The AMD Mantle Thread

[HurleyBird]

So atm with my 7990 I can look forward to even better frame pacing with 14.1. I will have to wait for Mantle support down the road, which will probably give me ~20-25% better performance due to a somewhat old CPU.

As per AMD, the driver coming out today will support your card with Mantle.

 

[Despoiler]

Aha. So why do we need Mantle at all?
Everybody with an Intel CPU have this kind of performance since the start of BF4 in october.

Nobody is pairing a low end CPU with a $500 card. There is no advantage for people playing with an iGPU outside of 720p and low detail settings. And mid-range gpu buyers have no advantage from Mantle because he will always run into a gpu-limit.

So, an useless API for 95% of the market.

80% of all statistics are made up on the spot. Lets look at a well known hardware sample to see if your guess pans out. Over 50% of steam users only have single or dual core CPUs. 43% of intel CPUs are sub 3 ghz. 54% of the video cards are laptops. I'm going with 95% not being even remotely accurate.

http://store.steampowered.com/hwsurvey?platform=pc

 

[Spjut]

I was not expecting the huge boost for Crossfire - was that expected? Is that just down to a better AFR implementation in mantle? Really curious to see how my AMD rig in sig below does when I get home from work today.

I don't see what's so surprising. CF/Sli systems with high-end GPUs gain alot of performance when overclocking, and that test was just running at 1080p

 

[NTMBK]

So, an useless API for 95% of the market.

Errm, no. The blog post demonstrates significant improvements in frametime consistency and framerate for rigs ranging from Kaveri on iGPU, through a fairly midrange setup with 7970 with FX processor, up to a top end Intel processor with Crossfire 290X. That's basically the entire damn market.

 

[Grooveriding]

I was not expecting the huge boost for Crossfire - was that expected? Is that just down to a better AFR implementation in mantle? Really curious to see how my AMD rig in sig below does when I get home from work today.

50%+ is pretty massive, that is at 1080P using 2x290X though. I'd expect the gain to be smaller at 2560x1600 or surround.

I get that there are a small few who have been dying to crap all over Mantle for months, but most forget that the number of people spending over $300 on a discreet graphics card is about 3% of all sales.

On AT it's easy to forget that the overwhelmingly majority of users are not running 290X, 290, 780ti, 780, 770, 280X etc.

 

[NTMBK]

Maybe because an incredibly small percentage of people are actually able to accomodate the power-hungry ( >700w load system power consumption) and expensive ($1300-1400 street price) dual R9 290X?

It's also a massive improvement for Kaveri. That frametime consistency is insane, that's something you'd expect to see on a console.

 

[Paul98]

Those GPU/CPU graphs are rather interesting in BF4, GPU usage gets very smooth. In the 8350+7970 in directX ms time
GPU avg: 18.55, min 15.70, max 23.93
CPU avg: 18.87, min 14. 32, max 24.66
in Mantle
GPU avg: 11.61, min 11.44, max 11.86
CPU avg: 15.08, min 12.88, max 21.86

I will be interested to see or do some major testing on this, looks like different configurations will get drastically different results. Plus more gains to come.

 

[ASM-coder]

I think it's easy: Show me someone who has combined a Kaveri with a 290X.

Kaveri is still very new, and before Mantle there hasn't been a reason to do this.
The better question is... In 6 months or a year, will we see this type of rig?

I think maybe so.

 

[Skurge]

This has nothing to do with what I quoted and responded to.

Talking about pairing a $100 CPU with a $500 GPU on anything but Mantle, it will not do well.

Sorry, I was using my phone. meant to quote someone else. Well, you can pair a $200 CPU and not be worried about bottlenecks in Frostbite games and get good performance in Most other games.

I got a 4770K cause I was getting dual 290s so I spent a $150 more on the CPU so I wouldn't get a bottleneck cause I was only going to use a single 1440p or 1080p monitor. The only modern game I play is BF4, the other game is starcraft and rFactor.

 

[Skurge]

With mining tax, the number may end up smaller than 5%

I'm not from the states, so I don't pay the mining tax and neither does the rest of the world.

 

[frozentundra123456]

It is an iGPU.
It is amazing something as weak as that 512core@720MHz is actually being bottleneck by the CPU (I would wager the performance gains is mostly to memory improvements than anything else).

I guess you were expecting 50% faster?

Well, it would take something like 50% improvement to make it a game changer. Even with a 14% increase, and assuming mantle gave no increase with a low end cpu/ HD7750, the increase in mantle performance will still not allow it to match a low end cpu with a low end discrete card (GDDR5 HD7750).

So you can play at a crappy resolution at a crappy image quality and get a few FPS faster. Big deal.

 

[el etro]

I think it's easy: Show me someone who has combined a Kaveri with a 290X.

Did you read the article?

• A10-7700K CPU + R9 290X GPU
  • 1080p, Ultra Preset, 4xAA: 40.9% improvement
  • 1600p, Ultra Preset, 4xAA: 17.3% improvement

So, an useless API for 95% of the market.

Change useless for useful in your statement. Will seem legit.




---------------------------------


Anyway, still no one can say if Mantle will really drive AMD to the graphics leadership, it will depend on market acceptance.
Today's tests were good to prove that Mantle existence is, after all, reasonable.

Maybe because an incredibly small percentage of people are actually able to accomodate the power-hungry ( >700w load system power consumption) and expensive ($1300-1400 street price) dual R9 290X? Maybe because almost no one with such a setup would actually run the game at 1080p resolution?

Tell this to people with older CPUs or with AMD CPU setups. My FX-4100 is waiting what mantle can help me in terms of bottleneck.
People thinks too much about eliminating CPU bottlenecks in your setups, and forget that in Gaming performance today relies more on powerful GPUs.
Pairing weak CPU with strong GPU showed a good/big frametime performance hit as techreport tests show. And then Mantle came to reduce this CPU dependancy.

 

[tonyfreak215]

Anyway, still no one can say if Mantle will really drive AMD to the graphics leadership, it will depend on market acceptance.
Today's tests were good to prove that Mantle existence is, after all, reasonable.

Exactly. These improvements are just from an API change that was applied retroactively. Imagine when games are designed from the ground up to use Mantle. (Star swarm, etc)

 

[AtenRa]

Because any game that doesn't use Mantle will not run well?

There are countless Games that will run just fine with a $100 CPU. Just to name a few , any game with Unreal Engine 3 for example.

 

[NTMBK]

Well, it would take something like 50% improvement to make it a game changer. Even with a 14% increase, and assuming mantle gave no increase with a low end cpu/ HD7750, the increase in mantle performance will still not allow it to match a low end cpu with a low end discrete card (GDDR5 HD7750).

So you can play at a crappy resolution at a crappy image quality and get a few FPS faster. Big deal.

We're really focusing on average FPS? Good god, it's like "Inside the Second" never happened. Those Kaveri frametime graphs are ridiculously improved- that gameplay is going to be a LOT smoother than it was under DirectX.

 

[96Firebird]

There are countless Games that will run just fine with a $100 CPU. Just to name a few , any game with Unreal Engine 3 for example.

I get that, I'm talking about modern, cutting-edge games. You know, the games most of us here build our PCs to run...

Anyways, looking forward to review site results, and see how they differ from the results AMD gave.

 

[Gloomy]

My replacement PSU is still in the post so I can't tear into it right now. But it looks good so far.

Personally it seems I'm not in the ballpark for the most gain. Those frametime graphs look pretty darn good though, I'm down for any increase in smoothness... my 4770K should still get some right?

 

[HurleyBird]

Those GPU/CPU graphs are rather interesting in BF4, GPU usage gets very smooth. In the 8350+7970 in directX ms time
GPU avg: 18.55, min 15.70, max 23.93
CPU avg: 18.87, min 14. 32, max 24.66
in Mantle
GPU avg: 11.61, min 11.44, max 11.86
CPU avg: 15.08, min 12.88, max 21.86

I will be interested to see or do some major testing on this, looks like different configurations will get drastically different results. Plus more gains to come.

Yes, in the multiplayer scenario the GPU frame time is more or less a flat line (!)

Interestingly, without Mantle the scenario was equally GPU and CPU bound, but with Mantle things look very CPU limited because the GPU frame time benefited far more than the CPU frame time. This is the opposite one would expect, and even contradicts the other scenarios . I wonder why this is happening? Perhaps it has something to do with the 64 player map already loading the CPU quite heavily, so Mantle isn't able to free up as many resources for the renderer?

 

[Gloomy]

Yes, in the multiplayer scenario the GPU frame time is more or less a flat line (!)

Interestingly, without Mantle the scenario was equally GPU and CPU bound, but with Mantle things look very CPU limited because the GPU frame time benefited far more than the CPU frame time. This is the opposite one would expect, and even contradicts the other scenarios . I wonder why this is happening? Perhaps it has something to do with the 64 player map already loading the CPU quite heavily, so Mantle isn't able to free up as many resources for the renderer?

Wow, good catch. A 4670K would really let that 7970 stretch its legs.

 

[ASM-coder]

I get that, I'm talking about modern, cutting-edge games. You know, the games most of us here build our PCs to run...

Give me some examples, and I will tell you if they run fine on my <100 CPU.
BF4, BioShock Infinite, and Diablo III all run fine. BF4 struggles a bit, like I said.

 

[Nevk]

Aha. So why do we need Mantle at all?
Everybody with an Intel CPU have this kind of performance since the start of BF4 in october.

Nobody is pairing a low end CPU with a $500 card. There is no advantage for people playing with an iGPU outside of 720p and low detail settings. And mid-range gpu buyers have no advantage from Mantle because he will always run into a gpu-limit.

So, an useless API for 95% of the market.

3970X+7970CF+Mantle = 58% fast

 

[Teizo]

Aha. So why do we need Mantle at all?
Everybody with an Intel CPU have this kind of performance since the start of BF4 in october.

Nobody is pairing a low end CPU with a $500 card. There is no advantage for people playing with an iGPU outside of 720p and low detail settings. And mid-range gpu buyers have no advantage from Mantle because he will always run into a gpu-limit.

So, an useless API for 95% of the market.

Well, that is interesting. They said during the announcement that Mantle will enable current GPU's to function like a 2016, or what is it 2018, GPU...right now.

Looking forward to Anandtech's review because I really am hoping that AMD does good with this because only that will bring about necessary improvement to other competing APIs.

 

[DiogoDX]

8350 + 7970 is 25% better in Shangai 64 players. Nice boost. Maybe I'll see this gain in my brother's rig with i7 920 + 7970.

 

[ShintaiDK]

Well, that is interesting. They said during the announcement that Mantle will enable current GPU's to function like a 2016, or what is it 2018, GPU...right now.

Looking forward to Anandtech's review because I really am hoping that AMD does good with this because only that will bring about necessary improvement to other competing APIs.

Thats simply a hype. The only benefit is essentially CPU due to API. But AMDs DX driver already leaves much to be desired.

AMD could essentially have done in Mantle what they could have done in DX:

 

[smackababy]

I am going to try this out tonight with an i7 and a 290, so we'll see just how much more FPS I average.


Anyone get this early?