The AMD Mantle Thread

[blastingcap]

If Mantle truly provides the performance advantages AMD has talked up, then I'm glad if they diverted resources away from enabling DirectX 11 multithreaded rendering over to Mantle.

I agree, though if Mantle fails, I hope AMD gets on that DX MTR pronto!

 

[dacostafilipe]

It makes it even more fun...when people tlak about AMD being "fully" DX11.2 compliant...when their driver isn't multithreaded.

And I suspect you are correct...AMD has been real long in the tooth making a DX11 multithreaded driver.
One can wonder why AMD's tries Mantle..and not just fix the driver first.

Do NOT confuse DX Multithread Rendering with Multithreaded Driver. AMD's Driver is Multithreaded, it just don't support DXMTR.

The idea behind DXMTR is to allow multiple threads to create rendering commands. You are still limited to one thread when sending this commands to your driver (aka the actual rendering).

 

[Carfax83]

Deferred context is a massive failure. Mantle will provide a much better way for multi-threaded rendering, because the command buffers can be re-used. This is how deferred context should work in DX11, but MS choose a very inefficient implementation to ensure backward compatibility.

Is deferred context a massive failure? In adoption perhaps, because of AMD's lack of implementation.

But from a performance standpoint, the performance gains from using it have been fairly impressive it seems. Civilization 5 had a massive increase from using that feature, about 50% or so initially versus IC.

Assassin's Creed 3 also got a nice boost, a conservative 24% gain according to this NVidia white paper. Project C.A.R.S is also supposed to support it.

Anyway, I can understand Microsoft's willingness to preserve backward compatibility, but eventually, they will have to sacrifice that if they want to move forward.

New driver models is part of that, which I suppose is why they are no longer upgrading Direct3D for older operating systems. Still, I can't wait to see how much of an improvement Mantle brings to the table. If it's really that good, then I would hope that NVidia and Intel will adopt it as well.

 

[zlatan]

Is deferred context a massive failure? In adoption perhaps, because of AMD's lack of implementation.

It has nothing to do with AMD or others. We just don't want to mess up with more active driver threads, which we cannot control. In many situation these will eat up the resources from the engine tasks. If you want to do some very complex simulation on the scene, than this scenario is unacceptable.

In a strategy game, the main issue will be the batch count. So there will be many situations where deferred context will increase the performance. There should be some scenes where it just run slower, but that's fine if you gain in many cases.
It maybe possible to increase the speed in a TPS, but you have to aware to the complex simulations, because the driver will eat up more and more CPU resources, and you may have left many free CPU-time, but it will unusable.

A real issue here that most developers don't want to decrease the gameplay quality just for an efficient deferred context implementation. Mantle will bring up a solution where the engine will control everything. You control everything, and there won't be any hidden driver threads and other limitations. Of course the Mantle will eat some resource, but you control how much. This will lead to predictable performance in any scenario, just like on the consoles.

 

[BallaTheFeared]

Nice try. GCN is in APUs and the latest AMD GPUs, which total more than 5% of the market.

Would be nice if you could actually produce numbers to support that.

Are you looking at current sales, or actual market representation?

 

[Carfax83]

It has nothing to do with AMD or others. We just don't want to mess up with more active driver threads, which we cannot control. In many situation these will eat up the resources from the engine tasks. If you want to do some very complex simulation on the scene, than this scenario is unacceptable.

I see. BTW, I have a question for you zlatan. Is it possible to have multiple rendering threads using IC rather than DC? There are some efficient engines like the CryEngine 3 which don't use DC, but still manage to extract high performance from multicore processors.

A real issue here that most developers don't want to decrease the gameplay quality just for an efficient deferred context implementation. Mantle will bring up a solution where the engine will control everything. You control everything, and there won't be any hidden driver threads and other limitations. Of course the Mantle will eat some resource, but you control how much. This will lead to predictable performance in any scenario, just like on the consoles.

That makes sense. Like I said before, I'm willing to support Mantle if it's as revolutionary as is being suggested, and more importantly, if it's made into a TRULY open standard, with collaboration from Intel and NVidia.

It won't matter how good it is, if it's only going to be limited to the relatively small percentage of GCN users.. How many checks is AMD going to be able to write to entice developers to implement a Mantle code path in their games when they are already cash strapped?

 

[zlatan]

I see. BTW, I have a question for you zlatan. Is it possible to have multiple rendering threads using IC rather than DC? There are some efficient engines like the CryEngine 3 which don't use DC, but still manage to extract high performance from multicore processors.

Yes it is possible. It need more work and more testing, but in the end the results will be better.
A CryEngine is an engineering masterpiece. I don't saw the source code, but I see that they do many extremely tricky things, and they gain speed of it. There are engines that do some things better the CryEngine, but overal I think Crytek has the best solution for the developers who don't write an own engine.

How many checks is AMD going to be able to write to entice developers to implement a Mantle code path in their games when they are already cash strapped?

They don't pay for them to use Mantle. Some developers ask it, they have it, and for now they use it. As far as we know all Frostbite 3 games will support Mantle, so from EA DICE, Ghost, BioWare, PopCap, Criterion Games and Visceral Games are on the board now. Nixxes also support it, so almost all Square Enix games will use Mantle in the future. Rebellion, Oxide, and RSI announces their support too. In fact Chris Roberts (from RSI) also want to create a "close to the metal" support to the Intel and NVIDIA hardwares. You can read about it here. I respect him for this, and I hope Intel and NVIDIA will help this project.
And in the future there will be many new products where Mantle will help a lot. Oculus VR for example.

 

[smackababy]

Do we have any real Mantle demos yet? Is BF4 update out yet? Any real data out yet or just more promotional slides and talk about how great it is?

 

[Stewox]

Deferred context is a massive failure. Mantle will provide a much better way for multi-threaded rendering, because the command buffers can be re-used. This is how deferred context should work in DX11, but MS choose a very inefficient implementation to ensure backward compatibility.

It's all a set of tradeoffs, while the average people call it "downsides" or "cons", places like wikipedia. (but im sure it's still a stub there, im just not interested in wiki rules and arguing so I didn't contribute much yet)

Those developers who are going to use mantle don't actually care about the downsides because the upsides are well worth it and that's what they're looking for.

I'm sure when they were asking for it they were hoping it would work on more than the 5% of the market GCN represents.

I despise the usual lazy corporate business idea of doing something if there is a demand for it from consumers, kind of "developing for existing market", I rather build something so the market comes to it when the thing is done, that is real innovation, that's exactly what Mantle's going to do.

While I would point out your referenced link is 2 years 8 months old, and therefore way out of date, I also noticed this fascinating little gem tucked away near the end of Ryan Smith's article:

"Coincidentally, last month's interview with AMD's Richard Huddy at Bit-Tech also has a lot in common with this. AMD says DX11 multi-threaded rendering can double object/draw-call throughput, and they want to go well beyond that by bypassing the DX11 API."

This from 16 March, 2011. Looks like AMD had a Mantle twinkle in it's eye from shortly after they won the PS4 and Xbox One contracts.

I have pointed that out a few times elsewhere as this is a big proof point against the inexperienced users talking how "no developer will adopt mantle, it's doomed", you are the first one I saw that mentioned this by him self, thank you for having a brain compared to the masses out there.

Microsoft is requiring the developers to use it's Xbox One custom DX API instead of direct to the metal PS4 style for a reason and I doubt it's to benefit PC gaming as a whole.

I heard by cutting off some hardware access Microsoft can keep the clocks locked and can increase them at a later date, I knew about this trick before, however, I just happen to be one of those who can see ...

... which means, total speculation, but this thought makes a bit more sense now ... NSA Backdoor.

 

[Paul98]

Is there any official word on if EA halting other BF4 development till main game is fixed is delaying mantle on BF4? I would assume those working on mantle are a different team than what is dealing with the problems with BF4. Seems like we would have heard something about mantle if it was still on track for this month as there isn't much time left this month.

 

[smackababy]

Is there any official word on if EA halting other BF4 development till main game is fixed is delaying mantle on BF4? I would assume those working on mantle are a different team than what is dealing with the problems with BF4. Seems like we would have heard something about mantle if it was still on track for this month as there isn't much time left this month.

I would certainly hope that the issues being fixed in the DX version of BF4 are making it into the Mantle build, otherwise, it would have to be delayed further. Perhaps, the delay is from merging fixes? I haven't heard anything, so that is just me guessing.

 

[Lonbjerg]

Do you think this is because the rendering API is superior on PC? Because that;s what this thread is about. It's not PC vs. Console.

Why refrase my question?
Show me a console game with better I.Q. than on the PC?

If not possible...I rest my case ^^

 

[Imouto]

Would be nice if you could actually produce numbers to support that.

Are you looking at current sales, or actual market representation?

Same can be said about your 5% claim.

 

[gorobei]

Do we have any real Mantle demos yet? Is BF4 update out yet? Any real data out yet or just more promotional slides and talk about how great it is?

if you watched the oxide video, you would know their initial alpha engine numbers and when the starswarm demo they showed is coming out for anyone to test and mod.

 

[Stewox]

I would certainly hope that the issues being fixed in the DX version of BF4 are making it into the Mantle build, otherwise, it would have to be delayed further. Perhaps, the delay is from merging fixes? I haven't heard anything, so that is just me guessing.

Issues pertaining GFX problems in the existing version will not even be present in Mantle to begin with. The first mantle release might not be perfect so it'll have it's own issues, but this is Mantle, they can fix those very fast, you don't need to wait for a driver update from AMD that takes 2 or more weeks.

All other fixes obviously carry over since Mantle does not affect the rest of the engine.

There might be some little differences if they deem to be important enough to cause mismatches under certain conditions they will separate the patching process, but that's not really such a problem since mantle version will have to get it's own updates for the opimizations anyways, Oxide had some small changes to the rest of the engine, but we don't know if that affects gameplay to a level that it would be treated as 2 separate games.

 

[blastingcap]

Would be nice if you could actually produce numbers to support that.

Are you looking at current sales, or actual market representation?

Current sales, but even if you are looking ONLY at installed base, GCN was already more than 5% of the market.

http://store.steampowered.com/hwsurvey/videocard/

7850 - 1.25
7770 - .99
7870 - .81
7950 - .74
7970 - .70
7750 - .51 (inferred by the DX11 market share)

= 5% which doesn't include APUs with GCN (especially important for mobile devices and the upcoming Kaveri looks particularly interesting), nor R7/R9 GPUs. Also, I think the SHS undercounts the average GPU strength of gaming rigs, since SHS counts so many Intel iGPUs, and because so many people have secondary HTPCs or kids' rigs or whatever that get old GPUs thrown into them. Also, keep in mind that Mantle is still in infancy; as time moves on and Mantle exits beta stage, we'll have more and more GCN GPUs as a percentage of the gaming rig pie. Those SHS data are from LAST month, so this month's percentage is likely higher just from the passage of time alone.

 

[Stewox]

Also remember that these 5 or more percent are the vocal group of people who will make their decisions and opinions visible on the webs, compared to the all the other IGP percentages that don't really count as gamers ...

So I never like to look at situations based on some corporate graphs of percentages that business people like michael pachter loves to throw around.

 

[nenforcer]

Also remember that these 5 or more percent are the vocal group of people who will make their decisions and opinions visible on the webs, compared to the all the other IGP percentages that don't really count as gamers ...

So I never like to look at situations based on some corporate graphs of percentages that business people like michael pachter loves to throw around.

http://www.businesswire.com/news/home/20131120005559/en/Intel-Nvidia-Raised-Shipments-Levels-Quarter

Here's the latest Jon Peddie Research data which is generally the graphics industry standard. It's much more representative of the industry as a whole since people running the Steam Hardware Survey are generally gamers / enthusiasts which is actually a relatively small percentage.

While everyone's sales volume has decreased due to the PC industry in general declining, AMD still maintains a 20% global marketshare.

Radeon HD 7000 GCN cards have been on the market since Late 2011. I think it's fair to say that the GCN market share is much greater than 5% considering that GCN cores have been in their APU's ever since Trinity hit the market a year ago.

There are probably next to no one purchasing Radeon HD 6000 (VLIW5?) series graphics cards at this point.

 

[stahlhart]

If you're all done talking about Mantle now, this thread is going to be closed for good. Stay on topic if you want it kept open.
-- stahlhart

 

[VulgarDisplay]

I'm just really curious about what mantle will bring to me. Currently with a single 7970 in BF4 I play on low settings on purpose (gives you an advantage) and I get around 150-200fps at 1080p. Will mantle get me 500fps?

Doesn't seem like the game to test it out on, and I don't expect the performance impact to be that ground breaking in an FPS. In an RTS with a ton of units on screen it could be a game changer though.

 

[blastingcap]

I'm just really curious about what mantle will bring to me. Currently with a single 7970 in BF4 I play on low settings on purpose (gives you an advantage) and I get around 150-200fps at 1080p. Will mantle get me 500fps?

Doesn't seem like the game to test it out on, and I don't expect the performance impact to be that ground breaking in an FPS. In an RTS with a ton of units on screen it could be a game changer though.

One dev said 20% is not unrealistic, though it varies depending on the exact kind of game and what bottlenecks it has. So I'm guessing 10-20% on average, with extreme cases of zero and say, 50%. Similar to your RTS comment, what I'd really like to see is better massive simulations like that space RTS video above, or a true sequel to Rome Total War where you have WAY more details and individualistic traits and huge numbers of soldiers and such rather than lots of clones and small unit counts.

 

[VulgarDisplay]

One dev said 20% is not unrealistic, though it varies depending on the exact kind of game and what bottlenecks it has. So I'm guessing 10-20% on average, with extreme cases of zero and say, 50%. Similar to your RTS comment, what I'd really like to see is better massive simulations like that space RTS video above, or a true sequel to Rome Total War where you have WAY more details and individualistic traits and huge numbers of soldiers and such rather than lots of clones and small unit counts.

That's the thing, that RTS video would likely not run on Direct X at all, so how can you quantify the percentage increase over DX?

 

[gorobei]

I'm just really curious about what mantle will bring to me. Currently with a single 7970 in BF4 I play on low settings on purpose (gives you an advantage) and I get around 150-200fps at 1080p. Will mantle get me 500fps?

Doesn't seem like the game to test it out on, and I don't expect the performance impact to be that ground breaking in an FPS. In an RTS with a ton of units on screen it could be a game changer though.

the initial wave of mantle uses are likely to be purely number gains(more polys, more units, more blades of grass, more frames per sec, more explosion particles, etc). you will probably be able to turn up your quality settings and maintain your 150-200fps.

the dx batch limit elimination means that more cpu time is free for other functions. you could get better animations thru conditional contraints, better deformations thru morph targets, better collision detection, or fuller hair and cloth simulation.

since you are now sending the gpu all the data it can handle and spending less time memory managing, you get the choice of using those previously idle gpu cycles for more of the same or more compute stuff. the oxide guys and some of the other devs in the after session indicated that you had just tons of time where the gpu was waiting for cpu threads to finish sending data. Grenady also says you can offload specific compute tasks to addtional gpu or the apu. so all the hair and cloth sim or any non immediate calculation could be run on any spare compute units.

the really radical stuff probably wont happen until the second generation of game engines using mantle come out. the current engines are still using a lot of rendering pipeline code designed for dx9. even the oxide engine only recently switched to a less rasterized motion picture type pipeline.

 

[brandon888]

where is the mantle ? or time to sell my 7950 ? ;/

 

[ShintaiDK]

where is the mantle ? or time to sell my 7950 ? ;/

The first Mantle patch should be in late december. But they are awfully quiet about that and still no performance numbers. And the game to carry the first patch is having anything but success at the moment.