Hardware AMD Mantle - Battlefield 4 Performance

We test AMD's latest discrete and integrated GPUs using its new Mantle API in Battlefield 4.
http://www.bit-tech.net/hardware/graphics/2014/02/05/amd-mantle-bf4-performance/1

 

want to borrow my AMD 9590 system to see the gains for that??

 

Would have been good to see a lower end cpu paired with a higher end gpu (one of the apus with a 290 for instance, or an i3), to see if mantle will mean in the future you can really focus on getting the best gpu possible (it's already leaning that way, but in theory mantle will push it even further)

 

Would it be fair to say that the Kaveri results are promising, but it perhaps shows that AMD are missing about 25% CPU grunt to make it a viable gaming APU?

 

Guys is it me, or the DX11 results on the 290X with the 14.1 beta drivers, are worst than before?
On previous reviews here, (not only on 290X but 780s as comparison), with same 3570 at 4.2Ghz,

2,560 x 1,600, 4x AA 16x AF, ultra detail settings, DirectX 11
The 290X (normal???) with Cat 13.11 was Minimum 36, Avg 44.


Now on "Uber" mode with Cat 14.1
2,560 x 1,440, Ultra Detail (no description about AF & AA though)
Min 40 Avg 48.

I doubt 4AA, 16AF and 410,000 pixels can provide only 10% hit. Or do we miss something from the settings used?

Other than that, great review proving that all these 45%-50% performance on graphics alone, was a "myth". Yes on the low end, but on overclocked i5 and i7 the gain is minimal. (hell my 4820K is at 5Ghz).

 

I would have liked to see the corresponding NVidia cards performance in the charts too to show if it was actually worth buying an AMD card for mantle or if NVidia cards run better under DX11 than AMD's under mantle.

 

AMD7950 Fx8320@4.3

Mantle test Paracel Storm same empty server
High textures, High AF, Medium Shadows, Low FXAA, everything else low

DX11 B flag

Stuttery mess when moving, cpu gpu graph fluctuates a lot.
minimum 41 fps average 60
Unplayable

Mantle B flag

Smooth cpu gpu graph very smooth
minimum 59 fps average 74
Very smooth and playable. 59 fps feels very smooth no stuttering.

All this on an empty server. Even 64 man server minimum is 59 fps.

Using Mantle in sp gives you nothing. In multiplayer though is a Amazing.

 

Yeah I think sadly this really was a case of going with multiplayer, whether it's as consistent or not. Seeing as that's where the focus of the game lies, what most people will be using and is where the CPU is more limiting, it seems like a lost opportunity not testing it.

Could you not have just played and logged 10 games on each system and then taken your averages? I understand that this is still a small sample set, but at least it would be somewhat indicative of what owners of these cards might be able to expect?

 

Will wait and see where this goes with other games. I like my r290 and mantle was a big reason I went AMD this time. These preliminary results though almost make me wish I opted for gsync. That seems like a more revolutionary improvement in image quality.

 

Biggest problem for the iGPU in the 7850K is the lack of memory bandwidth available. This is made more obvious when comparing either the iGPU to a HD7750 (512 GCN) or a HD7750 GDDR5 to a HD7750 GDDR3.

If they could get a quad channel memory controller onto the next APU it would show a massive performance improvement (at 1080p) even keeping everything else the same.

 

Would 2x8GB of 2133mhz RAM be a better set-up then for an iGPU set-up or is the dual channel already saturated?

 

apparently if you use double sided rather than single sided dims you get a slight speed increase as well at any given speed , which does show that dual channel is limited the AMD iGP`s

 

Unless AMD makes a quad or tri channel board duel channel is what you get.

Reviewers have speculated before that the Gpu is bandwidth starved. The fact mantle is not a huge performance boost may show that side of it.

Testing an empty server is hardly testing anything. Each player movement requires CPU and Gpu and network overhead. That's why 64 player bf4/3 is not on consoles to begin with.

People have wondered if they could sim a run with fake players to auto attack the terrain on repeat.

As I've said in other topics thief will show what mantle is capable off more than bf4.

Bf4 its about 7% with windows 8 giving a free 5-10% that's a useful performance boost. Not sure I expected more than 7-10%. I know AMD fans where hoping it was like 30% + but no site has got close to those numbers that I'd trust for benchmarks.

Be happy with 7-10% free performance is what I'd say. You never know in a year it could be 20%. AMD now needs to push it to developers and hope for uptake.

 

rollo - I`ll join you on BF4 64 player on my xbox one if you would like.....

 

Even at 3Ghz+ the memory is still a massive limitation on the "64 bit" bus.

 

Would be interesting to see how much of an improvement it would make for an AMD CPU with a discrete card.

 

Wasn't this supposed to be awesome?

 

According to the Techreport results the 290X shows significant improvements when linked with the 7850K (IIRC in the region of 60%).

Interestingly it also touchs on what other sites have mentioned before - Nvidia seem to have a better handle on CPU limited performance under DX (In the CPU limited test the 780TI was faster than the 290X with DX & Mantle).

 

That just shows how slow that CPU is and how badly it's capping the performance of AMDs top end cards.

Be more interested in seeing results for the 8350 with 290x as that's a more realistic combo. The 8350 is not even a lot more expensive than the 7850k. Nobody is ever gonna pair a £500 Gpu with a £100 CPU it's just not logical.