The AMD Mantle Thread

[Paul98]

So, there is a memory leak? In both mantle and DX? That could explain the worse performance compared with Nvidia. Is it only with AMD video cards?

I don't have a memory leak unless I use mantle, DX works fine.

 

[Will Robinson]

n0x1ous
Senior Member



Join Date: Sep 2010
Location: USA
Posts: 935

Quote:
Originally Posted by ASM-coder
I would say that the majority may still be happy, because they have seen what is possible, and can look forward to the improvements.
I don't even have a GCN card, but I am happy with it, and I have decided to buy one.
Just waiting for the dust to settle, maybe for 14.2(or whatever.)
Had AMD/Dice waited, we would all be more frustrated, so I think it was the right move.

You think releasing broken software that promises to be fixed later is better than delaying and fixing before release? Do you also think BF4 should have been released as it was or should Dice have had a functioning game before release?

You are certainly entitled to your opinion, but I just dont see how releasing something that knowingly doesnt work makes any sense. It just makes them look bad to the users who are perhaps not in the know to the same degree of AT forum members....
__________________

I'm not sure how much BF4 you play but I found the game quite "functional" thanks.
I think your "knowingly doesn't work" comment is being a little careless with the facts.

 

[n0x1ous]

I'm not sure how much BF4 you play but I found the game quite "functional" thanks.
I think your "knowingly doesn't work" comment is being a little careless with the facts.

I play plenty of BF4. Its fine now; I am referring to its state at the time of its release. It was a disaster - one of the buggiest releases I can remember.

"knowlingly doesnt work" however was in reference to Mantle on HD7XXX series.

 

[kane15]

I play plenty of BF4. Its fine now; I am referring to its state at the time of its release. It was a disaster - one of the buggiest releases I can remember.

im a diehard BF player always have been and i agree 100% it should have been delayed but the big shots over at EA didn't want it to bump into titanfall and they wanted to be out before CoD ghost

i dont agree however when it comes to the beta drivers... that is what beta means to test and use the community to help find bugs

 

[7stars]

I don't have a memory leak unless I use mantle, DX works fine.

yes, here just Mantle..
but the not clear thing to me is:
is that noticeable with 7000 series only now (or for less than 3GB vram) or even with the 290/x ?
'cause if also 290, i think that the BF4 benchs over the net are a bit distorted or anyway to not take them in serious consideration until they fix it...
that memory leak involves an apparent discrepancy in the framerate and also frametime...'cause the frame drops are evident every x secs and not only "under the hood"... so it distorts the graph and the bench for sure IMHO

 

[ASM-coder]

You think releasing broken software that promises to be fixed later is better than delaying and fixing before release? Do you also think BF4 should have been released as it was or should Dice have had a functioning game before release?

You are certainly entitled to your opinion, but I just dont see how releasing something that knowingly doesnt work makes any sense. It just makes them look bad to the users who are perhaps not in the know to the same degree of AT forum members....

If you've worked for a software company, then you'll know that a Beta release is an important tool used to find/shake out bugs. It's virtually impossible to reproduce all situations and variables that exist on customer setups, and multiplayer creates an entirely different set of difficulties.
Since there is an on/off setting for Mantle, I see absolutely no harm done.

I also played BF4 on some of the earlier iterations(but not day one), and never experienced a crash, or serious problem. Sound stuttering was the biggest problem. But we don't really know if the Dice group was surprised by what some players experienced, or if they knew about all the bugs. Some pretend to know though.

 

[7stars]

If you've worked for a software company, then you'll know that a Beta release is an important tool used to find/shake out bugs. It's virtually impossible to reproduce all situations and variables that exist on customer setups...

ok, but I would have presented the "revolution" better than this...and above all already released a fix after the 14.1...
If they said since the beginning that ALL GCN are supported... they can't openly declare (releasing these beta drivers) to privilege the new cards owners...they clearly know that 7000 was released in 2011...alpha or beta IMHO doesn't matter...the 7000 had to be the most proven setup. The opposite is really annoying.
Otherwise they could say that this "revolution" was reserved to new 290 and next only...or do everything at the same time and THEN release (even if beta). And yet, how is possible that someone, after 2 years (at least) of development, doesn't see a memory leak with the "old" cards (or not only?), available to them for so long or has not an already and efficient "thin driver"? That's a basic thing. OR 2-3 years of development is a fake OR they are so slow like snails :biggrin:

 

[ASM-coder]

And yet, how is possible that someone, after 2 years (at least) of development, doesn't see a memory leak with the "old" cards (or not only?), available to them for so long or has not an already and efficient "thin driver"? That's a basic thing. OR 2-3 years of development is a fake OR they are so slow like snails :biggrin:

Memory leaks are notoriously tricky business, and we are still only assuming it's a memory leak, based on
some comments on the BF4 message board.

 

[7stars]

...assuming...

how you define it when the memory usage goes always higher than your available Vram, regardless of any graphic setting? and then what does that entail?
e.g. i have 2GB of Vram.... BF4 goes to 2800+ MB (even at low settings)... at my home 2GB are 2048 MB... maybe something sounds weird to you?

 

[ASM-coder]

how you define it when the memory usage goes always higher than your available Vram, regardless of any graphic setting? and then what does that entail?

OK, say that's the case. Then my "guess" is that it will be fixed in a BF4 patch. I think Mantle gives the developer much more control over memory management, and someone didn't do some required housekeeping.

 

[Paul98]

Not sure if the memory leak is on BF4 side or driver side. with my 7850, and i5, after a round or two I am using all 8GB of memory.

 

[7stars]

OK, say that's the case. Then my "guess" is that it will be fixed in a BF4 patch. I think Mantle gives the developer much more control over memory management, and someone didn't do some required housekeeping.

tested again and again...and i think that there's an issue with GPU-Z and Mantle rendering...seems it does conflict 'cause HWINFO64 doesn't show that vram usage, instead...and the game doesn't stutter in the same way despite the stuttering is sometimes noticeable yet...so maybe it takes purely optimization
some of you launched GPU-Z while ingame?

 

[ASM-coder]

tested again and again...and i think that there's an issue with GPU-Z and Mantle rendering...seems it does conflict...

Which goes with what I was saying before.
It's doubtful that AMD or Dice used GPU-Z in their testing process.

 

[Carfax83]

Mantle effectively halves CPU requirements in games. It makes desktop CPU upgrades optional for at least one more generation and has tremendous implications in mobile gaming.

If you don't see how this translates into more GPU and more games sales combined then you couldn't possibly be NVidia's CEO in the first place.

If it's not Mantle, it will be something else, but it's gonna happen, whether some people like it or not. The $$$ implications are just too big.

I thought about this after posting my OP, and I was going to edit my post but it was so late that I just said, "meh." Good call out.

Of course you're right, Mantle does have huge implications for mobile gaming, and even for low to mid tier desktop gaming.

But as I mentioned before, Mantle's technical virtues likely won't determine it's long term success or failure. AMD's market share in the tablet space is fairly marginal from what I understand, and since Mantle is currently tied to GCN, Mantle will not (and may never) have a major impact on mobile gaming for a long time.

If Microsoft releases DX12 with Windows 9 with greatly reduced CPU overhead and increased thread management capabilities (same for OpenGL), then Mantle is history.

 

[ICDP]

If Microsoft releases DX12 with Windows 9 with greatly reduced CPU overhead and increased thread management capabilities (same for OpenGL), then Mantle is history.

A lot of ifs in that statement.
Windows 9 is planned for mid-late 2015 depending on the rumours you believe. Even then there are no guarantees that DX12 will be part of that release.

I find it amazing that the vast majority of Nvidia fans are happy to predict Mantle's demise before it has been given a chance. Instead of waiting to see if it can bring benefits to PC gaming they are much quicker to declare it pointless. Even after many previews/reviews show that there are definite benefits for even enthusiast level PCs many Nvidia fans declare it "meh".

It's in alpha at the moment and will continue to improve. Hopefully to the point where other devs and even Nvidia take notice.

Something tells me they actually would prefer if the PC gaming industry stagnated rather than accept AMD tech in their Nvidia cards. I wonder how they feel about Eyefinity or GDDR3/4/5.

 

[Spjut]

DX12 won't get adopted fast either unless Microsoft decides to backport it to Windows 7. Who knows if even Win 8.x will get it...

 

[wand3r3r]

Is true audio coming to BF4? I guess it will be in Thief, but am unsure if it's coming in anything else. I don't even know if it came with mantle or not yet.

 

[bystander36]

A lot of ifs in that statement.
Windows 9 is planned for mid-late 2015 depending on the rumours you believe. Even then there are no guarantees that DX12 will be part of that release.

I find it amazing that the vast majority of Nvidia fans are happy to predict Mantle's demise before it has been given a chance. Instead of waiting to see if it can bring benefits to PC gaming they are much quicker to declare it pointless. Even after many previews/reviews show that there are definite benefits for even enthusiast level PCs many Nvidia fans declare it "meh".

It's in alpha at the moment and will continue to improve. Hopefully to the point where other devs and even Nvidia take notice.

Something tells me they actually would prefer if the PC gaming industry stagnated rather than accept AMD tech in their Nvidia cards. I wonder how they feel about Eyefinity or GDDR3/4/5.

Just because someone doesn't like the direction Mantle may take us (fragmentation), doesn't mean we prefer one brand over the other. And just because someone doesn't like Mantle, doesn't mean they don't recognize what it does well. Just because someone likes Nvidia or AMD, doesn't make them blind, or require them to be a fanboy.

It would be better for all of us if DirectX were to advance in a way that is less hindered by the CPU, and has better multithreading support. DirectX can't go away anytime soon, and it would be better if we were using 1 API, rather than 2+. Alternatively, Mantle could become less low level and support more hardware, in which case it could replace DirectX, but that is not what they appear to want with Mantle, so DirectX improvements are a better future for us. Mantle may be what helps MS see what has to be done.

 

[Will Robinson]

I don't see any way MS could produce a low level API like Mantle without have full access to the hardware specifications of each generation.
No one outside of AMD or NVDA has that other than specific game developers.
So unless MS had a "Game optimizations" dept I don't see how DX could ever work as closely with the hardware as an individual API created by the chip designers themselves.

 

[bystander36]

I don't see any way MS could produce a low level API like Mantle without have full access to the hardware specifications of each generation.
No one outside of AMD or NVDA has that other than specific game developers.
So unless MS had a "Game optimizations" dept I don't see how DX could ever work as closely with the hardware as an individual API created by the chip designers themselves.

I don't think they have to go low level like Mantle. I don't even think they need to have as good of performance as Mantle. DX just needs to be close enough for it to be a superior choice.

As we have seen, the primary advantage to Mantle is removing the CPU bottleneck. 10%-15% boosts are not worth fragmentation. What may be worth fragmentation is the 50% boosts gained in CPU bottleneck situations. That is all MS needs to address. Give DX a multithread overhaul, and make it a little lighter per call would go a long ways to being good enough and not require fragmentation.

 

[Despoiler]

I don't think they have to go low level like Mantle. I don't even think they need to have as good of performance as Mantle. DX just needs to be close enough for it to be a superior choice.

As we have seen, the primary advantage to Mantle is removing the CPU bottleneck. 10%-15% boosts are not worth fragmentation. What may be worth fragmentation is the 50% boosts gained in CPU bottleneck situations. That is all MS needs to address. Give DX a multithread overhaul, and make it a little lighter per call would go a long ways to being good enough and not require fragmentation.

They would have to rewrite it from the ground up. What they have right now is a bandaid to an outdated design. I don't think a high level API is ever going to be able to compete with a low level API in terms of performance. That is kind of why you use a low level API. Raw performance. High level APIs leverage other strengths like easy of use, wide compatibility, among other things.

 

[bystander36]

They would have to rewrite it from the ground up. What they have right now is a bandaid to an outdated design. I don't think a high level API is ever going to be able to compete with a low level API in terms of performance. That is kind of why you use a low level API. Raw performance. High level APIs leverage other strengths like easy of use, wide compatibility, among other things.

As you have seen from a lot of benchmarks, there has been very little improvements if the game is GPU bound, when using Mantle. Sometimes 0 improvements. That goes to show that DX isn't completely holding people back. The problem comes with the draw calls. A huge performance improvement could be had simply by making the API multithread friendly. Instead of being held back by a single thread, if you could leverage 2-4 cores, you remove most existing bottlenecks on a balanced system.

Cutting back how much time each call takes may be more difficult, and much more limited with such a high level API, but multi-threading should be something that can be done, and remain high level. Which means better support for a wider range of hardware.

 

[Despoiler]

As you have seen from a lot of benchmarks, there has been very little improvements if the game is GPU bound, when using Mantle. Sometimes 0 improvements. That goes to show that DX isn't completely holding people back. The problem comes with the draw calls. A huge performance improvement could be had simply by making the API multithread friendly. Instead of being held back by a single thread, if you could leverage 2-4 cores, you remove most existing bottlenecks on a balanced system.

Cutting back how much time each call takes may be more difficult, and much more limited with such a high level API, but multi-threading should be something that can be done, and remain high level. Which means better support for a wider range of hardware.

I don't believe you read anything I just said nor do you understand what DX is doing right now. DX 9/10 was designed for a single thread and they've already put the bandaids in so DX11 can use 2-3 threads effectively. DX11 doesn't scale beyond that. DX11 mutithreading can, but it shunts the responsibility to the driver, which is basically like building a mutithreaded OS at the GPU level. Not even remotely trivial. So we already know that MS doesn't want to do the heavy lifting themselves. Your idea that MS is going to do this has already been disproved. All that aside, DX11 mutithreading implementation only doubles your object/draw-call throughput. Mantle is so far beyond double the draw calls.

 

[bystander36]

I don't believe you read anything I just said nor do you understand what DX is doing right now. DX 9/10 was designed for a single thread and they've already put the bandaids in so DX11 can use 2-3 threads effectively. DX11 doesn't scale beyond that. DX11 mutithreading can, but it shunts the responsibility to the driver, which is basically like building a mutithreaded OS at the GPU level. Not even remotely trivial. So we already know that MS doesn't want to do the heavy lifting themselves. Your idea that MS is going to do this has already been disproved. All that aside, DX11 mutithreading implementation only doubles your object/draw-call throughput. Mantle is so far beyond double the draw calls.

1st off, I didn't say what MS is doing. I said what they can do. I said IF they do this or that and you are the only one saying that they will or will not do.

I realize multithreading isn't trivial, but it IS something that can be done with high level API and give immediate results. And it is something that would be better than having Mantle and DX.

 

[Deadjester451]

This is probably going to be a totally ignorant question, but from what I was reading here. DX is working from the upper end and Mantle from the lower end, could their effects be combined for a superior out come by working together?