The AMD Mantle Thread

[LittleNemoNES]

I have a

3770k @ 4Ghz
Sapphire 290x tri-x OC
16GB RAM
1TB 840 SSD
Windows 8.1

I will report my experience here.

Do you folks think that Mantle will improve 4K performance? I have BF4 @ 150% rendering; everything else maxed. Would be nice to hit 200% without a performance impact

 

[bystander36]

I have a

3770k @ 4Ghz
Sapphire 290x tri-x OC
16GB RAM
1TB 840 SSD
Windows 8.1

I will report my experience here.

Do you folks think that Mantle will improve 4K performance? I have BF4 @ 150% rendering; everything else maxed. Would be nice to hit 200% without a performance impact

Mantle shouldn't help with allowing you to choose higher IQ settings. Mantle should help with trying to achieve higher average FPS when lowering settings.

You probably won't see much of a difference in your case.

 

[Erenhardt]

That Star Swarm demo worked great on my Phenom II 965 3,4Ghz and hd7870... at least for a little while. It was droping to 16fps on low settings when things happened.

I was thinking about upgrading to fx8320/6300 because I have am3+ mobo. If mantle can make my phenom last until new arch I would be happy

 

[el etro]

OK I played through the scene in windowed mode, and using CPU-Z, verified that yes, my CPU was running at 4.5ghz..

Not sure if that explains the huge discrepancy between the Mantle off FPS, and what I got though. A 3960x has a max turbo boost of 3.9ghz, which isn't too far off from 4.5ghz.

CPU@4,5Ghz clock(we don't know if stock 3960x can run at his max boost all the time, its 600Mhz from stock, other SB/IB processors can't run max boost with all cores), GPU@1280Mhz clock(close to golden), different settings, no info about movement on the scene, Don't know if 290x is running only at Quiet Mode.

 

[Shivansps]

let me see if i got this right, Mantle is a success even if can only bring insame performance gain under CF and allow to pair a 290x with a 7770K?!

Thats worth implementing a new vendor specific API plus having to optimise for every card(from what im reading regarding 280x, and others)?

The PCPer shows AMD own numbers of 9% for a single 290x and a I7 and a joke number for a low end 260x, THAT is a dissaster if true.

 

[Slomo4shO]

Mantle shouldn't help with allowing you to choose higher IQ settings. Mantle should help with trying to achieve higher average FPS when lowering settings.

You probably won't see much of a difference in your case.

I am sorry but you are basing this conclusion off what available data?

 

[Imouto]

Mantle is designed to get rid of CPU bottlenecks among other benefits.

If you have a GPU bottleneck it will free resources from your CPU.
If you have a CPU bottleneck it will lessened or removed.

As everything your mileage may vary.

 

[Shivansps]

Mantle is designed to get rid of CPU bottlenecks among other benefits.

If you have a GPU bottleneck it will free resources from your CPU.
If you have a CPU bottleneck it will lessened or removed.

As everything your mileage may vary.

Yes, thats exactly what i think it is from day 1, not that black magic that could give AMD cards 900% boost no matter what.

The important, the really important thing about it, in order to take advantanges of what it can do, you need to get rid of DX, i think the star swarm demo is a small example of that.

 

[cmdrdredd]

CPU@4,5Ghz clock(we don't know if stock 3960x can run at his max boost all the time, its 600Mhz from stock, other SB/IB processors can't run max boost with all cores), GPU@1280Mhz clock(close to golden), different settings, no info about movement on the scene, Don't know if 290x is running only at Quiet Mode.

You can force it to run max clock across all cores. That's part of what you do when you overclock.

 

[ASM-coder]

The important, the really important thing about it, in order to take advantanges of what it can do, you need to get rid of DX, i think the star swarm demo is a small example of that.

One of the things to consider is this: The best thing about DX is the worst thing for performance. To support legacy video cards, to support multiple vendor cards and their multiple card architectures, and for game coders to support legacy versions of DX, and multiple OS versions, all this requires multiple code paths, sub-optimal methods, sub-optimal memory management, and plop-on-top code.

If you can simplify all this, you reduce a lot of unneeded overhead.

 

[tonyfreak215]

You can force it to run max clock across all cores. That's part of what you do when you overclock.

We know you can, but the CPU DICE was using was only at 3.5.

Since the stock clock is 3.3 we can assume there was a slight overclock and since they overclocked it, all 6 cores were probably locked to 3.5.

 

[cmdrdredd]

We know you can, but the CPU DICE was using was only at 3.5.

Since the stock clock is 3.3 we can assume there was a slight overclock and since they overclocked it, all 6 cores were probably locked to 3.5.

The post I quoted said SB/IB can't run max turbo clock on all cores. It most certainly can.

 

[Shivansps]

Low level != simplify

Everything gets more complicated as you go lower.

I can understand that new things can be done with it, thats fine, but as long as your game has to be also be playable on DX you will be constrained with it, so Mantle does not solve anything at all, unless you are planning to ditch DX.

 

[cmdrdredd]

Low level != simplify

Everything gets more complicated as you go lower.

I can understand that new things can be done with it, thats fine, but as long as your game has to be also be playable on DX you will be constrained with it, so Mantle does not solve anything at all.

Well, right I guess. You'll probably never see a game go Mantle only so you'll always have DX in there (or maybe OpenGL but more rarely).

 

[bystander36]

I am sorry but you are basing this conclusion off what available data?

All the benchmarks in this thread that show under CPU bottlenecks, Mantle helps a lot, without a CPU bottleneck, Mantle doesn't help hardly at all (1-2%).

Raising IQ settings does not usually add more stress to the CPU (there are exceptions), so increasing visual settings will likely not net you any improvements in performance compared to DX. However, lowering settings will eventually increase FPS to a point where the CPU becomes a bottleneck, allowing Mantle to improve performance.

People with high resolutions are less likely going to experience CPU bottlenecks, so Mantle is less likely going to help. Exceptions may be if you have a slow CPU and or extreme GPU setup (Crossfire 2-4 way).

 

[ASM-coder]

Low level != simplify

Everything gets more complicated as you go lower.

I would disagree, but everybody has their own opinions and coding methods. From my own experience, as you get lower, and less is done for you, you have more control and know exactly what needs to be coded. It's liberating.

 

[bystander36]

I would disagree, but everybody has their own opinions and coding methods. From my own experience, as you get lower, and less is done for you, you have more control and know exactly what needs to be coded. It's liberating.

That doesn't contradict what he said.

Lower level is more complicated, but gives you more options.

Higher level is simpler, but limits your options.

 

[tonyfreak215]

The post I quoted said SB/IB can't run max turbo clock on all cores. It most certainly can.

But not stock. You have to force it to.

 

[Bubbleawsome]

http://www.tomshardware.com/news/battlefield-mantle-catalyst-drivers,25908.html
Is this news to anyone? I didn't know.

 

[GaiaHunter]

All the benchmarks in this thread that show under CPU bottlenecks, Mantle helps a lot, without a CPU bottleneck, Mantle doesn't help hardly at all (1-2%).

As of the moment.

The CPU bottlenecks were the easiest fix to implement.

 

[Rickyyy369]

http://www.tomshardware.com/news/battlefield-mantle-catalyst-drivers,25908.html
Is this news to anyone? I didn't know.

It is known.

 

[smackababy]

I would disagree, but everybody has their own opinions and coding methods. From my own experience, as you get lower, and less is done for you, you have more control and know exactly what needs to be coded. It's liberating.

It is definitely not simplified though. If it was, everyone would be using machine code.

 

[krumme]

That doesn't contradict what he said.

Lower level is more complicated, but gives you more options.

Higher level is simpler, but limits your options.

And it explains why mantle is preferred todya and not 10 years ago. Today we have a good handfull of engines as platform for most games unlike 10 hears ago where there was a magnitude more engines. As the work for creating an engine specializes and cost goes up the need and benefit of the extra control outweights the cost.

The change of api is primarily a result of the structural organization on the engine market with fewer enginess. Add newer gpu tech, more cores and newer methods for programmability.

 

[Cloudfire777]

Is now the time to say I told you so?

We will see how this will turn out.

I say max 10% ish extra FPS for most people with high end systems

45% faster on systems with weak APU and a powerful GPU.
Mantle reduce overhead and reduce bottleneck, making FPS much higher since the GPU can do its magic undisturbed.

For systems with a powerful enough CPU/APU which doesn`t have any bottleneck, less overhead means very small gains in FPS compared to DirectX.

Thats why they say "UP to 45% faster".
Those who think they will gain 45% more FPS on R9 290x and a i7 CPU are out of their minds.

Pretty sneaky marketing, but hey, its AMD

• Core i7-4960X CPU + R9 290X GPU
  • 1080p, Ultra Preset, 4xAA: 9.2% improvement with Mantle
  • 1600p, Ultra Preset, 4xAA: 10% improvement with Mantle
• Core i7-4960X CPU + R7 260X GPU
  • 1080p, Ultra Preset, 4xAA: 2.7% improvement
  • 1600p, Ultra Preset, 4xAA: 1.4% improvement

 

[Slomo4shO]

All the benchmarks in this thread that show under CPU bottlenecks, Mantle helps a lot, without a CPU bottleneck, Mantle doesn't help hardly at all (1-2%).

Raising IQ settings does not usually add more stress to the CPU (there are exceptions), so increasing visual settings will likely not net you any improvements in performance compared to DX. However, lowering settings will eventually increase FPS to a point where the CPU becomes a bottleneck, allowing Mantle to improve performance.

People with high resolutions are less likely going to experience CPU bottlenecks, so Mantle is less likely going to help. Exceptions may be if you have a slow CPU and or extreme GPU setup (Crossfire 2-4 way).

The only semi-comparative bench was the following and it still showed substantial performance boosts. I'll wait for real reviews to draw an factual conclusion :whiste:

Test case 2: Standard 64-player multiplayer
CPU: AMD FX-8350, 8 cores @ 4 GHz
GPU: AMD Radeon 7970 3 GB
Settings: 1080p ULTRA 1x MSAA
OS: Windows 8 64-bit
Level: Siege of Shanghai
Level was tested with 64 “pseudo players” that we have for our own internal testing that simulates heavy game workload that we have in multiplayer in order to get more deterministic results compared to full real multiplayer. 64 players on the large Battlefield levels is really demanding of the CPU so this test case is primarily CPU-bound.
Result: 18.87 ms/f -> 15.08 ms/f = 25.1% faster