Quick comparison: Geforce GeForce 378.66 versus 378.78 DirectX 12 performance

DX11 launch games were the same. Alot of them only had DX11 api path just for having AA with deferred lighting. After about a year of DX11 games they really started to shine.

Sadly it will take time but I think DX12 will be around a long time. We are 7 years into DX11 titles now.

I didnt expect anything mind blowing graphics wise from games. And honestly didnt expect them to run better either.

DX11 didnt really run better then DX10 or DX9 games. Infact most of the time DX9 ran better because of less graphics features being used.

 

Hitman is the only game that performs poorly on Nvidia hardware, everything else is probably as good as it can get.

DX12/Vulkan are still in the beta stages, and not hugely exciting for Nvidia owners, but AMD owners get the have more fun with it though.

 

Well
dx9 was limited to ~9-11k drawcalls
Dx11 up to ~18k drawcalls
Opengl 4.5 ~ 28k calls
Dx12 up to ~50k or more, apparently up to ~ 100k drawcalls

The higher the calls number less driver/ api bottlenecks, potentially making richer worlds and more interaction with it. And that's what 3dmark api overhead benchmark shows. But in reality it doesn't deliver, now idk why but I think its mostly because of excessive postprocess fx crippling gpu pipeline.

E.g. overdone DOF fx, brutal shadow techniques - hfsl, pcss (both bs to sell nv gpus), excessive tessellation, more postprocess bloom, blur on top, etc.

If all that was made in optimization in mind it would be a lot different, but hey you gotta make fx to cripple gpu to sell new ones.. hypocrite development at best.

 

Do games ever need that many draw calls?

3D Mark looked impressive with the api test, but it's not relevant for the majority of games, even RTS games don't benefit, and in some cases regress under DX12 on Nvidia hardware.

 

I mean all those techniques enhance the image quality.. what would you rather them do? Just increase polygon count to 500 billion? Let framerates go up to 10,000 fps?

And I know you're probably going to respond with "improve lighting with raytracing" or some nonsense, which is completely outside the realm of current generation performance. All the HBAO/GI/HFTS techniques you're describing as "fx to cripple GPU" are all optimized systems designed to replicate raytracing and in some cases implement raytracing techniques.

There has also been giant strides in material shaders and stuff too - things like photogrammetry significantly improve the look of materials with very little performance impact.

So really not sure what else you'd want them to do.

 

I wonder how many people are gonna' be suckered into getting Windows 10 over this.

 

Yes they are, but there is a difference if a thing is done right or tasking on purpose. And that's what most fx are, there are exceptions. E.g. ID6 engine fx or Unreal engine to certain point and few more, Snowdrop engine is ok too - ok The Division was crippled compared to initial showcase rls, but I guess they had to drop gfx detail so those "fancy nv shadows" don't take too much of gfx pipeline.
Imo default shadows @ high looks good enough if not better at times, compared to both pcss & hfsl combined.

I would like to see both richer worlds and of course better detail, idc about accurate raytracing bs, mimic is enough as long as it still looks ok and Im 100% sure it can be done in effective way.



Another good example was Deadrising3 or Metro2033 DOF, in both cases it was made to cripple gpu on purpose not that it somehow looks more advanced and accurate, Metro2033 Redux - fixed its DOF I wonder why heh


We could have CGI mimic gfx 2-3yrs ago np, but no. They rather push "new gfx techniques" so it tasks gpu more while still looking like some 2010 or older game. They showcased Unreal Samaritan in 2012 running on a midget GK104 for crying out loud and fast forward 5yrs later, hm hm.. Yaa, see where Im going at, BatmanAK or MKX is somewhat a small glimpse of that, but still nowhere near.

 

Just played some Titanfall 2 on this driver, with fps counter active, and fps seemed & felt higher. I have recently, a few days before reduced Titanfall 2 graphics settings to hit 144fps with more regularity, so I haven't played all the maps on the old vs new driver under the new graphics settings, but unscientifically I think this driver has some improvements in this game, seems pegged at refresh rate more often.

I have compared synthetics & game benchmarks between previous hotfix driver and this latest one, and all were the same:
-3DMark Firestrike
-3DMark Timespy
-3DMark11
-F1 2012
-F1 2015
-Batman Arkham Knight

 

Sorry the drivers didn't work out for you. I am seeing large gains ashes of the singularity and rise of the tomb raider, both which are cpu heavy games.

 

Because dx12 is supposed to help with cpu bottlenecks and he uses a cpu where he would be the least bottlenecked, then criticizes the results. :stewpid:

 

I'm never going to criticise graphical additions, the best image quality quality is a combination of all these small things.
Samaritan demo itself is full of these sort of post processing effects, and is all scripted instead of gameplay, modern games have already passed it.

So...anyone tested these with Hitman on older Nvidia hardware, or am I going to have to delete some games and test myself.

 

http://forums.guru3d.com/showpost.php?p=5404488&postcount=7

 

Cheers, might try the game again as before it was just under my ideal performance.

 

so... directx 12 on nvidia hardware is still mostly emulated by software?

 

Why do you say that?

 

Because of gains via graphics driver in a low level api. staying outta this one. lol

 

No GPU is emulating anything via software. NVIDIA's scheduler is statically partitioning the GPU on the SM level for Graphics/Compute/Copy tasks between each draw call, AMD does it on the SIMD level and it schedules in hardware.

They both follow the Graphics/Compute/Copy engine paradigm of DX12, just with different approaches. NVIDIA's is more efficient, but harder to use for a lot of different programs running on the GPU, AMD's is less easy to fully utilize but better once you have a lot of small programs running on the GPU.

Where "programs" any kind of compute/graphics/copy task running concurrently.

 

I just was wondering from what that performance increase come... In 9xx nvidia cards directx12 conservative depths and sad4 was emulated by software, so driver-software changes could impact performance..

 

Do you have a source for any of this ?

 

@hilbert

I'm really confused by this quick review thing you did here... Specifically this

This makes no sense whatsoever in light of your conclusion, you basically ran a test to validate the claims made by NV but you tested in a completely different context.

NV didn't make ANY CLAIMS WHATSOEVER regarding a performance improvement from the last WHQL.

This is for all intents and purposes, a futile test. If you said "hey we are going to test performance improvements between 378.78 and 378.66 " that would have been fine, but then you can't make any conclusions whatsoever about the claims nvidia made.

It's very easy to test, use old driver, benchmark. install new driver, benchmark. Game updates have absolutely nothing to do with it, because you're testing the latest game binary with older and newer drivers.

What next, let's say nvidia claims 10% improvement from core OC on the 1080ti (based on FE), then you tested a high end AIB card that's already at 2GHz and conclude that nvidia lied and you only get 2% improvement.

It's utterly retarded