AMD Mantle API [updating]

Which confirms my assumption that not all cryengine games will support mantle? Or is this a case of Ryse being on am older code base?
 
kukreknecmi said:
So Mantle support/can execute hand written AMD_IL code? I'm not %100 sure but we cant use AMD_IL on Windows for Dx, right? From my understanding if Mantle allows directly to use AMD_IL, it may mean allowing a bit more flexibility.

On Rendering BF4 with Mantle presentation it was said that ""Could write AMDIL directly "", so it may also accept AMD_IL as code input?

Side Question : Is there any way to code GCN ISA directly on Windows, without need an external Assembler? Somehow Mantle/AMD will allow that?
Click to expand...

Yes Mantle accepts straight AMDIL, and correct you can't do that on DX11
 
Those are some really nice gains and a nice writeup too. The numbers will probably look even more impressive on AMD CPUs. Hopefully a preview of great things to come in the DirectX 12 era.
 
dice + BF4 and Mantle = still unplayable using Mantle for weeks...!!
dice sucks as a developer is all I can take note off.
 
Well, in the end Civilization Beyond Earth didn't officially support any NVIDIA features, however the support for HBAO+ was added via NVIDIA driver hooks and can be forced in. There are many conclusions to be drawn from this, especially that HBAO+ can't be forced in any game, it needs at least some level of developer integration.

With Ryse/Alien Isolation/Lichdom not supporting Mantle, and Call Of Duty Advanced Warefare officially featuring NVIDIA features such as HBAO+, it looks like the previous leak about upcoming Mantle games was fake.
http://videocardz.com/51018/exclusive-upcoming-games-support-mantle
 
None of those games was in the "Confirmed Mantle support" list, they were mentioned in the speculative part.
 
Two multi-platform titles are showing -what I think could be- signs of less efficient AMD DX11 drivers, ACU and COD AW.

Ubi shifted the blame on AMD for bad ACU performance on their GPUs, AMD fired back with a statement stating the game is overloading DX11 with draw calls.

"The game (in its current state) is issuing approximately 50,000 draw calls on the DirectX 11 API. Problem is, DX11 is only equipped to handle ~10,000 peak draw calls. What happens after that is a severe bottleneck with most draw calls culled or incorrectly rendered, resulting in texture/NPCs popping all over the place. On the other hand, consoles have to-the-metal access and almost non-existent API Overhead but significantly underpowered hardware which is not able to cope with the stress of the multitude of polygons. Simply put, its a very very bad port for the PC Platform and an unoptimized (some would even go as far as saying, unfinished) title on the consoles".

http://www.tweaktown.com/news/41110...n-s-creed-unity-s-poor-performance/index.html

However NV is handling the same load much better than AMD, (and thus providing much higher performance), likely owing to it's much better DX11 drivers.

DF found that COD AW caused much higher CPU load with AMD cards, than NVIDIA, necessitating the use of more powerful CPUs with AMD GPUs:

Our observation that the use of AMD graphics cards seems to create a much higher CPU load than utilization of Nvidia equivalents. This week, AMD released a new beta driver optimised for Call of Duty. We saw a boost in frame-rates compared to our measurements taken last week, but the underlying problem remains.

If you have a Core i3 - or an alternative offering much the same processing performance - paired with an enthusiast AMD card, the game often hits CPU limits during both cut-scenes and gameplay, resulting in a collapse in frame-rates.

We saw the same results on both our Core i3 gaming PC, and also when we disabled two cores on our i7, downclocking it to simulate an i3's capabilities. In both cases, the Nvidia cards' performance barely saw any difference at all. Basically, if you're using an AMD card, you'll need a more powerful CPU too - the i5 2500K in the recommended spec, for example.
Click to expand...
http://www.eurogamer.net/articles/digitalfoundry-2014-call-of-duty-advanced-warfare-pc-performance

The same behavior is noticed with ACU too:
Even with the i5 installed, we also found that the Radeon R7 265 and the R9 280 didn't quite deliver the performance of their direct Nvidia competitors - the GTX 750 Ti and the GTX 760.
Click to expand...

http://www.eurogamer.net/articles/digitalfoundry-2014-assassins-creed-unity-face-off

I can't draw solid conclusions from this, but I am inclined to think this can't be a coincidence, though the whole thing is open for discussion.
 
I wonder if this is related to those DX11 improvements that were made a couple of months ago?
It'd be interesting to see if this difference in performance is also present in drivers earlier than that.
 
Actually, COD AW run pretty well on AMD gpu's ... http://www.guru3d.com/articles_page..._graphics_performance_benchmark_review,6.html

As for Assassin Creed, this is a different story.

FC4 have been released and it run "suprisingly" bad on AMD GPU's... i will like to get a real full analysis on both thoses 2 games.. They both look completely unoptimized.

Dont forget, in both cases, thoses are "gamework" titles, and its sure that Nvidia have more control on what is made by the driver in any case, or even the games. Even if in thoses 2 titles, the performance are not really stellar too on Nvidia gpu's. ( FC have less problem than AC this said ).
 
Last edited:
Perhaps the gameworks libraries themselves are using more draw calls than necessary? Wouldn't be the first time something similar to this has happened with Nvidia sponsored titles (flat walls and objects with extreme tesselation that could have done the same with single digit triangles) when doing so would obviously hurt the competition.

Regards,
SB
 
lanek said:
Actually, COD AW run pretty well on AMD gpu's ... http://www.guru3d.com/articles_page..._graphics_performance_benchmark_review,6.html
Click to expand...
It's not a matter of running bad on AMD GPUs, it is a matter of requiring more CPU performance to run well. Guru3d benchmarks are done using high end CPUs, so the problem is not apparent, but switch to a Core i5 or something similar and AMD GPUs stall.

Silent_Buddha said:
Perhaps the gameworks libraries themselves are using more draw calls than necessary? Wouldn't be the first time something similar to this has happened with Nvidia sponsored titles (flat walls and objects with extreme tesselation that could have done the same with single digit triangles) when doing so would obviously hurt the competition.

Regards,
SB
Click to expand...
AMD didn't blame GameWorks this time, and they have a history of mouthing it off whenever necessary, besides the effects are only limited to HBAO+ and PCSS, don't see how these two could quadruple draw calls count to that extent.
 
Silent_Buddha said:
Perhaps the gameworks libraries themselves are using more draw calls than necessary? Wouldn't be the first time something similar to this has happened with Nvidia sponsored titles (flat walls and objects with extreme tesselation that could have done the same with single digit triangles) when doing so would obviously hurt the competition.

Regards,
SB
Click to expand...

I dont know, i will not imply something like that, im really curious to see the FC4 performance analysis, AC seems to get anyway a big problem of performance ( maybe the drawcall ), but looking the situation with this game without proper analysis of the drawcall ( gpuperf maybe ), i think that Nvidia have just maybe got the possibility to find a workaround on an overall problem. This could explain why tesselation have been removed too from the game and "should" be patched later.

DavidGraham said:
It's not a matter of running bad on AMD GPUs, it is a matter of requiring more CPU performance to run well. Guru3d benchmarks are done using high end CPUs, so the problem is not apparent, but switch to a Core i5 or something similar and AMD GPUs stall.
Click to expand...

Yes, i have understand it no worry, but, no explanation, on COD you could be right.. or just some optimization to do.

Then again for AC. looking how bad the game have been tested on amd gpu's, the overall performance problem, the driver will get more overhead on the actual state, any bug, problem in the engine will ask more for the driver. i dont think AC is really a good example anyway for use it as a test for known if this can be an overall better DX11 driver (it is too bad optimized even on nvidia gpu's ). We dont even know what optimisation from Nvidia have been included in return during developpement when they had face such performance problems.

This said, without further and more in dept analysis, i have no idea, what can imply "gamework" features, specially if they are not optimized by developpers on the AMD driver side.
 
Last edited:
Hawaii has 8 ACEs. Anyone studied the effect of AMD GPUs which have only 2 ACEs (most of the HD7000/R200 cards) versus those with 8?
 
You must log in or register to reply here.

Similar threads

S
SDL_Gpu API, SDL 3.0 ... New cross-platform graphics API.
Replies
2
Views
2K
Dominik D
Dominik D
Kaotik
AMD & Sony project Amethyst
Replies
4
Views
481
Shifty Geezer
Shifty Geezer
D
DX12, DX12U, DXR, API bias, and API evolution ..
2
Replies
37
Views
6K
Lurkmass
L
R
State of GDeflate *spawn
2
Replies
25
Views
2K
Remij
R
F
RDNA5 Speculation
Replies
8
Views
3K
BRiT
BRiT
Back
Top