The AMD Mantle Thread

[dacostafilipe]

I had to chuckle when motion blur was disabled and DX fps shot past Mantle fps.

Yes, but they did not show Mantle running without motion blur, so ...

 

[BrightCandle]

Motion blur is a really interesting setting for a variety of reasons:
1) It uses information from the previous frame, potentially the previous frame itself. This is something that is kind of annoying in DX and is normally simulated.
2) The motion blur could be per object not screen based, which would be a lot of additional draw calls.
3) Lots of people hate motion blur, and typically in games its worth almost nothing in terms of performance, turning it off rarely nets even 1 fps. Yet in this tech demo its a dramatic impact (5x) and entirely responsible for DX showing poor performance. I do wonder why that is.

So in many ways its an unfortunate setting to show the impact of, because normally motion blur has no real performance impact at all and it looks like its been done in a way to show off Mantle rather than how its typically done in DX, resulting in a skewed benchmark (as I think I predicted about 100 pages ago).

 

[Hitman928]

The narrator in the video said that to make it run smooth on dx they must turn off high end graphical effects like motion blur and MSAA. That makes a lot more sense as MSAA would cause a huge performance hit with so many individual objects being rendered. Along with motion blur, turning on (up?) MSAA could easily bring that system down to single digit fps. It's impressive that the mantle version allows the system to handle it pretty well.

 

[Gloomy]

They didn't turn off MSAA though. They turned off "multi-sample motion blur", whatever that is.

 

[Hitman928]

They didn't turn off MSAA though. They turned off "multi-sample motion blur", whatever that is.

Listened again and you are correct, she says "motion blur multi-samples" and I originally inserted a comma in there.

 

[BallaTheFeared]

A joke?

 

[sushiwarrior]

Dual graphics (APU + dGPU) was mentioned a few pages back, with Kaveri dual graphics performs pretty well (Oland is a perfect match for an A10, both same arch and number of cores, only difference is memory interface) and the frame pacing patch has been expanded to dual graphics in some current drivers, guessing they'll be released at the end of Jan.

They were absolutely terrible before (look up Toms Hardware review of it, it's abysmal) but now that APU's and GPU's are on the same architecture, it's working a lot better.

 

[ASM-coder]

So many of you suspect a conspiracy and that Oxide used motion blur
simply to burn DirectX, and make Mantle look good.
Are we all so jaded?
But is it possible that they use motion blur because it's a space
game, and one of the coolest effects in any space show/movie
(Star Trek, Star Wars) is when objects suddenly blur past you.
They did say they were trying to achieve Hollywood effects in real time.

I don't think BF4 or Thief are going to be influenced by motion blur.

 

[Hitman928]

After some brief research it looks like motion blur multi-sampling is a more accurate/better way of doing motion blur. Rather than just "smearing" the screen when in motion (I am one that will always turn this off), it will take multiple samples of where the object is along the path (according to an oversampling rate) so as to give you a clearer view of objects with motion blur effect around the object, if that makes sense. Not sure how they do this with a dynamic, user controlled motion source, but that's why they're the developers. It makes sense why it would take such a huge toll now, if I'm understanding it correctly.

 

[Gloomy]

Mantle enables next-generation non-blurry motion blur technology!

 

[Skurge]

After some brief research it looks like motion blur multi-sampling is a more accurate/better way of doing motion blur. Rather than just "smearing" the screen when in motion (I am one that will always turn this off), it will take multiple samples of where the object is along the path (according to an oversampling rate) so as to give you a clearer view of objects with motion blur effect around the object, if that makes sense. Not sure how they do this with a dynamic, user controlled motion source, but that's why they're the developers. It makes sense why it would take such a huge toll now, if I'm understanding it correctly.

Doesn't that then meant that MSAA will also be accelerated to as similar degree?

 

[BrightCandle]

After some brief research it looks like motion blur multi-sampling is a more accurate/better way of doing motion blur. Rather than just "smearing" the screen when in motion (I am one that will always turn this off), it will take multiple samples of where the object is along the path (according to an oversampling rate) so as to give you a clearer view of objects with motion blur effect around the object, if that makes sense. Not sure how they do this with a dynamic, user controlled motion source, but that's why they're the developers. It makes sense why it would take such a huge toll now, if I'm understanding it correctly.

I suspected as much. This is a very draw call heavy way to do motion blur because its applied per object. Its more accurate than screen based motion blur. Its basically impossible to do this with DX with that many objects by calling on them all individually, you would need to batch them together. That would actually perform better in all cases (but be less accurate).

I think this is a classic example of the sort of problem you get comparing the two APIs. You would never choose this algorithm for DX, it makes no sense to spend 5k draw calls just motion blurring 5k objects when the impact of the effect is going to be minimally better than a much cheaper effect. The GPU can do it but the CPU is going to struggle. Its both a good reason for Mantle but also a bad choice of algorithm for a benchmark, because its obviously not something you would do on DX at all.

 

[Gloomy]

So Mantle does enable new stuff? Good to know.

 

[moonbogg]

So Mantle does enable new stuff? Good to know.

I'll wait to see a straight comparison before passing judgment. Motion blur is the reason my 120hz monitor becomes useless, which is why its not useless because I don't use motion blur. If I wanted motion blur, I'd just do it right and buy a 60hz IPS monitor.

 

[DaZeeMan]

Mantle enables next-generation non-blurry motion blur technology!

:biggrin:

 

[Gloomy]

Sorry, "its obviously not something you would do on DX at all" sounds like an excuse to me even if it's motion blur I would not use.

 

[ASM-coder]

because its obviously not something you would do on DX at all.

But that was one of the arguments for Mantle. It allows game developers
freedom to do things that couldn't be done before.
And I know the counter argument... Just because you can, doesn't mean you should.
But certainly game developers should make those decisions. Not the API.

Conformity is about the only thing that DirectX can do better, that's clearly apparent.

 

[Paul98]

It's too bad they didn't show the frame to frame time, batch count, and other information shown at the end of the video while DX was active.

 

[desura]

So...Mantle looks like it might be a big deal?

http://www.engadget.com/2014/01/14/oxide-star-swarm-real-time-strategy-mantle-demo/

Was reading this earlier.

There's something like 300% frame rate increase. in there...

Seems similar to what SteamOS is trying to do...like, they're trying to go straight to the video card, and are reporting decent increases in framerate.

So we can finally have big expansive worlds in gaming instead of tiny ornate rooms which have pretty much been what's been going on this gen. Just a bunch of tiny really detailed rooms, pretty but not vast.

 

[MutantGith]

How is running a motion blur technique that floods the system with predictive calls to render individual objects along a path any different than arbitrarily telling all of the terrain in a flight sim to render 1x1 pixel tesselation? Alternately, compare the motion blur to rendering full, high level tesselation on moving water layers, that only exist under the ground layer of the city you're playing an FPS in.

It seems like both are specific alterations to a code designed to highlight the capacity of a particular rendering system, be that hardware or software. Neither are applicable or practical in most cases. Yet it seems like people might be applying selective logic and calling one ground breaking and the other dirty pool, depending on their intended outcome.

This is an interesting technique, but I'm not sure one developer that's already said that they completely started writing their own engine from scratch to work only with highly threaded 64bit systems is necessarily concerned with worrying about how practical their engine techniques are to the majority of systems currently out there. The value of this technique might be in figuring out a rendering method that does a good job or approximating it, rather than brute forcing through the calculations.

 

[bystander36]

So...Mantle looks like it might be a big deal?

http://www.engadget.com/2014/01/14/oxide-star-swarm-real-time-strategy-mantle-demo/

Was reading this earlier.

There's something like 300% frame rate increase. in there...

Seems similar to what SteamOS is trying to do...like, they're trying to go straight to the video card, and are reporting decent increases in framerate.

So we can finally have big expansive worlds in gaming instead of tiny ornate rooms which have pretty much been what's been going on this gen. Just a bunch of tiny really detailed rooms, pretty but not vast.

I thought SteamOS just a Linux OS that uses OpenGL, which is pretty similar to DX.

 

[VulgarDisplay]

I think another thing to remember is that this is with a 6 core 12 thread CPU which could be why DX is able to push as many objects as it is in this demo. They have really pushed DX multithreading according to all their videos on the engine to a place it hasn't gone before.

An interesting thing would be if mantle would allow a dual or quad core to keep similar levels of performance to the DX version of the benchmark. Even with these demos we really need more information to draw decent conclusions still.

After another watch, the motion blur really does look damn good in the demo. It's not something I like in mutliplayer FPS games, but the effects on the lasers of the motion blur looks great. In the DX with no motion blur they just look like colored rectangles flying by.

 

[Erenhardt]

Interesting...
Tablet/mobile technology after all!
I imagine, that high-end desktop users will find many new reasons to justify buying expensive server chips just for the sake of having the best. But in the mobile world - could be game changer.

GCN SOC could see amazing uplift when under mantle load, compared to competition running on DX. Chances for competition (intel/arm) to run mantle are slim. That would mean they put GCN in their SOCs, or mantle gets all architectures support, but that would make it another DX.

 

[ASM-coder]

Mantle is not hardwired to GCN, but does require a minimum feature level from the hardware.

 

[BrightCandle]

Mantle is not hardwired to GCN, but does require a minimum feature level from the hardware.

Yes the GCN feature list, something not found on anything other than GCN hardware.