Of course the matter takes more work by developers, so again we will see smaller indie devs use middleware platforms like gamesworks to reduce development costs.

 

[vintage-x]

What people will probably entirely overlook is minimums when it comes to Dx12 games. Benchmarks have and will continue to do absolutely nothing for me, including those game benchmarks which don't actually reflect real in game performance at all but just use the same engine with higher/lower settings than the real game.

With Mantle in almost all the games the minimums was the biggest difference and caused the biggest improvement in gameplay and smoothness. Without remembering the exact numbers you would see in Thief maybe a 10% raise in average fps but minimums going from 30 to 50fps. You might only have 1% of all frames being that low and as such the average doesn't go up drastically but it's the drops from say 70fps to 30fps that you notice. The smaller the change in frame rate the less obvious the drop and the better the perceived smoothness.

People may(or may not) be disappointed in max frame rate improvement but overlook minimums improvement and general improvement in feel of games.

This is talking about games which basically port or add in DX12 purely for a speed boost. Other games will enable many more effects using way more draw calls under DX12 and then you just have to see how well performance copes with higher IQ. Dx12 will be capable of a hell of a lot more in higher settings and more effects usage. I'd expect the former type of game to appear basically straight away and most games maybe xmas onwards to be DX12 supporting. Games that really go big for DX12 and can effectively not run under DX11 will be rare for a while but should still appear much faster and in much greater quantity than any other new graphics API is launched.

That to me is what excites me most.
I have never really cared that I can get 100000fps stood alone in a dark room with no effects going on in the background but I do care when I'm in the middle of an intense fire fight and my fps drops from 60 to 30 causing me to die because my controller response doubles and makes it harder to shoot where I want