The AMD Mantle Thread

[Despoiler]

Just because you doubt it was running at 4.5, doesn't mean you can claim it as a comparison.

This. Please set your CPU clock to match theirs to get a more accurate test.

 

[NTMBK]

Maybe the frametimes are improved, but it is still at 720p low settings. Even if you could force it to medium or high on 720p, you would still be better off with a console.

No it isn't, it's at 720p medium. Seriously, did people even read the article?

Test case 1: Low-end single-player
CPU/GPU: AMD A10-7850K (‘Kaveri’ APU), 4 cores @ 3.7 GHz
Settings: 720p MEDIUM settings.

And a console doesn't have a 15 year back catalogue of games that I can play, and requires me to throw out every single game I bought for the last generation. I'd rather have one box that can play Battlefield 1942, Battlefield 3, and Battlefield 4. But that's just personal preference; in terms of raw performance/$, yes, you can't beat a PS4.

 

[wand3r3r]

A 770SLI is NOWHERE NEAR a 290X CF setup, the 290x is ~25% faster. Your screenshots dont even look the same as the ones submitted by AMD, and the test setup, hardware, software, etc. is not even the same. Basically a pointless claim.

PS: 290X is 40% faster in BF4 according to this review Link You really think you can touch a 290X CF with a pair of 770? lol

Not only are his 770's "faster" by his own testing, he makes classic claims like this (which is "proven" through his super fast 770's :whiste:

AMD's drivers have inferior CPU utilization..

 

[Carfax83]

Wait, you don't think that taking a screen while someone is actively playing (moving) is better than standing still?

If you want the results to be as reproducible as possible for the sake of comparison, of course not.

 

[Imouto]

This. Please set your CPU clock to match theirs to get a more accurate test.

No overclock at all would be better since OCed CPUs don't go to base clocks when stressed.

 

[ASM-coder]

If you want the results to be as reproducible as possible for the same of comparison, of course not.

And what about "difficulty" settings? We don't know what Dice used, but the number of enemies and the AI work may make a big difference.

 

[Carfax83]

How about waiting for a review? Your claims of a SLI 770 setup being faster than a CF 290X setup are just straight wrong. As proven by MANY review sites.

I really wish people would stop misquoting me, or learn to read properly.

I never said that the GTX 770 SLI was outright faster than the 290x CF. That's absurd.

I was claiming that because AMD's drivers (Direct3D path) were suboptimal when it comes to feeding the GPUs, causing the GPUs to underperform. Now Mantle has rectified that issue.

That's what I'm claiming.

 

[Paul98]

I really wish people would stop misquoting me, or learn to read properly.

I never said that the GTX 770 SLI was outright faster than the 290x CF. That's absurd.

You are saying the 770 SLI out performs 290x CF in BF4 on DirectX

 

[Paul98]

Any news when the driver will be released, and star swarm?

 

[ASM-coder]

I never said that the GTX 770 SLI was outright faster than the 290x CF. That's absurd.

But you believe that your GTX 770 SLI setup already outperforms the Dice demo(pre Mantle), and it's not because of your CPU. So it must be... what?

 

[Carfax83]

Turbo wouldn't be at 3.9 using all 6 cores. More likely 3.5-3.7ish depending on cooler.

Right, because if that scene could cause my CPU to go full speed @ 4.5ghz due to how CPU intensive it is, the slower stock 3960x wouldn't be at 3.9ghz..

That really makes a lot of sense.. That scene by the way had my CPU at 50% usage at the highest peak if I recall.

And yes, I do have speedstep enabled, and C1E..

 

[Imouto]

Interesting to say the least.

Something is broken there. Going by those times in the last testing with the 3960K + 290X CF it's impossible to have 116 fps with a GPU time of 14.02 ms.

I think the perf overlay is adding the times of both GPUs so it would be massively CPU bound in the first screen and slightly CPU bound in the second (14.02/2=7.01 < 8.38).

 

[Carfax83]

You are saying the 770 SLI out performs 290x CF in BF4 on DirectX

Yes, but not because the 290x CF is slower. The hardware is obviously considerably faster than the GTX 770 SLI. It's the software that's holding it back.

Anyway, I'm going to do another test. I think I figured out why AMD's screenshot looks so differently.. And it may not be due to motion.

In that particular area, there's an enemy with a grenade launcher. I think AMD took the screenshot when the grenade exploded near the main character, hence the sparks and red hue.

I'm going to see if I have any frame rate drops when that incident occurs..

 

[Stuka87]

I really wish people would stop misquoting me, or learn to read properly.

I never said that the GTX 770 SLI was outright faster than the 290x CF. That's absurd.

I was claiming that because AMD's drivers (Direct3D path) were suboptimal when it comes to feeding the GPUs, causing the GPUs to underperform. Now Mantle has rectified that issue.

That's what I'm claiming.

Except the benchmarks that I have looked at (and that were posted above by another poster) shows an SLI 770 is quite a bit slower than a CF 290X in BF4. Which you are claiming is wrong apparently?

 

[Shmee]

My understanding is mantle is released in the patch today for Bf4. Patching now

 

[tonyfreak215]

Right, because if that scene could cause my CPU to go full speed @ 4.5ghz due to how CPU intensive it is, the slower stock 3960x wouldn't be at 3.9ghz..

That really makes a lot of sense.. That scene by the way had my CPU at 50% usage at the highest peak if I recall.

And yes, I do have speedstep enabled, and C1E..

Talking stock; non overclocked.

Unless I misunderstand, a stock 3960x wouldn't hit 3.9ghz on all 6 cores due to how the turbo works.

Even then they state it was at 3.5Ghz.

 

[Slomo4shO]

Star Swarm benchmark is available on Steam...

 

[Imouto]

Previous post elaborated:

Test case 3: High-end single-player with multiple GPUs

DX
CPU 13.24 = 75 FPS
GPU 18.38 / 2 = 9.19 ms = 109 FPS
Screen showing 76 FPS = Massively CPU bottlenecked

Mantle
CPU 8.38 = 119 FPS
GPU 14.02 / 2 = 7.01 = 142 FPS
Screen showing 116 FPS = Slightly CPU bottlenecked

 

[Paul98]

Star Swarm benchmark is available on Steam...

sweet, I had checked just a couple minutes ago and it wasn't there.

 

[Carfax83]

Except the benchmarks that I have looked at (and that were posted above by another poster) shows an SLI 770 is quite a bit slower than a CF 290X in BF4. Which you are claiming is wrong apparently?

I'm just going by what I personally have done, and AMD's screenshots.

Anyway, I re-tested that area and sure enough, AMD had taken screenshots during grenade explosions, which explains the red hue and all that.

And yes, there was a frame rate drop, although not drastic depending on how close it was. If it was a direct hit, the frame rate would drop into the 80s.

If it was close but not direct, then frame rate would stay above 100 FPS, but below 110 FPS.

Judging by AMD's screenshots, I can tell the grenade explosions weren't direct hits because there are no injury indications on screen ie blood splatter.

 

[Dresdenboy]

I was claiming that because AMD's drivers (Direct3D path) were suboptimal when it comes to feeding the GPUs, causing the GPUs to underperform. Now Mantle has rectified that issue.

And what did Microsoft contribute to DirectX? Only the include header files and docs?

 

[Carfax83]

Talking stock; non overclocked.

Unless I misunderstand, a stock 3960x wouldn't hit 3.9ghz on all 6 cores due to how the turbo works.

Even then they state it was at 3.5Ghz.

OK I did some checking, and you are right. I think that particular scene had a workload that used 4 cores, so that would put it around 3.5 to 3.7ghz..

 

[Dresdenboy]

Something is broken there. Going by those times in the last testing with the 3960K + 290X CF it's impossible to have 116 fps with a GPU time of 14.02 ms.

I think the perf overlay is adding the times of both GPUs so it would be massively CPU bound in the first screen and slightly CPU bound in the second (14.02/2=7.01 < 8.38).

Could that be caused by AFR, so that each frame takes longer, but drawing them in CF doubles fps?

 

[bepo]

I'm going to wait for a less biased review than benchmarks given by AMD/DICE before drawing any conclusions. It's incredibly easy to cherry pick a single scene that shows 45% for a given setup but that doesn't mean it's representative of average or even normal game play. For all we know they could have recorded a play through of this level with both setups and then simply grabbed frames from the scene where the difference was the largest.

 

[Imouto]

I'm just going by what I personally have done, and AMD's screenshots.

Anyway, I re-tested that area and sure enough, AMD had taken screenshots during grenade explosions, which explains the red hue and all that.

And yes, there was a frame rate drop, although not drastic depending on how close it was. If it was a direct hit, the frame rate would drop into the 80s.

If it was close but not direct, then frame rate would stay above 100 FPS, but below 110 FPS.

Judging by AMD's screenshots, I can tell the grenade explosions weren't direct hits because there are no injury indications on screen ie blood splatter.

You're doing it wrong. You're comparing your OCed CPU dealing with frames at a rate of 9.30 ms (107 FPS) with a stock CPU dealing with frames at a rate of 13.24 ms (75 FPS).