AMD Mantle API [updating]

There is a balance to be had between sensible GPU and CPU pairings, but with performances like mantle can show the onus of the spend can certainly shift from CPU to GPU a little more. Having said that, there will be a lot of systems out that that pairs FX's/i5's with GPU's up the higher end of the stack that will see great performance and playability gains. Reviewers sticking to their pathological high end performance rigs will not represent what the "average" user will gain from this.
 
Battlefield-4-Mantle-API-Benchmark.png

when CPU is an overclocked to 4.5 GHz Core i7-4770K is used, which renders on Radeon R9 290X in Uber mode . In the game we activated the resolution of 1,920 × 1,080 with maximum details, 4xMSAA, FXAA and 16x anisotropic filtering.

http://translate.google.it/translat...-mit-mantle-erster-eigener-benchmark/&act=url


original link: http://www.computerbase.de/news/2014-02/battlefield-4-mit-mantle-erster-eigener-benchmark/
 
Congratulations!

Well, well. Impressive after all! Love to see that i3 being pushed up near i7 levels!!!
Congratulations! Good job!

Now, to be fair, when AMD Product Manager Devon Nekechuck said this:

http://www.maximumpc.com/amd_r9_290x_will_be_much_faster_titan_battlefield_4

.. he should have just shut up, since that did not happen, was not supposed to happen as it seems, and only hyped up things to an unnecessary extent.

Nevertheless, this is great news and lets hope more developers (as well as IHVs) come on board!
 
I found it's hard to believe that with i7-4770K and i5-4670K in DX, the differences between GTX 780 and R9 290X are that great. I know it's multiplayer so things are hard to control, but something isn't right, the condition must differ greatly.
Yeah that's a little weird, in those pclab.pl benchmarks in DX 11.2 the 780 seems to stomp the 290x, which I haven't seen in benchmarks before.
 
No but, at the least, the BF4 patch should have been launched only after the driver was released. That would have avoided all this mess.


Anyway, the driver 14.1 was initially released on the same day, they have discover some late bugs and the decision have been taken to "remove it " at last minutes for fix it.. Dice have decided to release the patch anyway ( because there is not only Mantle on this upate but some fix for other problems too and they was not see any reason to decompile the patch for remove the Mantle update ).. peoples so will be too able to update their games with the patch they was waiting about.. AMD have decided in the same time to first send driver to reviewers like that they could have the week-end for work on the article ( this is not a bad idea ).. monday we should have the 14.2 beta driver ..
 
Last edited by a moderator:
Anyway, the driver 14.1 was initially released on the same day, they have discover some late bugs and the decision have been taken to "remove it " at last minutes for fix it.. Dice have decided to release the patch anyway ( because there is not only Mantle on this upate but some fix for other problems too and they was not see any reason to decompile the patch for remove the Mantle update ).. peoples so will be too able to update their games with the patch they was waiting about.. AMD have decided in the same time to first send driver to reviewers like that they could have the week-end for work on the article ( this is not a bad idea ).. monday we should have the 14.2 beta driver ..

I am not sure that the initial driver was released at all.
But still, would such an apparent critical bug not be detected in advance, if a good quality control was performed?
Not that the other side (nVIDIA) hasnt screwed up either before, probably even more (damaging cards), but its AMD I was talking about.

Anyway, it looks like Mantle itself is a good thing after finally having been launched, Pointless to keep that discussing that now. I just hope AMD takes lessons from hiccups like this.

Off-topic: thanks for the removal of the personal attack.
 
Last edited by a moderator:
I am not sure that the initial driver was released at all.
But still, would such an apparent critical bug not be detected in advance, if a good quality control was performed?
Not that the other side (nVIDIA) hasnt screwed up either before, probably even more (damaging cards), but its AMD I was talking about.

Anyway, it looks like Mantle itself is a good thing after finally having been launched, Pointless to keep that discussing that now. I just hope AMD takes lessons from hiccups like this.

Off-topic: thanks for the removal of the personal attack.

It have been introduced during some minutes... Anyway press and reviewers do their test with the 14.1 ( not the 14.2 as i was initially think ) ( the initial 14.1 have been issued to the press before the release of the patch, it depend of the reviewers to use one or not ( something i like, this way you know what is doing the reviewers ( some use the 14.1 other use the 14.2 ) even Guru3D Hilbert H. have post the issue list AMD have given to him with the driver ( why HH post this, dont ask me, as only press have got this driver ( )..

Outside performance gains, where im the most surprised is by the way you absolutely dont see any difference on quality of the graphism, different API, same graphics.. Ok, AMD/ATI is on the base of nearly all features used on DirectX. But this is absolutely more visible when you see Mantle running.

Personally i use an I7 4930k and play on 2560x1440p so cpu bottleneck wise im not on the list, but its interesting too with cfx ( i have 2x 7970 ) for seen than in cfx where you hit cpu bound situation, the impact is clear.
 
Last edited by a moderator:
Yeah, you sure told them with your handful of internet forum posts :LOL:

Of course I am not talking about what I wrote, but from their own experience with this. I am sure who is in management is not happy with how things developed (or should not be). An organisation that learns with its shortcomings is a stronger organisation.
 
Outside performance gains, where im the most surprised is by the way you absolutely dont see any difference on quality of the graphism, different API, same graphics.. Ok, AMD/ATI is on the base of nearly all features used on DirectX. But this is absolutely more visible when you see Mantle running.

I am not sure that in most of the graphic specific aspects Mantle differs that much from DirectX. I remember reading somewhere that it could make use of DX HLSL as well. That is also why I found AMD answer about "Mantle complementing DirectX", rather than competing with it, interesting.

Further, it didn't look like DICE took that much time to adapt BF4 to Mantle. Granted that Mantle work started 2 years ago according to AMD, but I would guess most of that time the ball was mostly on AMD side, albeit with inputs from DICE. So maybe its not like DICE had to build Mantle codepath from totally zero. Maybe they could take some bits from DirectX as well.
 
So according to AMD's own measurements, something like +10% is expected at maximum settings with good CPUs. I would like to think there is even more than that. Either some Mantle functions are yet to be exposed, or Mantle's implementation in the FrostBite engine is not as deeply rooted as advertised, which is quite logical considering the current version is built upon version 2(would like to hear Repi's thoughts on this).

The problem as I see it, 99% of games now are GPU limited at max settings, I have never seen a game max out 4 physical cores .. not to mention the logical ones .. the situation will change with next-gen only games for sure. and I hope AMD will be ready there.

Yeah that's a little weird, in those pclab.pl benchmarks in DX 11.2 the 780 seems to stomp the 290x, which I haven't seen in benchmarks before.
The last patch added some important optimizations for NVIDIA GPUs. In fact it added even more vital CPU optimizations. Now the game actually uses logical (HT) cores, CPU usage is spread across all threads now. which makes me wonder, why release all of that in conjunction with the Mantle patch? why not from the start?!
 
Last edited by a moderator:
So according to AMD's own measurements, something like +10% is expected at maximum settings with good CPUs.(at least in SP) I would like to think there is even more than that. Either some Mantle functions are yet to be exposed, or Mantle's implementation in the FrostBite engine is not as deeply rooted as advertised, which is quite logical considering the current version is built upon version 2(would like to hear Repi's thoughts on this).

The problem as I see it, 99% of games now are GPU limited at max settings, I have never seen a game max out 4 physical cores .. not to mention the logical ones .. the situation will change with next-gen only games for sure. and I hope AMD will be ready there.



The last patch added some important optimizations for NVIDIA GPUs. In fact it added even more vital CPU optimizations. Now the game actually uses logical (HT) cores, CPU usage is spread across all threads now. which makes me wonder, why release all of that in conjunction with the Mantle patch? why not from the start?!


. ofc most games ask finally more to the gpu than for the rest because we have headed thoses last year to transfer many computing ressouces to the shaders of the gpu ( instead of keep them locked to the cpu )... i use SLI and CFX since they exist ( H2o and high OC ( 1300mhz on my 7970's ) and i run all my cpu's extremely overclocked 24/24 for this simple reason.. i allways hit the cpu limitations on a time or another. for feed thoses gpu's, i need increase the cpu performance even at 2560x1440p.....

But to a certain extend, many games are cpu bound due to the limitations of the code, who dont use properly the cpu ressources available... ( proper multithreading in nearly all cases )
 
Last edited by a moderator:
Even if that specific site shows 290x on Mantle faster than GTX780Ti (numbers are all over the place, on multiplayer especially), that is far from "ridiculing" the opponent.

Yes, it faster than 780Ti, but Devon Nekechuck said about the Titan, and not about 780Ti.
 
Yes, it faster than 780Ti, but Devon Nekechuck said about the Titan, and not about 780Ti.

1) GTX780Ti is not THAT faster than a Titan;
2) Titan was not a full GK110 chip anyway, so it was bound to be a faster SKU of it than Titan, probably even before Mantle was released, as it happened.
 
The problem as I see it, 99% of games now are GPU limited at max settings, I have never seen a game max out 4 physical cores .. not to mention the logical ones
That's the point - currently the workloads are not being distributed across those threads effectively. Guess what's happening with Mantle - CPU utilisation is shooting I
up. You'll be surprised how many "GPU killing" games are still very CPU bound.

Oh for the days of the old B3D fillrate graphs, it used to highlight this very well.
1) GTX780Ti is not THAT faster than a Titan;
2) Titan was not a full GK110 chip anyway, so it was bound to be a faster SKU of it than Titan, probably even before Mantle was released, as it happened.
What bearing does that have? The quote was taken prior to Ti's release / announcement.
 
Back
Top