The Division DX12 Patch Benchmarks

[Bacon1]

https://www.overclock3d.net/reviews..._the_division_directx_12_performance_review/1

1080 shows improvements @ 1080p due to reduced CPU usage

https://www.computerbase.de/2016-12/the-division-dx12-benchmark/

 

[Maverick177]

AMD Finewine technology.

 

[Sweepr]

https://www.overclock3d.net/reviews..._the_division_directx_12_performance_review/1

1080 shows improvements @ 1080p due to reduced CPU usage

Who cares where the improvement comes from? GTX 1080 gains more from DX 12 than Fury X according to ComputerBase at 1080P, with both CPUs. Very impressive 13% boost.

 

[Bacon1]

GTX 1080 gains more from DX 12 than Fury X according to ComputerBase at 1080P and 4K, with both CPUs

The 1080 goes slower @ 4k from DX12 not faster. That's why I specifically mentioned gains @ 1080p. Not sure why you re-posted the same images from my post.

 

[Sweepr]

The 1080 goes slower @ 4k from DX12 not faster. That's why I specifically mentioned gains @ 1080p. Not sure why you re-posted the same images from my post.

The gains for AMD cards at 4K are minor at best, margin of error? The real suprise here is GTX 1080's bump at 1080P.

 

[raghu78]

The gains for AMD cards at 4K are minor at best, margin of error? The real suprise here is GTX 1080's bump at 1080P.

The GTX 1080 shows a 5% decrease in perf at 4k with 6700k. But you wouldn't want to point out that.

 

[richaron]

I don't understand why they even bother doing a piece about DX12 performance without including results from systems with mid or low range CPUs.

They're testing a thing with a major selling point of reducing CPU load with a high end CPU and Ultra gfx settings... Nice. So they're printing the worst performing, and likely least representitive data (I'm guessing most people game with mid-range or old high end CPUs).

If anything that leaves the reader less informed about actual performance; as they say: A little knowledge is a dangerous thing.

 

[96Firebird]

Hope they didn't work too hard on this near-pointless patch.

 

[TheELF]

I don't understand why they even bother doing a piece about DX12 performance without including results from systems with mid or low range CPUs.

They're testing a thing with a major selling point of reducing CPU load with a high end CPU and Ultra gfx settings... Nice. So they're printing the worst performing, and likely least representitive data (I'm guessing most people game with mid-range or old high end CPUs).

If anything that leaves the reader less informed about actual performance; as they say: A little knowledge is a dangerous thing.

Dx11 driver for medium settings is ~10% for the weakest haswell celeron
(nwgf2=nvidia geforce) so that's the theoretical most you could gain if dx12 where to be magic that doesn't need any driver,it's lower level and needs less cpu but it get's redistributed to the other threads so good luck finding out how much it needs.
Gain is one thread less going from 60 to 59 threads...
You can see that in dx12 the game runs with ~80 instead of with ~77% which could also be statistical margin of error and the CPU running ~2% idle,so total of 5% cpu usage difference(also not very convincing that it's not statistical margin of error), but the GPU going from ~91 to ~99 is definitely a good thing.
Dx11 would have little freezes now and again on both benchmark and game play dx12 had nothing of the sort fps would be a bit lower on more demanding scenes but it would be a smooth transition.

 

[richaron]

Dx11 driver for medium settings is ~10% for the weakest haswell celeron
(nwgf2=nvidia geforce) so that's the theoretical most you could gain if dx12 where to be magic that doesn't need any driver,it's lower level and needs less cpu but it get's redistributed to the other threads so good luck finding out how much it needs.
Gain is one thread less going from 60 to 59 threads...
You can see that in dx12 the game runs with ~80 instead of with ~77% which could also be statistical margin of error and the CPU running ~2% idle,so total of 5% cpu usage difference(also not very convincing that it's not statistical margin of error), but the GPU going from ~91 to ~99 is definitely a good thing.
Dx11 would have little freezes now and again on both benchmark and game play dx12 had nothing of the sort fps would be a bit lower on more demanding scenes but it would be a smooth transition.

I was sort of counting the multi threading thing as reducing load also. But even 5% reduced driver overhead is nothing to sneeze at.

And I can come up with an extreme example also:
A quad core CPU bound with a single thread can become 300% faster when properly threaded in DX12!

I suspect the results in reality would be somewhat less, but it's hard to find data which isn't skewed. Like I said, relying on most of what's out there will most likely skew ones ideas also.

 

[Red Hawk]

Hope they didn't work too hard on this near-pointless patch.

The point is probably to get accustomed to DX12 development. We've seen that there's a learning curve to it - like the id devs said in their interview with Digital Foundry, it's best to start as early as possible. Even if the benefits aren't huge at first.

 

[greatnoob]

They should really start testing these DX12 games on lower end cards too...

Hope they didn't work too hard on this near-pointless patch.

Hope you didn't work too hard on this near-pointless comment.

 

[frozentundra123456]

Those results look strange. The computerbase test with a 6700k shows an increase in performance, while the overclock 3d test with a hex core (which should show more gains with DX12) shows a regression. Max performance is about the same though, so maybe ~ 100 FPS is the gpu limit.

 

[Bacon1]

The gains for AMD cards at 4K are minor at best, margin of error? The real suprise here is GTX 1080's bump at 1080P.

The exact same bump you said "Who cares" about when I pointed it out in the OP?

The similar bump that a Fury X gets? 11-14% @ Guru3d, 9% @ computerbase.

Guru3d showed the 1080 still decreasing @ 1080p as they had higher FPS already, so not as CPU bound as computerbase.

Not sure why people are trying to say that DX12 is bad, it obviously can help. Once engines can move away from their DX11 requirements and go pure DX12 we'll see better gains as well.

 

[SlickR12345]

If tacked on DX12 support can bring in up to 13% gains for AMD cards, then imagine what a pure DX12 engine and game, optimized from the start for DX12 render and DX12 cards can do!

 

[TheELF]

If tacked on DX12 support can bring in up to 13% gains for AMD cards, then imagine what a pure DX12 engine and game, optimized from the start for DX12 render and DX12 cards can do!

Exactly the same!
dx12(d3d12) is only the graphics/rendering part,the game logic and all of the rest of the game still has to follow the same rules as ever.
Just look at console games where something like dx12 exists forever it's always only better graphics while the game itself is either very simple so it runs easily on small cores or it runs crappy even though the game is designed for low level from the get go.
Remember that most if not all console games ARE being coded for dx12 from the ground up,devs just use dx11 for windows because it's optimized like crazy.

 

[TheELF]

And I can come up with an extreme example also:
A quad core CPU bound with a single thread can become 300% faster when properly threaded in DX12!

You don't need dx12 to do this and no it won't.
It might be able to push 3x the fps when it's only rendering graphics(cinebench difference between 1 and all cores) but then that would be a pretty boring "game".

 

[Dygaza]

TheELF's process hacker pictures show how good threading the game has even under DX11. DX12 version seems to be actually quite good to even get that kind of benefits. If only all DX11 games would have this equal threading, but usually you have main game logic thread being a lot heavier than others, making game rely on single thread performance.

 

[Dygaza]

You don't need dx12 to do this and no it won't.
It might be able to push 3x the fps when it's only rendering graphics(cinebench difference between 1 and all cores) but then that would be a pretty boring "game".

Yep. 3dmark API test pretty much shows you what kind of game you would be playing.

People seem to forget that d3d12 only can split rendering workloads between cores. It doesn't do anything for AI, physics, gamelogic, netcode etc. These are things outside of graphics API. You need to build your engine very well, and really build for that parellism. If your game is AI heavy for example, and the engine is only handling AI on one core, chances are very high that you will always have that thread running on high load, no matter what API you use.

Ashes of singularity is one of the best examples how you really build your engine to spread everything across multiple cores. Deux Ex DX12 showed that even when rendering was split between multiple cores, main thread was still limiting factor. It's a bit better after few patches.

 

[richaron]

You don't need dx12 to do this and no it won't.
It might be able to push 3x the fps when it's only rendering graphics(cinebench difference between 1 and all cores) but then that would be a pretty boring "game".

No you don't need DX12, and 300% more = 4x the fps. But as I stated it was an extreme example, but still within the realm of possibility. An example showing a very CPU bound situation, which barely any of these "DX12 tests" cover. And a CPU bound situation should be the first test they do given the selling points on DX12.

 

[frowertr]

The game is poorly optimised on PC for both DX 11 and DX 12. 1080 users are frequently complaining about frame rate stuttering and memory leaking. It is a fun game with patch 1.4 but its frustratung to play with the graphical glitches.

 

[tviceman]

Yet another title that shows Nvidia has "potential" at DX12 but fails to show any consistent gains - and actually loses performance - except in special case scenarios. Nvidia needs to work harder with their DX12 optimizations, whether it is on the driver side or (most likely) with developers. As I've said, the gains are there to be had but it "appears" to be more difficult and/or more time consuming to implement the necessary code to make said improvements.

 

[TheELF]

The game is poorly optimised on PC for both DX 11 and DX 12. 1080 users are frequently complaining about frame rate stuttering and memory leaking. It is a fun game with patch 1.4 but its frustratung to play with the graphical glitches.

Yeah unless you play it with frame limiter it gets really bad,thus my screenshots are both at 30fps locked.
But it's not badly optimized,you get good fps it's just badly ported,I mean comm'on invisible enemies galore...
unless this happens on the console as well then yes, bad coding.

 

[96Firebird]

The point is probably to get accustomed to DX12 development. We've seen that there's a learning curve to it - like the id devs said in their interview with Digital Foundry, it's best to start as early as possible. Even if the benefits aren't huge at first.

Sure, but does patching a DX11 game with DX12 lead to the same experience as creating a DX12 game from scratch? We keep hearing that the real benefits of DX12 will come from those built from the ground up with DX12 in mind, which leads me to believe the learnings from a patch will not transfer to a from scratch game.

Hope you didn't work too hard on this near-pointless comment.

Pot... kettle.

 

[Red Hawk]

Sure, but does patching a DX11 game with DX12 lead to the same experience as creating a DX12 game from scratch? We keep hearing that the real benefits of DX12 will come from those built from the ground up with DX12 in mind, which leads me to believe the learnings from a patch will not transfer to a from scratch game.

Getting experience with coding and programming for DX12 should be helpful when it comes time to make a game/engine for DX12 from the ground up. Doesn't make sense for it not to.