DX12 Is Able to Handle 600K Draw Calls, 600% Performance Increase Achieved on AMD GPU

[Silverforce11]

The min fps are on the graph I posted. About the same ~10% increase.

In non CPU limited scenarios, how is it even boosting performance at all? It gets some gains in Sniper Elite 3 at very high quality settings too.

We've seen in BF4 64 MP, Mantle does & can give major performance boost to min/avg fps, especially for multi-card users.

That should apply to DX12 as well.

Minimize the effect of CPU bottleneck occurs for low-end setups (weak GPU) or high-end setups (multi-GPU). For a balanced setup, the gains will be minimal.

 

[Shehriazad]

In non CPU limited scenarios, how is it even boosting performance at all? It gets some gains in Sniper Elite 3 at very high quality settings too.

We've seen in BF4 64 MP, Mantle does & can give major performance boost to min/avg fps, especially for multi-card users.

That should apply to DX12 as well.

Minimize the effect of CPU bottleneck occurs for low-end setups (weak GPU) or high-end setups (multi-GPU). For a balanced setup, the gains will be minimal.

Actually...wouldn't the CPU/driver overhead with DX9/11 also literally "slow down" the GPU?(or make it "wait")

Now I'm not a pro with these things...but last time I checked a GPU still heavily relies on a CPU to accompany it...so if a CPU needs like wayyyy less time to send the data through its' pipeline, wouldn't the GPU benefit as well?

So I'd obviously expect balanced systems to also see visible gains since basically the application is now "closer to the metal".

 

[bystander36]

In non CPU limited scenarios, how is it even boosting performance at all? It gets some gains in Sniper Elite 3 at very high quality settings too.

We've seen in BF4 64 MP, Mantle does & can give major performance boost to min/avg fps, especially for multi-card users.

That should apply to DX12 as well.

Minimize the effect of CPU bottleneck occurs for low-end setups (weak GPU) or high-end setups (multi-GPU). For a balanced setup, the gains will be minimal.

That really depends on the games you play. That may be true for the games you play, but not for at least a couple genre's. MMO's and RTS's in particular, are very CPU bound in general. Most online games are much more CPU bound as well.

Starswarm is very much a RTS demo. I do agree that we won't see massive improvements, like the 600% quoted, but we will see good gains in certain genres, even on balanced systems. It may also be related to your FPS target as well.

 

[DarkKnightDude]

I'll believe it when I see it.

Also, "DX9 uses max 6000 draw calls vs 600,000 on dx12." I'd love to see some evidence of this instead of a random forum post from a banned user.

 

[desprado]

I hope it does both on AMD and Nvidia GPU.No doubt this year GDC will be the be best.

 

[boozzer]

how well would this translate into pc gaming performance? no way it is the 600% right? that would be insane and probably the biggest innovation in pc gaming that I can think of. nothing brings 600% increase just like that.

 

[Red Hawk]

Yeah but AMD's DX11 drivers suck in this game (compared to Nvidia's DX11). It would not be unrealistic for their DX 12 drivers to need work as well. Keep in mind i'm comparing it to a well done DX 12 implementation (Nvidia) who is edging out mantle significantly.

And it's not possible that AMD's Mantle drivers also could use some work and still have potential performance to unlock? Comparing Nvidia's DX12 performance to AMD's Mantle performance is pointless when they're different hardware at clearly different points of optimization. Comparing AMD's Mantle and DX12 performance is more appropriate.

 

[III-V]

The 600% increase is so high because AMD does so terribly at DX 11.

They do so poorly in that test because their DX11 drivers do not support multithreaded rendering. Mantle and DX12 do. Nvidia's supported it for a while, e.g. Civ V.

And it's not possible that AMD's Mantle drivers also could use some work and still have potential performance to unlock? Comparing Nvidia's DX12 performance to AMD's Mantle performance is pointless when they're different hardware at clearly different points of optimization. Comparing AMD's Mantle and DX12 performance is more appropriate.

That'd be a bit silly for their own API to have a significant amount of room left for improvement.

 

[3DVagabond]

I hope it does both on AMD and Nvidia GPU.No doubt this year GDC will be the be best.

As with the other PC API's, all the IHV's have to do is decide to support it.

 

[ayylamayo]

OpenGL-Next. Of course, when you will see a ES version remains to be seen.

Nope, no more ES versions. Mobile GPUs and OS' have started transitioning into desktop-level APIs.

 

[Red Hawk]

That'd be a bit silly for their own API to have a significant amount of room left for improvement.

It's not the API that would have room for improvement, it's their hardware that would have room for improvement with the API, just like with DirectX 12.

 

[ayylamayo]

Here's the go:
1) click bait title
2) supposed "600%" increase in DRAW CALLS which isn't that big of a deal
3) would come down to a small percent jump in fps of course [way] below < 30%

This is similar to allowing devs to adding slightly more details on characters, environments, etc. at the same fps you're used to.

 

[Dribble]

Starswarm just exists to show off draw call improvements. If someone had actually tried to write that demo properly the improvement would be minimal. In other news it showed AMD haven't bothered to optmise DX11 probably because they were to busy working on mantle. I suspect most AMD users would have been better off without mantle which was mostly broken and instead had those same devs make DX11 work better.

 

[3DVagabond]

Starswarm just exists to show off draw call improvements. If someone had actually tried to write that demo properly the improvement would be minimal. In other news it showed AMD haven't bothered to optmise DX11 probably because they were to busy working on mantle. I suspect most AMD users would have been better off without mantle which was mostly broken and instead had those same devs make DX11 work better.

There is no evidence at all that AMD has compromised their DX11 drivers. Take the Omega release as an example. They're still supporting DX as much as they ever have.

 

[dacostafilipe]

In other news it showed AMD haven't bothered to optmise DX11 probably because they were to busy working on mantle. I suspect most AMD users would have been better off without mantle which was mostly broken and instead had those same devs make DX11 work better.

This mostly shows that investing so much time/money in DX11 optimisation was not worth it. What was needed, was a new API, not optimising for a broken one.

Mantle was a proof of concept that made the industry move, and now we have DX12 and glNext is just around the door. There no reason to bash around Mantle, it did it's job and should continue doing it in glNext if the rumours are true.

 

[Dribble]

There is no evidence at all that AMD has compromised their DX11 drivers. Take the Omega release as an example. They're still supporting DX as much as they ever have.

The evidence is in the graphs above, AMD does really badly under DX11 in starswarm, where as nvidia does much better. This because nvidia having no mantle optimised DX11 as best they could to make it work ok, where as AMD didn't bother relying on mantle. Unfortunately most games don't support mantle, and even when it does work it's often buggy. Having no mantle and a DX11 driver that performed like the nvidia one would have worked out better most of the time.

This mostly shows that investing so much time/money in DX11 optimisation was not worth it. What was needed, was a new API, not optimising for a broken one.

You optimse for whatever the industry are using broken or not as that's what users need to play the games. Most games right now use DX11 so we need the best DX11 drivers possible.

 

[Hitman928]

The evidence is in the graphs above, AMD does really badly under DX11 in starswarm, where as nvidia does much better. This because nvidia having no mantle optimised DX11 as best they could to make it work ok, where as AMD didn't bother relying on mantle. Unfortunately most games don't support mantle, and even when it does work it's often buggy. Having no mantle and a DX11 driver that performed like the nvidia one would have worked out better most of the time.



You optimse for whatever the industry are using broken or not as that's what users need to play the games. Most games right now use DX11 so we need the best DX11 drivers possible.

The problem is your theory falls apart when comparing any games that don't support mantle. AMD has actually increased their position against nvidia across the spread of dx11 games, contrary to what you are proposing.

 

[dacostafilipe]

You optimse for whatever the industry are using broken or not as that's what users need to play the games. Most games right now use DX11 so we need the best DX11 drivers possible.

You are absolutely right and thats why AMD didn't implement DX11's Command Lists, because the benefit is really small and most games not even bother using it.

Edit: Obviously, I forgot something

 

[PPB]

Command lists are fubar according to the frostbite engine dev. I would rather trust someone that touched a line of game code, let alone produce one of the best optimized game engines to this day, than to the usual viral poster of this forum.

 

[ShintaiDK]

Command lists are fubar according to the frostbite engine dev. I would rather trust someone that touched a line of game code, let alone produce one of the best optimized game engines to this day, than to the usual viral poster of this forum.

The same guy that championed mantle? ()

 

[Headfoot]

Command lists are fubar according to the frostbite engine dev. I would rather trust someone that touched a line of game code, let alone produce one of the best optimized game engines to this day, than to the usual viral poster of this forum.

100000% agreed.

I bet you the engine devs from DICE and Crytek would have something to say about Frostbite 3 and CryEngine (3, e.g. Crysis 3 iteration) being called DX9 era. Yeah, hardly. I dont exactly recall any game scaling to 8 cores during the DX9 time frame...

 

[witeken]

What about Intel? How will DX12 affect (low power) laptops?

 

[Midwayman]

It would not be unrealistic for their DX 12 drivers to need work as well. Keep in mind i'm comparing it to a well done DX 12 implementation (Nvidia) who is edging out mantle significantly.

You mean AMD who has literally spent *years* ahead of this on a similar low level API? If anything its reasonable to expect AMD to have better low level drivers out of the gate than Nvidia as long as we're going for baseless speculation. In a few years, who knows. I do trust Nvidia's driver team a bit more, but lets not pretend AMD doesn't have a head start here.

That doesn't even take into consideration that it's AMD hardware in both consoles and there is reason to believe that more time is going to be spent making AMD hardware work than Nvidia. If mantle has taught us anything, its that the work done by the game dev is going to have even a greater impact on how well the low level stuff works.

 

[PPB]

The same guy that championed mantle? ()

The same guy that also made one of the most optimized dx11 games without command lists. I bet 10 bucks he knows something more about dx11 optimizations than both of our heads combined.

 

[ShintaiDK]

The same guy that also made one of the most optimized dx11 games without command lists. I bet 10 bucks he knows something more about dx11 optimizations than both of our heads combined.

Since its not BF4, what game are you refering to?