DX12 here?

[monstercameron]

I will investigate
http://hexus.net/tech/news/graphics/85385-dx12-unreal-engine-4-elemental-demo-download-available/

In the video, above, you can see the UE4 Elemental Demo running with DX12. The demo has an FPS overlay in the upper right portion of the screen. Among the components in the PC system in use are; an Intel i7-4790K CPU @ 4.8Ghz, an EVGA Nvidia GTX 980 Ti @ 1366mhz core / 4100mhz memory and 32GB of DDR3-2400 quad RAM.

In the video comments CryZENx says that while the DirectX 12 version of the demo runs consistently at around 60fps, the DirectX 11 version fps sometimes drops as low as 42fps. This is on the same system, hardware, settings etc.

Other testers of the demo are having mixed results. Some have noted great uplifts in frame rates, like CryZENx, but others have not fared so well. Some people can't run the demo, despite seemingly having the requisite software/hardware. Such is the nature of test/in-development software.

 

[sam_816]

Sure, but why? Let the ppl who r paid to do it.... do it. U have a good setup to play games, go play some games mate. Having a technology is one thing and how devs understand it and choose to implement it is another. No offence (to anyone) but ppl on tech forums speculate and argue more about gaming technological advances then actually play n enjoy those advances.

Just my 'imo', no offense again.

 

[Nhirlathothep]

we re not all gamers!

 

[monstercameron]

WOW, cant believe how good that looked.
problems:
kaveri laptop struggles 80-120MS frames
cant tell if is d3d12 or d2d11, though there is a cmd argument to mess with
ran out of video memory 1GB allocated

 

[Sabrewings]

I have a nearly identical setup to the one in the link and I can confirm their results.

 

[monstercameron]

strange, the dx11 version seems to be smoother than the dx12 version. I think.
the cmd argument to enable -D3D12 .

d3d11

d3d12

 

[hawtdawg]

edit: delete this

 

[hawtdawg]

Here are my DX11 results using a 3770K at 4.6 and a 980ti at 1354 core/3506 VRAM

frametimes:

FPS

rank

cpu usage

 

[hawtdawg]

DX12 results

frametimes

FPS

frameranks

cpu usage

 

[hawtdawg]

I got very similar results with DX11 and DX12. maybe slightly better minimums with DX12 over DX11, but certainly nothing like OP's post indicates. Neither truly dipped below 60fps (or 80 for that matter) with the exception of a few outlier frames, but both have a handful of frames below 60fps. My CPU is not as good as OP's article, and I left my 980ti at stock settings, so I shouldn't have gotten better results like i did. I also dont really see any significant difference in CPU usage.

 

[Skurge]

Did Epic port this?

 

[AtenRa]

Intgeresting,

Core i5 2500K Default
HD7950 @ 1GHz

Win 10 64bit
Catalyst 15.7.1

DX-12 no Vsync

 

[shady28]

Would be a lot more useful if you guys would click your CPU graph and select it to show CPU usage by logical processors rather than the aggregated graph (exc hawtdog who posted a graph of all cores over time) along with a graph of %GPU usage / % Memory controller usage (from GPU-Z).

It may be that the GPU load is not being evenly spread out across all CPU cores (hawt's graphs indicate this).

It could also be that the GPUs are simply not able to handle many more draw calls or become limited elsewhere first - so the main 'advantage' of DX12 becomes moot. Many developers have stated that DX12 will only result in small gains in the real world and not the huge boosts the press / media have built it up to be, we might be seeing that.

 

[Flapdrol1337]

Would be a lot more useful if you guys would click your CPU graph and select it to show CPU usage by logical processors rather than the aggregated graph.

Not that useful, threads often get moved from core to core faster than the graph logs.

Here I have a single thread of prime, 50% on each core.

 

[AtenRa]

Ok, run FRAPs with it and i didnt like what i show,

System as above

Frametimes

FPS

Rank

 

[shady28]

Not that useful, threads often get moved from core to core faster than the graph logs.

Here I have a single thread of prime, 50% on each core.

I know that, but if it were 100% true then hawt's graphs wouldn't show 5 cores loaded and 3 with little load. Hyperthreading makes it even more complex.

You can use process explorer to see what is going on with the threads though :

 

[shady28]

This demo appears completely GPU limited in DX11 and seems heavily threaded to me (see process explorer threads).

I didn't do FRAPS (uninstalled last week, might try re-install later) but framerates didn't fall below high 40s that I saw. GPU usage stays at 98%, with ~50% memory controller usage.

I got the impression there is a ton of physics going on.

 

[omek]

It's literally rendering in DX11 while using the 12's API and extensions. It's the same thing as rendering using DX10/10.1 while using DX11's shell with a command line argument I believe, not full-on but it probably is benefiting in certain ways.

This also wasn't an official release from Epic btw, a random user created a bat file to use the arguments and enable DX12.

I wonder if a low resolution test would show some type of benefit because at 1080p it's all GPU on a 290x - low resolution may actually show the benefit if there is any. The train whistle is being yanked and people are jumping on board without the wheels moving at all, myself being one of them. I really want to see what it offers but this is not it.

 

[Hi-Fi Man]

Ran it with the hardware in my sig. Direct3D 11 was actually a little bit faster for me than Direct3D 12. CPU usage was lower on 12 though.

Beginning scene peaked at ~100fps on D3D11 and ~90fps on D3D12.

@omek: it's called using feature levels. It seems like it's using D3D12 with feature level 11.

 

[monstercameron]

It's literally rendering in DX11 while using the 12's API and extensions. It's the same thing as rendering using DX10/10.1 while using DX11's shell with a command line argument I believe, not full-on but it probably is benefiting in certain ways.

This also wasn't an official release from Epic btw, a random user created a bat file to use the arguments and enable DX12.

I wonder if a low resolution test would show some type of benefit because at 1080p it's all GPU on a 290x - low resolution may actually show the benefit if there is any. The train whistle is being yanked and people are jumping on board without the wheels moving at all, myself being one of them. I really want to see what it offers but this is not it.

It isn't maxing out my cpu even at vga, it is gpu limited as denoted by the massive throttling on my laptops cpu

 

[SlowSpyder]

I don't know how to benchmark it, but it runs fairly decent on my FX @ 4.9GHz and 7970 @ 1050Mhz. During the demo it's 40 - 70FPS, when I can run around the map after the demo finishes it's pretty much stuck at 60-70FPS.

*edit - DX12

 

[AtenRa]

Ok just OCed to 4GHz, same problems with DX12. There is a lot of stutter with DX-12, but DX-11 was much much better as you can see from the pics bellow.

Core i5 2500K @ 4GHz
HD7950 @ 1GHz

Win 10 64bit
Catalyst 15.7.1

DX-11

Framtimes

FPS

Rank

 

[Simco50]

The demo is maybe running using dx12 but you wont see much of a difference with dx11. The demo is not optimized using the new features of dx12. In fact, dx11 and dx12 are very much alike and dx11.3 which is released together with dx12 will still be used in the industry because dx12 is more advanced, difficult and only useful for specific optimizations. Certain things you can code in dx11 in one line, may need a complex paragraph of code in dx12. The programmer said it was a port, so we don't really know how much is rewritten to get the benefits from dx12

 

[Despoiler]

Did Epic port this?

Nope. This is a demo put together by some random guy on the internet. It runs like garbage for a lot of people. It's not indicative of anything.

 

[cmdrdredd]

Yeah it's kind of useless. Need to use DX12 more fully to see any real difference.