The AMD Mantle Thread

[Venomous]

Mantle won't even work on my system, so yes, I wouldn't consider it a fluid launch especially for those with gcn 1.0 hardware. Reminds me of the whole 3DNow hype back in the k6 days.

 

[PPB]

Or maybe it's a bug?

http://forums.anandtech.com/showpost.php?p=36023377&postcount=123

Nope! Repi is paid PR by amd, it is sure amd lowering IQ in their drivers!!!one!eleventy


Moving on...

 

[Grooveriding]

[H]ard's Mantle review

http://hardocp.com/article/2014/02/03/amd_mantle_performance_preview_in_battlefield_4

There are also single-player and higher resolution multiplayer runs in the benches.

 

[VulgarDisplay]

Mantle won't even work on my system, so yes, I wouldn't consider it a fluid launch especially for those with gcn 1.0 hardware. Reminds me of the whole 3DNow hype back in the k6 days.

Why does it not work on your system? Works fine albeit not optimized on my gcn 1.0 setup.

 

[bystander36]

Why does it not work on your system? Works fine albeit not optimized on my gcn 1.0 setup.

You can expect different setups to have varying results. One of the problems with lower level coding is it has a harder time accounting for a wider range of hardware.

I'm sure some of these things will get worked out, but not everyone is having smooth game play with Mantle as of now, but it is beta, so that is expected.

 

[Venomous]

Why does it not work on your system? Works fine albeit not optimized on my gcn 1.0 setup.

Turned out to be my air display adapter in device manager . After disabling it I got the mantle API to run last night. I went back to the d3d version due to the stuttering and lag. I tried enabling lightboost, vsync, etc and nothing helped.


I'm not to keen on having to disable things in device manager just to get an API to run. I'm throwing in the towel on this and moving on. I may look at AMD a generation or two down the road and see how things play out.

 

[Teizo]

[H]ard's Mantle review

http://hardocp.com/article/2014/02/03/amd_mantle_performance_preview_in_battlefield_4

There are also single-player and higher resolution multiplayer runs in the benches.

I haven't bothered to read the review yet, and not sure I will, but did they mention why they did not have AA enabled? A 20 FPS increase is good though.

 

[ASM-coder]

I haven't bothered to read the review yet, and not sure I will, but did they mention why they did not have AA enabled? A 20 FPS increase is good though.

Reading the review yourself is the best way to answer your question.
(and best of all, without any bias)

 

[cmdrdredd]

I haven't bothered to read the review yet, and not sure I will, but did they mention why they did not have AA enabled? A 20 FPS increase is good though.

1920x1080 and no AA is going to be more CPU limited perhaps? It's the situation where Mantle gives the most benefit. Cranking up AA and resolution and you begin to get GPU limited.

 

[Teizo]

Reading the review yourself is the best way to answer your question.
(and best of all, without any bias)

I know, just [H] is not a favorite site of mine...so I prefer not to visit there.

1920x1080 and no AA is going to be more CPU limited perhaps? It's the situation where Mantle gives the most benefit. Cranking up AA and resolution and you begin to get GPU limited.

I see, but I would like to see it tested in all conditions though. Was just curious if there was something off with the AA in the driver.

 

[cmdrdredd]

I know, just [H] is not a favorite site of mine...so I prefer not visit there.



I see, but I would like to see it tested in all conditions though. Was just curious if there was something off with the AA in the driver.

I hear you...I think it's just trying to show the benefit for CPU limited situations. DICE already released numbers where you were not very CPU limited and the difference was very low by comparison. 1 or 2% increase. Still an increase but not as dramatic.

 

[Gloomy]

I see, but I would like to see it tested in all conditions though. Was just curious if there was something off with the AA in the driver.

Mantle gets a 40% hit with 4xAA. The performance lead pretty much disappears.

I don't think there's anything off with the AA, rather that Mantle has a much bigger GPU lead so they get hit harder with the 'real cost' of the AA when there isn't a CPU bottleneck.

 

[Venomous]

Mantle gets a 40% hit with 4xAA. The performance lead pretty much disappears.

I don't think there's anything off with the AA, rather that Mantle has a much bigger GPU lead so they get hit harder with the 'real cost' of the AA when there isn't a CPU bottleneck.

Interesting about AA affecting performance. Maybe AMD is thinking, hey people are starting to move to higher resolution scaling and the odds of them cranking up aa is less likely?

 

[VulgarDisplay]

The AA performance hit is most likely due to the GPU enhancements not even really being touched yet. At least that's what PCPER and AT said in their reviews. I'm sure MSAA performance will increase in a few driver releases/game patches (who knows what needs to happen with this new API.)

 

[Venomous]

The AA performance hit is most likely due to the GPU enhancements not even really being touched yet. At least that's what PCPER and AT said in their reviews. I'm sure MSAA performance will increase in a few driver releases/game patches (who knows what needs to happen with this new API.)

Just a sit and wait game from here. I wonder if mantle updates will be speedy?

 

[Headfoot]

I noticed reduced IQ and periodic crashes in frametime/rate, but also a rather large general performance uplift on a 2500k and 290 running 5314x1050 resolution @ Ultra, no aa. I would drop below 60 fps on a few vegetation heavy maps before (Guilin Peaks, ~45-50 fps). Mantle gave me a solid 20 fps on the high end (from 65 to 85 on Golmud), but the hiccups are completely unbearable. I get them about once a minute. The upside is that the patch improved DX11 performance quite a bit as well. I no longer drop below 60 fps ever with those settings and 1000 mhz clock speed. Ideally, I would be thrilled if I could turn on 2xAA and stay at 60 fps but that may be a bit too optimistic given the GPU limited nature of turning on AA at high resolution.

Additionally, Vsync under Mantle capped me at 30 fps for some reason. I could not figure out why either. Had to turn off vsync altogether to regain any amount over 30. I personally notice screen tearing much more than the lag from vsync so that was a negative for me. At the end of it all, the performance uplift in DX11 was enough to get me to 60 fps without exceptions so I'm a happy camper

 

[DarkKnightDude]

Once the stuttering is gone I can see Mantle being used like crazy. Though I hope there's a way to record gameplay with it.

 

[DamnedLife]

The AA performance hit is most likely due to the GPU enhancements not even really being touched yet. At least that's what PCPER and AT said in their reviews. I'm sure MSAA performance will increase in a few driver releases/game patches (who knows what needs to happen with this new API.)

CPU optimization was the low hanging fruit, and with overhead reduction came real high increases in percentage, GPU optimizations are left in the control of the developers (as per Mantle's working schematics) so if developers wants they can take the time and optimize lets say DirectCompute AA with Mantle, or multi GPU scalability improvements, or rendering improvements, or culling improvements etc. It is up to developers to take to time and add these in the first place to begin with and then again taking time to optimize these things. So for indie developers highly unlikely but CPU optimizations are still there for them. That's why in the initial phase, we see only CPU optimizations heavy release, second phase will come later on which will widen the performance gap even further. AAA game titles will most likely include them for future releases. Like Mass Effect or Star Wars Battlefront etc.

 

[PPB]

I noticed reduced IQ and periodic crashes in frametime/rate, but also a rather large general performance uplift on a 2500k and 290 running 5314x1050 resolution @ Ultra, no aa. I would drop below 60 fps on a few vegetation heavy maps before (Guilin Peaks, ~45-50 fps). Mantle gave me a solid 20 fps on the high end (from 65 to 85 on Golmud), but the hiccups are completely unbearable. I get them about once a minute. The upside is that the patch improved DX11 performance quite a bit as well. I no longer drop below 60 fps ever with those settings and 1000 mhz clock speed. Ideally, I would be thrilled if I could turn on 2xAA and stay at 60 fps but that may be a bit too optimistic given the GPU limited nature of turning on AA at high resolution.

Additionally, Vsync under Mantle capped me at 30 fps for some reason. I could not figure out why either. Had to turn off vsync altogether to regain any amount over 30. I personally notice screen tearing much more than the lag from vsync so that was a negative for me. At the end of it all, the performance uplift in DX11 was enough to get me to 60 fps without exceptions so I'm a happy camper

When you go below a threshold of fps with vsync on with mantle (≈-10% of your refresh rate as fps) it will go to 30fps ,or 45 if triplebuffering is on (this last one is an assumption as I couldnt test it on this game). On dx it would show you a different fps number, but in reality you would be also be running at 30fps too (thats why on DX when going below your refresh rate as fps with vsync on looks horrid, even if fps number isnt that bad).

What bothers me that vsync on this game turns into a loss of performance, I PM'ed repi about this, hope he can look into it, as reaching constant 60+fps has never been easier with mantle. It's such a waste I cant turn it on just because I lose from 10 to 20 fps.

 

[bystander36]

When you go below a threshold of fps with vsync on with mantle (≈-10% of your refresh rate as fps) it will go to 30fps ,or 45 if triplebuffering is on (this last one is an assumption as I couldnt test it on this game). On dx it would show you a different fps number, but in reality you would be also be running at 30fps too (thats why on DX when going below your refresh rate as fps with vsync on looks horrid, even if fps number isnt that bad).

What bothers me that vsync on this game turns into a loss of performance, I PM'ed repi about this, hope he can look into it, as reaching constant 60+fps has never been easier with mantle. It's such a waste I cant turn it on just because I lose from 10 to 20 fps.

Triple buffering should allow for complete variable FPS, similar to not using V-sync. However, the stuttering caused from only being able to display frames of 16.7ms and 33.3ms, remains. Not using triple buffering will often result in locked at 30 FPS, unless you are able to get some frames rendered faster than 16ms.

I'm curious as to how Mantle would handle hitting your refresh rate with triple buffering and V-sync. Is it going to take an OpenGl approach or DirectX one. OpenGL has less latency in that situation.

 

[Deders]

When you go below a threshold of fps with vsync on with mantle (≈-10% of your refresh rate as fps) it will go to 30fps ,or 45 if triplebuffering is on (this last one is an assumption as I couldnt test it on this game). On dx it would show you a different fps number, but in reality you would be also be running at 30fps too (thats why on DX when going below your refresh rate as fps with vsync on looks horrid, even if fps number isnt that bad).

If Triple buffering is implemented properly you should get smooth fps all the way from 0 to 60 (or whatever the refresh rate of your monitor is).

I have to use D3DOverrider to get it working properly. When I measure the frametimw with D3DO it adjusts proportionally to the framerate, so it's not locked at anyting below the refresh rate like you described.

 

[3DVagabond]

Just a sit and wait game from here. I wonder if mantle updates will be speedy?

If I understand Mantle correctly, it won't be as much Mantle or driver updates that are needed as game updates.

It's been said by both Dice and Oxide that Mantle took 2 man months to incorporate into the game engines. They've never said it takes 2 man months to design a game using it. Not as far as I recall, anyway.

I think just like with consoles in the past the more time devs have to program and optimize code for specific hardware the better the performance gets. I think that there is going to be a similar learning curve with Mantle. What gives Mantle it's opportunity though is that learning curve should be able to be applied closely with the console's because most of the same optimizations should work with Mantle. Without the consoles using GCN I really doubt Mantle would have a lot of hope to be successful. The consoles using GCN provides opportunities to devs across both console platforms and PC's using the same GCN architecture.

Now, before anyone takes that and twists and spins it, I did not say that Mantle is used on consoles. I'm not saying that at all! What I'm saying is that the fact the consoles and Radeon graphics are both GCN means that many of the low level optimizations should be applicable to both.

 

[Dresdenboy]

Mantle gets a 40% hit with 4xAA. The performance lead pretty much disappears.

Using which CPU? Of course this is a matter of balancing the whole system. Even using Win 7 instead of 8.1 can change the picture in a similar way. In a cost neutral way one could save money on the CPU if 4xAA already saturates the GPU with the more expensive CPU.

So far CPUs < $200 (+ a $300 FX as accepted exception ) paired with a R9 290X benefit the most, even Pentium, i3 and i5 variants. An i3-4330 moves into i7-4770K territory within 5%:

Source: http://www.betazeta.com/#!/necronomicon/chw/post/amd-cumple-las-expectativas-con-mantle

BTW, how much more does the GTX 780 GHz cost? Looks like $550+ vs. ~$450 for the standard edition. I don't see many listings here.

 

[DamnedLife]

780 GHz didn't made it to retail SKUs AFAIK, just one Gigabyte 780 GHz model was retailed but then it halted production so no more sold, if you got one of those rare cards then it is like getting a car that is limited edition with numbers inscribed at the chasis not available anymore...

 

[3DVagabond]

Using which CPU? Of course this is a matter of balancing the whole system. Even using Win 7 instead of 8.1 can change the picture in a similar way. In a cost neutral way one could save money on the CPU if 4xAA already saturates the GPU with the more expensive CPU.

So far CPUs < $200 (+ a $300 FX as accepted exception ) paired with a R9 290X benefit the most, even Pentium, i3 and i5 variants. An i3-4330 moves into i7-4770K territory within 5%:

Source: http://www.betazeta.com/#!/necronomicon/chw/post/amd-cumple-las-expectativas-con-mantle

BTW, how much more does the GTX 780 GHz cost? Looks like $550+ vs. ~$450 for the standard edition. I don't see many listings here.

That's a strange comparison with that chart. It's shown as % vs the 4770K. It's showing the performance of the 780, for example versus itself on a 4770K, not versus the 290X in DX and under Mantle.

It is interesting that the 780 is less CPU bound on slower processors than the 290X in DX to a point. Below a certain point, the FX6350 in this comparison, the 290X is less CPU bound than the 780. I'm assuming it's showing the few % improvement nVidia gets due to DX multi threading and that doesn't really help with the APU and the dual core chips. Or maybe a lack of DX optimizing by AMD recently because of Mantle? Where it would be the opposite for nVidia, they would be trying really hard to optimize their DX drivers on BF4 to limit Mantle's advantage? A bit of both? Am I misinterpreting it altogether?

Also the 4330/4340 maintain a very high percentage of the 4770K's performance using Mantle.