Has DX12/Vulkan Performance Improved

I wanted to buy a GTX 1070 or maybe a 1080 but I'm a little concerned by the DX12 and Vulkan benchmarks I have found. Granted a lot of these are from summer time the reason I'm posting now is to ask has Nvidia been able to improve things with driver updates since then? Also does anyone know roughly when Vega is supposed to be released?

 

If you are comparing to amd there is your problem, DX11 was highly optimized and alot its overhead was eliminated from Nvidia drivers years ago compared to AMD DX11 which has alot overhead still in comparison.

And OpenGL is utter mess and missing alot optimizing in there driver unlike Nvidia, Vulkan is just Opengl next gen name, Seeing AMD DX11 had alot overhead and OPENGL on AMD was badly done and supported. AMD saw much better gains from both cause how much overhead AMD has in DX11 compared to nvidia and how badly there OpenGL support in there drivers is compared to nvidia.

IF you talking about how some games have horrible performance in DX12 and other do not it comes down to how well the dev optimized the engine for them, Cause both require dev to actual do extra work that DX11 and OpenGL did for them, if one thing has been learned dev dont really care to optimize thing. they just do enough to make passable.

If you over think it like me you will never get new card.

 

DX12 means ****.

As I predicted.

To find a better card - find the formula - you will see the improvement/price increases over the years and be able to predict, basically everything , specially with regards ti GPU's - probably the most highly overpriced product

 

If you are aiming at dx12 and vulcan especially AMD are much more optimized so far. Even the older generation cards. Just wait a little longer and see what January will offer from vega and 1080ti. Even if you don't by neither of these cards previous models will lose some value for sure

 

they not better optimized, there DX11 (Nvidia went threw the trouble of reduce overhead on this amd has not as far as I am aware) and OpenGL ( was out right missing lot feature and code that nvidia support) support were terrible compared to nvidia in optimized department and there. DX12 on AMD VS Nvidia for most part there card performance just as well when match up against similar powered cards.

Poor performance comes from the dev not wanting to do the extra work NEEDED for both DX12 or Vulcan. for both Nvidia/AMD also DX11/OpenGL did things Dev now need to do them selves + optimizing for Nvidia/Amd architecture they they need to do.

 

It's not as if DX12 and Vulcan performance is so bad it renders the card useless.
It simply does nothing for you, slightly improves FPS, or slightly hurts FPS.
Regardless, the 1070 and 1080 are still really fast great GPU's.

It might be a case of dev's learning the API, games like Deus Ex MD offering no improvement, but games like The Division offering a 10 FPS pick-up even while using an expensive CPU. And both of these games have DX12 tacked on with a patch, pretty much every implementation of low-level API's have been an afterthought.

However the cards lay, the GPU you buy will still be fast. At the very least, the popularization of Vulkan and DX12 will help to render expensive CPU's for gaming a thing of the past.

 

As per above, it's difficult to work out if it has.

DX12 works better in AMD because DX11 is very inconsistent on AMD hardware, and while Vulkan is even further behind DX12, because OpenGL is poor for them Vulkan looked great with Doom.

With Nvidia it's needed less since DX11/OpenGL are already running well.

Gears of War 4 is easily the most optimised DX12 game, probably the best use of any low level api.
The problem is that there is no DX11 version to compare it to, for all we know it could just be a very well optimised game.

 

On my 980 DX12 is really not to be recommended. I lose 25% performance in Deus Ex for example.

 

^

Which is a great case of Dev need to learn the API And Devs are well known for not wanting to put the effort into optimizing on PC before DX12 and Vulcan were those API did things the Devs need to program DX12/Vulcan to do them shelves and then optimize it on top of optimize the engine for what they need it to do and the 1000's of configurations there are, and games the have and API patched into game tend to suffer.

There is reason everything is Developed for Consoles then ported to PC now.

 

Either way isn't that the final result ?

 

There is no way of knowing. The common thought is no, it will improve or at least level out as games and game engines are made with DX12/Vulkan at the forefront. The fact that the results are sporadic should be an indication of that.

 

Since nvidia updated their Vulkan runtime to 1.0.24 i experienced a massive boost in Doom, removing the bootlenet caused by CPU as it can be observed in-game frametime stats. And it's not a subjective appreciation: http://www.pcgamer.com/doom-benchmarks-return-vulkan-vs-opengl/2/

On the other hand, it's a fact that the considerable gains for AMD under Vulkan/DX12 are based in the horrible implementation of DX11 and OGL by AMD. For Nvidia it's more difficult to improve what is already optimized

 

Isn't that a myth? An AMD R9 Nano performs the same as a GTX 980 Ti. Which is ridiculous.

Can you explain how this is "nothing more to optimize for nvidia"?

 

No, it's not a myth and it's not that uncommon for performance to be similar, especially in newer engines.

 

Why?

 

I think he mean that DX11 is quite slow on AMD software but very fast on nVidia software. DX12/Vulcan on the other hand seems to be quite fast on AMD but slower on nVidia. This means that there is little left to do in the optimization area regarding DX11 for nVidia users but a lot to do in the DX12 department. So far the only game that has given me a boost in DX12 is The Division, all others have a negative impact on the FPS.

 

Bunch of reasons.

For starters the longer consoles are running on AMD, the more that game developers are going to target engines for AMD hardware. For example, GCN has been relatively weak in geometry processing - developers know this so they will tend to favor designs that avoid geometry (which is Nvidia's strength and which is why most of their gameworks stuff is so heavy with tessellation). Polaris obviously made up some of the difference so it could shift around again (Scorpio) but the point is that devs are designing their stuff to AMDs strengths and in some cases those are Nvidia's weaknesses. Thus when newer games inevitably get ported AMD's older gen cards, based on the same arch, they simply handle the work better than Nvidia's. Compute performance is another aspect where Nvidia essentially falls behind. Kepler for example had terrible compute performance and as a result is falling off harder in more modern titles that make use of compute.

Async Compute is another factor - AMD's architecture is relatively wide compared to Nvidia's and has some balance issues as far as geometry and stuff go (well did, again Polaris improved in this area) there are a lot of idle portions of the GPU that are briefly doing nothing depending on the scenario. Async compute essentially makes use of those idle portions by creating a separate graphics/compute queue(s) and filling those idle spots when they become available. Nvidia doesn't benefit nearly as much because they don't have the idle spots to begin with.

The reason why Nvidia loses performance in some, if not all cases of DX12 is because some their optimizations are essentially being lost. Nvidia has in the past and I'm sure still does a ton of optimization at the driver level. Some of these optimizations are being replaced by developer code at the game level with DX12, so the performance enhancements they bring are lost.

As far as AMD's DX11 driver is concerned - it did have some problems with overhead but AMD has largely resolved that in the last year or so. Their DX11 performance has been catching up and in some cases older cards are eclipsing the Nvidia counterparts. The hardware was essentially being underutilized. PRMinisterGR has a good post about it here - http://forums.guru3d.com/showthread.php?t=398858 AMD is essentially fixing these problems and since the last 5 years of cards they've released have all been GCN based, those cards are all reaping the benefits.

But yeah, it's not really as simple as most people make it out to be. Both companies do things slightly differently and favor certain features over others and in the end performance can shift as some of those advantages/disadvantages are utilized or not.

 

As it stands now:

- Nvidia drivers aren't getting across the board improvements; maybe a game here and there.

- AMD drivers are getting much better. For example, when the RX480 was introduced, a GTX980 was much faster. As I type this, the difference is about ZERO in DX11 games (might differ game to game). The RX480 is faster in DX12/Vulkan games.

I wouldn't get an nVidia card now anyway. The cards aren't bad, but nVidia is pricegauging because they can get away with it at the moment.
They're incredibly overpriced and are going to drop in price the moment Vega is introduced.

 

Meanwhile over at AMD, perfect, 100% mGPU scaling,...

 

And people lol'd when I said 480X will be near stock 980Ti in certain situations, basically on par with stock 980GTX or just a little faster like Gromuhl'Djun said.


And what's surprsing to me lately, is how effective it really is considering it has only 32ROPs, now imagine full Vega with 64ROPs or maybe even 96?

I remember NV ROPs were optimized in the past, but lately not so much, or AMD really did a great job.



Speaking of DX12, I saw a boost in ROTR and in The Division compared to DX11, both at same settings, same with new v2 AoS benchmark.


Only Doom not so much, well it ran fast @ OpenGL by default, didn't test more with latest Vulkan driver.. Last time was ~ R370 but vsync felt really bad, so I stopped and returned to OpenGL.