• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Frame Rating: Battlefield 4 Mantle CrossFire Early Performance with FCAT

Caporegime
Joined
17 Mar 2012
Posts
50,839
Location
ARC-L1, Stanton System
http://www.pcper.com/reviews/Graphi...eld-4-Mantle-CrossFire-Early-Performance-FCAT

Its an interesting read, but....

Multi-GPU results are more confusing. The orange line that represents Mantle CrossFire results starts out "flat" like we saw on the previous page but then appears to open up and scales upward peaking around 34% faster than the single card option. At the same time, the D3D11 CrossFire result actually stays quite a bit lower, giving Mantle the advantage. If you look at our Frame Times graph you'll see that under DirectX, we have quite a bit of frame time variance that does not exist under Mantle.
These results were consistent and repeatable and make multi-GPU configurations at 1920x1080, even with the A10-7850K, less than desirable.
Things actually get even more interesting at 2560x1440. Mantle scales by as much as 85% going from a single 290X to a pair of them in CrossFire while DirectX only sees a 42% advantage. And again, the consistency of the frame times in the multi-GPU Mantle result is much better than what we are seeing with D3D1 - by a lot.
I just don't understand how he can be confused by this.

He gets better and more consistent CF scaling in Mantle than in DX because the A10-7850K CPU he is using for it is not nearly as bottlenecked by Mantle as it is by DX.

Does he not understand what Mantle does? that would be shocking for an apparently experienced reviewer.

Good Grief. all he has to do is this to see what's going on.

Mantle



DX

 
Last edited:
That's some impressively flat graph for mantle, regardless of the frame-rates.

I would probably take the lower-frame rate and a line as flat as that over faster FPS.

First time I have seen something like that, I bet they checked their software a few times to see if it was broken!
 
That's some impressively flat graph for mantle, regardless of the frame-rates.

I would probably take the lower-frame rate and a line as flat as that over faster FPS.

First time I have seen something like that, I bet they checked their software a few times to see if it was broken!

Here is another incredibly flat line :D

Mantle



And one thats not

DX

 
So from PCpers testing at 1080P and in Crossfire, it is better to stick with DX for the higher frames or the more smoother experience, go with Mantle. Very interesting.
 
I wonder if these test was done after last bf4 patch that added frame pacing? Because it's not even working I had to disable it in user.cfg

Has matt pointed out in the bf4 thread. We think dice as added the frame pacing option I to bf4 but amd need to release there driver to work with it better.
Has iof right now game play is better with frame pacing in game disabled.
 
Frame pacing by it's nature reduces FPS to some degree. You can't pace frames if you can't average them out, that means delaying some frames to change the difference. It's generally the whole point, but I don't know how much it's hurting performance in this particular situation.

Either way, Mantle across the board has been showing smoother frame rates, lower frame latency and less spikes(outside of some rarer ones in BF4 which I think is BF4 hitting memory limits which they've suggested they need a rewrite of the memory management to work with complete control, the consequence of Mantle being available way after the game was mostly done).

One of the reasons is, and pretty much every dev complained about this on DX for years, is that DX does what it wants. That means sometimes it will randomly compile stuff out of order and cause a latency spike before it can do the next thing.

On consoles in general and now Mantle, and most game design you work to a latency. You can precisely tell how long each part of a frame building takes, how long each part of the engine will take to process and you can design an engine to use say 8ms of compute per frame, then start adding on extra processing effects that add on a certain amount of latency. But these things are predictable. For instance TressFX added something like 4ms of work, and that in TressFX has been cut to 2ms or something, so effectively halving the performance hit for enabling it. Games are a series of things adding up to a certain amount of time each frame. When you can predict the precise order of execution you can control the latency every frame with no surprises.

DX is a black box, it does effectively what it wants after you send it data, latency spikes around as it does things.

A high level API that hides control will always be less predictable and have such spikes. The reason consoles often target an FPS is because precisely that they are targetting a x ms worth of work on each frame and know the exact hardware. The reason frame rate is more variable on consoles is because they can't control the ms per frame.

Mantle, and likely DX12 will always, always be smoother than DX11(other high level api's and all previous version).
 
Right on the money, pcper done test with dice frame pacing what is not working well.
https://mobile.twitter.com/ryanshrout/status/451398434440753152?screen_name=ryanshrout

I noticed a bug with Mantle and crossfire, dices implementation of frame pacing hurts fps badly. Disable it by typing 1/2
RenderDevice.FramePacingMethod 0 into the console.Once you do this fps shoots up and it remains just as smooth. Try it.
That explains the lower high FPS..

He probably needs to run this again.

And not be so surprised at the better more consistent CF scaling in Mantle :p
 
Last edited:
Back
Top Bottom