Mantle Feedback/Bugs

[pgi947]

^ What most of us have been saying.

Why didn't they add any aa? Surely full ultra is playable at 1080 on a single 290x

 

[shankly1985]

In case you missed it...

2nd Image confirms my finding the game is so much smoother under Mantle.

 

[SiDeards73]

do we know what CPU was used and what clocks etc?

 

[LtMatt]

Why didn't they add any aa? Surely full ultra is playable at 1080 on a single 290x

Don't know, but it won't change things at all. They did run some 1600P tests though where they used AA.

Zavod 311 - 2560x1600 2X AA - Ultra Settings

With the AA reduced down to 2X AA, which is at playable levels, the performance advantage with Mantle does go up more than the very GPU dependent 4X AA setting. Here AMD Mantle is almost 14% faster than D3D 11.1.

Under D3D this map is playable at 2560x1600 2X AA, but just barely. When we play under the AMD Mantle API the performance improves enough for us to notice it, and is more than playable. Therefore, AMD Mantle has helped improve upon the gameplay experience by giving us a super smooth at all times 2560x1600 2X AA gameplay experience.

In terms of frame times, AMD Mantle is clearly superior with more consistent, less "spiky" frames, and overall better numbers.

Siege of Shanghai - 2560x1600 4X AA - Ultra Settings

Under this GPU limited test, the results turn out to be not as extreme as our CPU limited 1080p with no AA test. Here Mantle is improving on performance, but not by a whole lot. AMD Mantle is providing a 7% performance improvement. It's not massive, but a free performance improvement is always welcomed.

At this setting, the game isn't exactly at the level of performance we'd deem playable on the R9 290X. We would want to maintain a consistent 60 FPS average to be playable. Under D3D performance is at 52 FPS, under Mantle it rises to 55-56 FPS. However, even that isn't quite enough to call these settings playable. Therefore, AMD Mantle did not improve the gameplay experience at this GPU dependent resolution and AA setting in BF4.
When it comes to the frame times, we found that once you start going toward GPU dependent settings the variance of D3D isn't as bad. However, AMD Mantle is still more consistent, and overall produces lower frame times, which is better.

Is AMD Mantle everything it is cracked up to be? The answer is potentially yes. As we stated, you first need to understand the goals of AMD Mantle, and the kind of benefits it can provide. Do not expect GPU dependent performance to get significantly better. Rather, if you are on a more mid-range system this should help more than someone who has the latest bleeding edge hardware overclocked to madness. At least, that is the trend we see so far. When we test lower-end video cards, it will really tell us just what kind of advantages Mantle has in store.

Mantle could potentially, drastically change the gaming world if more games used it. Lower-end computers, especially notebooks, and ultrabooks could see significant gaming performance increases. AMD isn't done with Mantle, this is just a first taste of it. We have to say, this first taste is sweet. We also experienced no issues, no crashing, it was rock stable for all the pounding we gave it the whole day. We look forward to future improvements from DICE and AMD, and hopefully more games that will implement Mantle. Remember, we have more evaluation coming, so stay tuned.

 

[dave_beast]

Not going to post benchmarks, but ran the Star Swarm benchmark under DX and Mantle and the difference was night and day on my 1150mhz 7970.

Good work AMD

 

[Besty]

Never mind, at least it sets them up nicely on the Steambox/IGP/Laptop front. Its not all doom and gloom, its just that their sales pitch was all wrong, especially on the 290x+Mantle= Titan Killer comments.

My beef is that AMD did not make it clear from the outset that GPU performance would not be boosted much by this tech, just CPU performance.

This still looks like a subtle bait and switch to me on the messaging they were giving at their presentations and what not.

 

[shankly1985]

Never mind, at least it sets them up nicely on the Steambox/IGP/Laptop front. Its not all doom and gloom, its just that their sales pitch was all wrong, especially on the 290x+Mantle= Titan Killer comments.

My beef is that AMD did not make it clear from the outset that GPU performance would not be boosted much by this tech, just CPU performance.

This still looks like a subtle bait and switch to me on the messaging they were giving at their presentations and what not.

My GPU's have boosted from it and my CPU is doing less??

 

[Martini1991]

My girlfriends dad just ran mantle on his 7850 and his fx6300 after I pimped it out, it's so broken for him he has to go back to direct x.

 

[Rroff]

Never mind, at least it sets them up nicely on the Steambox/IGP/Laptop front. Its not all doom and gloom, its just that their sales pitch was all wrong, especially on the 290x+Mantle= Titan Killer comments.

My beef is that AMD did not make it clear from the outset that GPU performance would not be boosted much by this tech, just CPU performance.

This still looks like a subtle bait and switch to me on the messaging they were giving at their presentations and what not.

Aside from the fact I have an issue with the whole way its marketted if they don't deliver most people don't really realise where the real win is with Mantle bit of a silly example but one of the scenarios faced with designers with DX is that the overhead would mean say it took 3ms* to render a simple bush but also 3ms to render a high polygon character so a designer could be there thinking do I flesh out the scene with lots of vegetation or do I skimp on the vegetation so I can add the extra character in the scene I wanted, but then it doesn't look as "alive" with the barer background.

With Mantle those simple bushes could be knocked down to say 1ms without the API overhead - letting you get both things you wanted in the scene without the performance drop you'd normally get with DX. In the short term this could produce performance gains in some games as while those high polygon characters are still taking say 3ms of your time spent rendering those bushes now take a lot less time but in the long run it lets designers produce games that are closer to their original vision without so many compromises for performance concerns.


* Would actually be a lot less than 3ms but those numbers are easier for illustration.

 

[humbug]

I ran some monitoring to see if I could find anything interesting in my BF4 + Mantle stutter then crash problems.

Below is a recording of my System and GPU RAM during Mantle game play.

The GPU vRam is perfectly stable all the way through, the System RAM built up to about 2GB and the Dropped about 800MB (40%) in 30 Seconds. at that point the Game Crashed.


Time,, .......RAM,, ........vRAM
15:02:57, ,2066.000 ,1779.383
15:02:58, ,2066.000 ,1779.383
15:02:59, ,2067.000 ,1779.383
15:03:00, ,2066.000 ,1779.383
15:03:01, ,2070.000 ,1779.383
15:03:04, ,2066.000 ,1779.383
15:03:05, ,1960.000 ,1779.383
15:03:06, ,1938.000 ,1779.383
15:03:07, ,1929.000 ,1779.383
15:03:08, ,1928.000 ,1779.383
15:03:09, ,1858.000 ,1779.383
15:03:10, ,1652.000 ,1779.383..... System RAM leak
15:03:11, ,1608.000 ,1779.383
15:03:12, ,1596.000 ,1779.383
15:03:13, ,1592.000 ,1779.383
15:03:14, ,1592.000 ,1779.383
15:03:15, ,1592.000 ,1779.383
15:03:16, ,1593.000 ,1779.383
15:03:17, ,1592.000 ,1779.383
15:03:18, ,1592.000 ,1779.383
15:03:19, ,1536.000 ,1779.383
15:03:20, ,1327.000 ,1779.383
15:03:21, ,1266.000 ,1779.383
15:03:22, ,1248.000 ,1779.383
15:03:23, ,1244.000 ,1779.383
15:03:24, ,1243.000 ,1779.383
15:03:25, ,1242.000 ,1779.383
15:03:26, ,1241.000 ,1779.383
15:03:27, ,1240.000 ,1779.383
15:03:28, ,1240.000 ,1779.383
15:03:29, ,1241.000 ,1779.383
15:03:30, ,1242.000 ,1779.383
15:03:31, ,1242.000 ,1779.383
15:03:32, ,1242.000 ,1779.383
15:03:33, ,1242.000 ,1779.383
15:03:34, ,1242.000 ,1760.223
15:03:35, ,1229.000 ,1761.234
15:03:36, ,1223.000 ,1764.949
15:03:37, ,1221.000 ,1766.129
15:03:38, ,1220.000 ,1766.129
15:03:39, ,1220.000 ,1766.129
15:03:40, ,1220.000 ,1766.125
15:03:41, ,1220.000 ,1766.125
15:03:42, ,1219.000 ,1766.145
15:03:43, ,1220.000 ,1766.125
15:03:44, ,1220.000 ,1766.125
15:03:45, ,1220.000 ,1766.125
15:03:46, ,1219.000 ,1766.125
15:03:47, ,1219.000 ,72.059 ......Game Crashed

 

[BuzzLloydyear]

I done some tests as specified by Matt in the Mantle Performance thread. With my 7990 (which as far as supported cards go, is probably the worst for Mantle right now). The performance of Mantle is actually slower in BF4 than Directx.

I am sure when they do the optimise and sort out any issues with crossfire, things will be different, but here are my results from 2 different resolutions.

2560x1440 @ Ultra

Mantle:

Min: 11.09
Max: 78.62
Avg: 58.34

Directx:

Min: 8.81
Max: 115.21
Avg: 75.77

1920x1080 @ Ultra

Mantle:

Min: 35.84
Max: 123.46
Avg: 80.24

Directx:

Min: 42.02
Max: 127.55
Avg: 82.75

Higher FPS is in green, lower FPS is in red.

 

[SimBy]

My beef is that AMD did not make it clear from the outset that GPU performance would not be boosted much by this tech, just CPU performance.

This still looks like a subtle bait and switch to me on the messaging they were giving at their presentations and what not.

One can just facepalm at arguments like that. And I'm reading it all over the place by green trolls GPU performance IS what was boosted. By removing CPU overhead. Simple as that. And those are just the low hanging fruits. More gains will most likely come when devs actually do some real lower level optimizing.

 

[breadturbo]

I've had to go back to 13.12 also, the new driver caused civ5 to ctd after 5mins

 

[humbug]

My beef is that AMD did not make it clear from the outset that GPU performance would not be boosted much by this tech, just CPU performance.

This still looks like a subtle bait and switch to me on the messaging they were giving at their presentations and what not.

One can just facepalm at arguments like that. And I'm reading it all over the place by green trolls GPU performance IS what was boosted. By removing CPU overhead. Simple as that. And those are just the low hanging fruits. More gains will most likely come when devs actually do some real lower level optimizing.

Exactly, AMD made it clear Performance would be boosted by removing the CPU Bottleneck.

What are you talking about, Besty?

 

[r7slayer]

Well I got over 30% boost in FPS and im pretty sure that's not just from removing the CPU overhead. So yea it has given my 290x a performance boost also. Anyway my only problem with Mantle/BF4 atm is ever since I updated to the beta drivers my game crashes when It tries to load a map/level or sometimes throws a DirectX error even though im using mantle lol. Restarted the pc a couple of times then it seemed to fix its self but apart from that im happy so far!

 

[AllBodies]

Never mind, at least it sets them up nicely on the Steambox/IGP/Laptop front. Its not all doom and gloom, its just that their sales pitch was all wrong, especially on the 290x+Mantle= Titan Killer comments.

You realise that the R9 290X is faster than a Titan even without mantle right?

And as others have explained, I don't believe AMD ever misrepresented what mantle was for, I think lazy readers and/or media misinterpreted it (like what happens with scientific articles being sensationalised all the time).

Mantle increases graphics rendering performance (i.e. frames-per-second) in CPU-limited scenarios - that's it. And that's what they said would be the case.

So what this essentially boils down to in reality is that it's of benefit to:

• 
  Multiplayer games (lots of players taxes the CPU)

• 
  People with weak CPU's but good graphics cards (e.g. i3 4130 with an R9 280X)

• 
  Anyone with Xfire

• 
  Likely other games with lots of units/AI, like RTS games

• 
  But likely NOT single-player games with not many units, like Tomb Raider, or COD, or Dark Souls, that sort of thing.

 

[Ian Evey]

I done some tests as specified by Matt in the Mantle Performance thread. With my 7990 (which as far as supported cards go, is probably the worst for Mantle right now). The performance of Mantle is actually slower in BF4 than Directx.

I am sure when they do the optimise and sort out any issues with crossfire, things will be different, but here are my results from 2 different resolutions.

2560x1440 @ Ultra

Mantle:

Min: 11.09
Max: 78.62
Avg: 58.34

Directx:

Min: 8.81
Max: 115.21
Avg: 75.77

1920x1080 @ Ultra

Mantle:

Min: 35.84
Max: 123.46
Avg: 80.24

Directx:

Min: 42.02
Max: 127.55
Avg: 82.75

Higher FPS is in green, lower FPS is in red.

Crossfire probably is not working at all using the mantle api, so effectively you could be using the equivalent of a 7970 using mantle and 2 7970 using dx, If so then you will be in for a big gain when they sort it out for the 7990.

 

[drunkenmaster]

You realise that the R9 290X is faster than a Titan even without mantle right?

And as others have explained, I don't believe AMD ever misrepresented what mantle was for, I think lazy readers and/or media misinterpreted it (like what happens with scientific articles being sensationalised all the time).

Mantle increases graphics rendering performance (i.e. frames-per-second) in CPU-limited scenarios - that's it. And that's what they said would be the case.

So what this essentially boils down to in reality is that it's of benefit to:

• 
  Multiplayer games (lots of players taxes the CPU)

• 
  People with weak CPU's but good graphics cards (e.g. i3 4130 with an R9 280X)

• 
  Anyone with Xfire

• 
  Likely other games with lots of units/AI, like RTS games

• 
  But likely NOT single-player games with not many units, like Tomb Raider, or COD, or Dark Souls, that sort of thing.

The thing to note is, single player games can see a boost... just not yet. Why? Because game designers currently try to design to a goal, currently that goal is a very limited number of draw calls. Draw calls govern everything being drawn, not just other players or units in a online game or RTS.

Currently what you could say is single player games are where game dev's set those targets. Balancing is a difficult prospect, how do you make an engine that runs perfectly in multiplayer, you turn down the detail/units/draw calls till the max number of players won't go past said drawcalls, that's fine but what happens when in BF4 case, 64 people aren't on screen? You can do more but you've designed a game that can't do more. What if you base is only on single player, then you get severe slowdown with more players in screen. So it's a balancing act but ultimately the point of the balance is to aim at a certain amount of draw calls/performance.

But if you remove this limit the design targets shift also. It doesn't matter what game it's in. In an RTS that can be more effects, more lighting, more bullets, or more units, anything you want. In a single player it can mean more NPC's on screen, more graphical effects, more objects, more destructible items, ability to draw more pieces of broken walls/bodies/anything.

This balancing act is one of the reasons that the high end of draw calls and reducing the overhead is so effective. because some games tend to naturally go over this limit in some circumstances and thus we get a huge performance gain.

But there is nothing preventing game devs just raising this target and aiming for something much more complex/better looking in single player.

The difficulty comes in with how do you design a game with more complexity/higher amount of things being drawn.... when half your audience can't use those settings.

The answer is with difficulty, however the improvement at the high end of the scale currently, huge multiplayer games, RTS's and smaller improvements elsewhere should drive customers to want Mantle, it should hopefully drive Nvidia to support it(or do similar) and once the whole customer base is off DX then every game can be targeted in the same way in the future.

There is nothing fundamentally different about single player games that means lower draw calls, it's just far far easier to control draw calls/generally everything in a single player game, so they rarely have to factor in balancing problems with many more players/objects on screen. We haven't seen these things turned up because they would do what DX does in the Oxide demo, create a stuttering mess of a game.

 

[Besty]

GPU was boosted by removing CPU overhead? How does that boost my performance if my GPU's are already at 98/99% and my CPU is less than 50%? It looks like it is marginal and the benefits get lower the higher the resolution goes, this was not made clear until the last minute.

The original presentation by AMD was in the context of 290/290x, most of us sat through the PR briefing on it, Mantle was not presented in the context of their CPU lineup and they did not make it clear that your CPU needed to be sub-optimal to see the bigger gains.

The way the hype machine rolled on through Roy did indicate that this was a GPU centric technology and not CPU optimisation story.

I am not arguing that the gains are not there to be had but this was marketed for enthusiasts in the GPU space, this is why it is in the graphics section of the OCUK forum. Guess what, it does not make your GPU much quicker if no CPU bottleneck. They should have called that out from the start.

Call me cynical but this looks like a really good play by AMD to get their CPU's competitive again at the expense of the Intel enthusiast, not their GPU's.

 

[BuzzLloydyear]

Crossfire probably is not working at all using the mantle api, so effectively you could be using the equivalent of a 7970 using mantle and 2 7970 using dx, If so then you will be in for a big gain when they sort it out for the 7990.

Both GPU's are in use whilst playing, as you can see from the below image (the dip in usage in gpu2 was when I tabbed out to check Afterburner)