Mantle API presentation by AMD, DICE and Oxide - AMD Summit 2013

[TheoneandonlyMrK]

Multithreading is a challenge for all developers, always. Games are especially difficult to multithread because most of it has to be symmetrical.

NVIDIA has a reputation for half-assing open implementations and pushing their own, proprietary designs. AMD knows that so they're not going to get all lofty with expectations. NVIDIA would rather reverse engineer it and sell it as their own product than cooperate with anyone else for an open standard.

I'm not interly convinced of that take unified memory in win 8.1 for eg nvidia intel and the whole hsa foundation are prepping and moving it along and don't think they are all doing it the same way.
It might make sense for nvidia to make there own mantel cutting all but two apis out and limliting other mobile space opportunities for eg

 

[Frag_Maniac]

Multithreading is a challenge for all developers, always. Games are especially difficult to multithread because most of it has to be symmetrical.

NVIDIA has a reputation for half-assing open implementations and pushing their own, proprietary designs. AMD knows that so they're not going to get all lofty with expectations. NVIDIA would rather reverse engineer it and sell it as their own product than cooperate with anyone else for an open standard.

Yes I know MT has always been a challenge, but we HAVE seen more and more devs use at least quad threading, it's getting fairly common. What I mean is I don't think the AMD driver dev would have even mentioned it unless there were yet another level of learning curves involved MTing with Mantle, esp regarding all the bindings and assignments he was referring to. Anytime you have lots of options in an API, the parts that were already tricky are going to get even trickier. In other words Mantle is going to require an extensive guide just to learn everything that's possible with it. It will take yet another level of learning, much of it hands on probably, just for the MT part.

LOL, yeah Nvidia WOULD stoop to covertly reversing it and selling it as their own, but you have to wonder if it has THAT much success to begin with, AMD could do a LOT of damage as far as market share before it even comes to that.

I've yet to watch the DICE and Oxide presentations because I'm not keen on registering on that site just to do so. I might wait and see if someone uploads it to YT. If not I'm not going to worry about it. I already heard pretty much everything I wanted to know anyway. The other presentations are only going to be more game specific.

 

[FordGT90Concept]

If Mantle optimizations can be used in all titles created on the latest Frostbite engine with little/no work from the title devs, it'll be very popular. That's why DICE is important here: EA is using DICE's Frostbite engine as much as possible.

 

[RoutedScripter]

Noteworthy stuff from heise.de interview (found by neogaf user Datschge)

- Effort to have console-like access and programmability on PC started about 5 years ago. Spoke to different companies including Nvidia and Intel.
- Respect for AMD being the sole company to realize his suggestions.
- Yearly meetings, long discussions. Mantle code was started one and a half year ago.
- Project was internally treated as top secret, similar to Eyefinity.
- Mantle is *not* a console-like AMD interface for GCN graphic chips. It allows finer grained access to the GPU. Too specific cases can be realized through extensions. Possible for other manufacturers to support Mantle.
- DirectX 11 compatible doesn't automatically mean potentially Mantle compatible. GPU architecture needs to fulfil specific requirements. Nvidia's Kepler should be able to do that.
- Wants to see Mantle everywhere (re smartphones, tablets, mobile, Linux and Mac OS).
- Still too early on how much faster Mantle will be over Direct3D.
- Porting console games using Direct3D may still be faster, Mantle may take longer but helps doing it well re performace, level of details and robustness.

Initial quote from Project C.A.R.S programmer:

“However, there isn’t any doubt in my mind that Mantle will solve the poor implementation of multi-threaded rendering in the PC space, more so for AMD since they don’t even support Driver Command Lists in DirectX! That alone will get them to 70-80% of next-gen console performance in terms of submitting draw calls to the hardware. When DICE reveal their Mantle version in mid November, you’ll see scalability that’s unheard of in the PC space e.g. some 6 or 8 core demo with submission performance many times that of DX, with lots of pretty charts.”

I've said it before, Multi-GPU people will be banging their heads all day long, out of joy.

Exactly It all comes down to seeing it in action not just presentations anyone remember the guys at Euclideon who claimed Unlimited Detail? You know with all the interviews and videos and bs yet people still loved every minute of it? Oh yea no one has heard a damn thing from them in nearly 2 years with even their latest stuff still not up to par to regular plain old traditional techniques

Unlimited Detail Real-Time Rendering Technology Pr...

Until Mantle is actually used and shows this 50-150% increase its pretty much just PR marketing speak and vague promises.

UDT wasn't a proven thing. This is. And it's not even a fair comparison, just something that promised "boost".


The performance boost for BF4 initial mantle demo release will most likely be at it's lowest, like 50% let's say, maybe higher if they manage to do it by december. But I hope it's a bit better to get the mainstream people talking, if it's not enough then mainstream will jump to conclusions and think that it's all Mantle's and AMD's fault and this thing would be as good as dead.

Next is CPU, significant improvements, and most of all, proper multithreading.

Multi-GPU, those charts can go as high as ~300% faster than DX, for Quadruple setups.


As I said, the problem with these numbers is, there would have to be a demo in same visual spec with DX and with super mantle implementation, but this will evolve over time, so these increases will come gradually and most people won't notice how much it has increased, also, developers using available performance for even more visuals, it won't have such a big apparent impact if you don't pay attention or track these things, I hope some GFX benchmarks out there make a controlled environment for comparison.

This is real deal analysis, this is the kind of info investors pay money for, too bad Pachter doesn't really know much about tech, this thing can take of so big it might cause nividia's profits to get a chainsaw just like another guy said around here.

So, their own words, 300% faster ... I might be totally wrong, I'm just trying to not bee too speculative, but deep down I think it's going to be a ton more than anyone expects.

It's also good to note that the 300% number there is the maximum, not average. The maximums will be higher since with proper control of the hardware the ALUs will utilize them 99%, not 50% of it being idle like it's now.


Also I have a big feeling that G-Sync might be totally obsolete for high-end markets, it already is obvious it's not targeted for high-end, it's for the 60hz monitor people with unstable FPS. I have a 144hz monitor and I have no problems with tearing since the tear probably appears too short of a time for the eyes to detect it, so it has no apparent tearing in practise, but I don't even care about visuals that much anyways.

Because, the high ends will go into multi-gpu setups, which means they definitely have enough money for 120-144hz monitors, even if you have 60FPS, just having a faster monitor practically makes tearing obsolete, but I'm still not an expert on tearing since I don't really care much about it unless it's super noticable, like on a 60hz monitor, I noticed it immediately when I tried it temporairly (not my own monitor), horrible feeling, had to enable Vsync for the first time in my life.

I really have to point out some clash of interests, I think there are some people trying for what-the-fick ever reason downplay Mantle, and even Carmack him self talking that he was pushing for G-Sync for years, it's totally clear he's humped on Nvidia, they're bringing up these "issues" of , I don't care about many APIs on the market, I care about the idea if you buy 2000€ machine it should be 10 times faster than a 300€ (vague but for an example) and whatever people use what they want and they please leave us the heck alone, unity fanboys are the biggest trolls, yes you can have it, make games however you like, just stop bothering us, we're not taking anything away from you.

All the idiots who said that Mantle won't bring advantages to linux, when mantle is the TOP thing that can basically set Linux and the whole PC platform to a giant golden age, this is what Linux needs, Linux doesn't need some company to release montly OGL extensions, it really is a rare occurance I don't agree with carmack and I mean that strongly, Oculus Rift is awesome stuff and I've been excited all the time, but look, OGL extensions don't make it full control, he didn't really explain much what nvidia's OGL extensions did that much good, it's still OGL, same principle and way things are don, still having to be inside the broader paradigm of devices, so Mantle is high-end only and that's how it should be, spin off them, they're too slow, the progress of software catching up hardware has been long too slow because these APIs hold us back, not any more. FFS

 

[RoutedScripter]

Perhaps we should all wait till December and make judgements based on whats actually released.

Nothing wrong with being informed and closer to everyone else. Similarly how half of the stories on mainstream media has been known in alternative media for weeks, months sometimes a full year.

It's AMD applying console techniques to PC. They're giving a direct path to developers to code for the GPU itself which eliminates a ton of overhead but at the cost of compatibility. It's the same reason why consoles can achieve so much with so few resources.

Exactly, it's just a trade off, get rid of some compatability, get rid of some stuff that is acceptable to sacrifice in return for top performance. That's all there is to it. All these people whining about compatibility, don't understand, that I don't really give a damn about it to be honest, I want top peformance, so the Crossfire people can actually use what they paid for, the top PC games will be there, it's the developers them selfs that wanted Mantle in the first place.


So, to make an Mantle engine it will take more time than it takes to build a DX renderer. But, since it's closer to how console renderers are developed, and with the bonus of transferring carefully created optimization code for GCN, the ports will be done so much faster for the time being, this bonus will ofcourse end when PC GPUs moves to new architecture.

AAA studios don't really care about the little bit of more cost and time it will take to build Mantle Renderers.

Once the mantle renderer is complete it can be improved on top, also there is much less time spent on fixing bugs post-release, because these have been eliminated in the first place (fine control).

Multithreading is a challenge for all developers, always. Games are especially difficult to multithread because most of it has to be symmetrical.

NVIDIA has a reputation for half-assing open implementations and pushing their own, proprietary designs. AMD knows that so they're not going to get all lofty with expectations. NVIDIA would rather reverse engineer it and sell it as their own product than cooperate with anyone else for an open standard.

Providing low level api is nothing of a secret they would have to reverse engineer other's implementation, plus, AMD's initial thing is GCN-only, Nvidia wouldn't benefit that much, they would pretty much save time by doing it from scratch on their own.

This is an interesting one to point out, this approach claims to revolutionize PC scaling!

More slides here:
http://www.mediafire.com/?756rc480otk58af
http://www.mediafire.com/?s4h21sjyvotdtd1
http://i.imgur.com/nt5217p.png

Oxide Slides: (for some reason they're not on official PDF, so i had to screenshot them my self)

Spoiler

Actually, people should see all these slides before they even talk here (i've watched all 3 presentations before even making this thread)

sorry for double post agian

 

[Steevo]

That's all incredibly vague. I'll believe it when I see it, and see it, I do not.

Somebody provide me with hard facts, then I'll become a belieber in Mantle.

They already do this with close to metal applications written in native code for consoles. Add in the fancy tessellation and other new hardware features and 90% of games is just setup calls and re-instancing of objects, T&L is already done at the hardware layer so no penalty for that.

 

[Frag_Maniac]

If Mantle optimizations can be used in all titles created on the latest Frostbite engine with little/no work from the title devs, it'll be very popular. That's why DICE is important here: EA is using DICE's Frostbite engine as much as possible.

Oh it will no doubt be used on all Frostbite titles. AMD are deep in bed with DICE now, whom are owned by EA, and literally every Frosbite game thus far and in the foreseeable future is under the umbrella of EA.

@Ruski
"AMD's initial thing is GCN-only, Nvidia wouldn't benefit that much, they would pretty much save time by doing it from scratch on their own..."

I was going to say this, but after I mentioned AMD half heatedly mentioned possible competitor support, I wasn't sure if he meant AFTER Nvidia used it in a build version modified to work with their architecture. I see only 3 scenarios, a version modified to work with Nvidia GPUs, Nvidia changing their architecture somewhat to make use of it, or building their own API, and the 1st two aren't very likely.

The 4th scenario of course is Mantle not catching on and everyone sticking with DirectX, which may be what Nvidia is hoping for. They've got a good thing going with much of their fanbase convinced we need buy expensive GPUs to stay ahead of the curve. Funny how I used to think AMD were the ones lazy (or incompetent) with software, but now it's looking like Nvidia are.

 

[DannibusX]

The biggest cockblock to Mantle is Microsoft. They won't allow it on their console, their console is gonna sell like hotcakes and developers/publishers will target it using DirectX. If it's going to succeed, they'll need help from Sony, Valve and Nintendo so more developers can target those platforms with lower development costs. Microsoft may have to change their tune after that or get shitty PC ports.

But they need to prove it works first. When it comes to BF4 we'll see. I'd love to see an open API take swings at DirectX, the best part is that AMD made it open and not propietary.

 

[HTC]

The biggest cockblock to Mantle is Microsoft. They won't allow it on their console, their console is gonna sell like hotcakes and developers/publishers will target it using DirectX. If it's going to succeed, they'll need help from Sony, Valve and Nintendo so more developers can target those platforms with lower development costs. Microsoft may have to change their tune after that or get shitty PC ports.

But they need to prove it works first. When it comes to BF4 we'll see. I'd love to see an open API take swings at DirectX, the best part is that AMD made it open and not propietary.

Agree.

As for the open part, it obviously should work better on CGN cards then on any other technology's cards so, in a way, it's still proprietary but it seems not closed to others like PhysX most definitely is and that, IMO, is something that just might make it succeed.

 

[Frag_Maniac]

The problem is, when one of the only two chip vendors for GPUs makes something "open", it really means, "OK, would you like to play our way?" The answer of course is typically "No thank you, ours is better", even if said competitor doesn't HAVE a version of their own way yet.

In this kind of scenario it generally takes a war, the likes of which played out between Sony and MS when Blu-ray and HD-DVD were going head to head. As you said that war is only so powerful on AMD's console side when MS wants to block it from being used on Xbox One, which is understandable since they make DirectX.

Valve won't likely participate in Mantle with their Steam Machine either, since Nvidia is strongly winning on Linux compatibility. If the PC platform is a wash being split between AMD and Nvidia users, then it comes down to PS vs Xbox and Steam Machine.

 

[droopyRO]

Don't you see it, SteamOS (clean, superfast linux code) Mantle Oculus Rift = The second PC golden age.

With the gameplay we get these days no way, where are the flight sims ? where are the oldschool RPGs, tactical shooters, GTR2 and so on games that are dead and buried not by graphics, physics or sound but by actual complex gameplay.

 

[RoutedScripter]

Update: Some numbers - "secret demo"

I keep seeing another demo around the web, seems like a site called hardware.fr has all the slides that some people were using in their forums, but I can't find the video, I keep searching, it's not in the official AMD Summit page, I searched for all the tags, it's just not there, and it's the presentation with SOME NUMBERS!


Seems like it was kind of a semi-secret presentation or what, they have a Benchmark all up and running called Nitrous or Swarm or whatever it is.

Here are the slides:

http://www.hardware.fr/news/13450/apu13-oxide-fait-exploser-limite-cpu-avec-mantle.html


Preview (some of them i choosed):

Spoiler

The biggest cockblock to Mantle is Microsoft. They won't allow it on their console, their console is gonna sell like hotcakes and developers/publishers will target it using DirectX. If it's going to succeed, they'll need help from Sony, Valve and Nintendo so more developers can target those platforms with lower development costs. Microsoft may have to change their tune after that or get shitty PC ports.

But they need to prove it works first. When it comes to BF4 we'll see. I'd love to see an open API take swings at DirectX, the best part is that AMD made it open and not propietary.

Thankfully, PS4 seems like the favourite around there.



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

Browsing through all the webs i keep seeing noobs doubting mantle so much, most of them don't know tech who criticize this, they keep saying "I don't believe until I see reviews not paid by AMD" - that's so ridicolous and stupid, Mantle doesn't do anything on it's own, AMD doesn't have games or demos to show this, the demo is called Nitrous I think OXIDE will release in early 2014 for the PC crowd to test and then the benchmarks will come, it has nothing to do with AMD anymore, they will keep updating DX and OGL Drivers in terms of optimizations, not Mantle. Mantle will only get API revisions here and there which wouldn't do anything directly to any game at the time of release, it all has to be supported by developer, and the developers are now assuming position and job of driver developers as well.

 

[Hilux SSRG]

The biggest cockblock to Mantle is Microsoft. They won't allow it on their console, their console is gonna sell like hotcakes and developers/publishers will target it using DirectX. If it's going to succeed, they'll need help from Sony, Valve and Nintendo so more developers can target those platforms with lower development costs. Microsoft may have to change their tune after that or get shitty PC ports.

But they need to prove it works first. When it comes to BF4 we'll see. I'd love to see an open API take swings at DirectX, the best part is that AMD made it open and not propietary.

By the same token MS might prevent Mantle running well on Windows period. I'd rather see innovation from AMD and MS than limitations.

 

[RejZoR]

Trust me, MS doens't want to go that way, because someone would notice it and that would be baaaad for MS...

 

[Crap Daddy]

The biggest cockblock to Mantle is Microsoft. They won't allow it on their console, their console is gonna sell like hotcakes and developers/publishers will target it using DirectX. If it's going to succeed, they'll need help from Sony, Valve and Nintendo so more developers can target those platforms with lower development costs. Microsoft may have to change their tune after that or get shitty PC ports.

But they need to prove it works first. When it comes to BF4 we'll see. I'd love to see an open API take swings at DirectX, the best part is that AMD made it open and not propietary.

How is it open if it runs only on Radeons? Am I missing something or it works only on GCN arch?

 

[The Von Matrices]

Maybe someone can enlighten me, but I am skeptical of how Mantle can make that much performance improvement by reducing the CPU bottleneck. The original assumption is that GPUs are idle a portion of the time waiting for data from the CPU due to DirectX overhead. However, current GPUs (I'm looking at the 290/X in particular) are already throttling when handling DirectX code. If the GPU is sent work more frequently, won't it just throttle more thus reducing the performance gains significantly?

 

[Frag_Maniac]

Trust me, MS doens't want to go that way, because someone would notice it and that would be baaaad for MS...

Perhaps even legally bad, it would essentially be corporate sabotage to put a hook in the OS that intentionally cripples Mantle performance, and the creators of Mantle could easily dissect it and prove it's use to be suspect.

We live in a day and age where everything programmed leaves a digital signature, and even average users can spot copy and pasted content in video games (as was shown recently in the CoD Ghosts cutscene).

I don't think we've seen the end of MS' console customers forcing change either. If the PS4 shows to have vastly improved performance with Mantle, LOTS of Xbone customers are going to insist MS allow it on their platform.

If they don't comply, many might jump ship to the PS4, esp when they're already getting 720p in a lot of big titles where the PS4 gets 1080p. In the tech industry nothing speaks louder than a consumer boycott.

 

[RoutedScripter]

Maybe someone can enlighten me, but I am skeptical of how Mantle can make that much performance improvement by reducing the CPU bottleneck. The original assumption is that GPUs are idle a portion of the time waiting for data from the CPU due to DirectX overhead. However, current GPUs (I'm looking at the 290/X in particular) are already throttling when handling DirectX code. If the GPU is sent work more frequently, won't it just throttle more thus reducing the performance gains significantly?

Without mantle, the driver takes CPU time because the GPU needs to be spoonfeed the commands, it needs the CPU to keep commanding GPU what to do. With this they're able to command the GPU them selfs, making GPU almost independent from CPU stuff, the CPU doesn't have to keep telling the GPU to do draw calls so frequently I think, I got this but I forgot the details my self, I just know the CPU doesn't have to spoonfeed the GPU, and it removes the need for the driver threads to be in the CPU all the time.

Imagine a giant man that is 50 meters tall, this giant man has his arms hurt and cant move them, but he needs to eat, so there's one little man that takes care of him by climbing up a series of ladders to spoonfeed him the food, the little man has to go down for more food from a big big big pile of food, the little man can only carry 20 KG, while the giant man's mouth can hold 200 KG at once, the time it takes for the little man to climb down, grab more food, and climb up, is the API overhead on DX.

With Mantle, you make the giant mans arms work again, he can carry 190KG of food in one hand, and he's using both to eat all day long, by the time he eats one load, another load is ready, which means a very short delay between. That mans the giant man doesn't need the little man anymore. And then the little man can only focus on farms, has more time to operate even more farms than previously, to keep producing more food (ex. geometry, number of units, AI), the little man doesn't have to directly assist the big man.

The big point is how you'll get this, is that the big guy doesn't have to wait for anyone, and the little guy has way more time for something else that he had to do at the same time. That probably holds but the deeper I go I'm not sure even my self, maybe it's better if I use two, one little guy was working in farm the other was feeding, now both can work on farm with Mantle.


Also what do you mean with throttle ? The GPU isn't sent more work from the CPU, ofcourse it will do more effective of a job, the API overhead doesn't mean it's empty space there, it's still work and calculations that get done, but those are totally wasted into heat. DirectX is basically the home heating software, keeping your legs nice and warm.

 

[The Von Matrices]

Without mantle, the driver takes CPU time because the GPU needs to be spoonfeed the commands, it needs the CPU to keep commanding GPU what to do. With this they're able to command the GPU them selfs, making GPU almost independent from CPU stuff, the CPU doesn't have to keep telling the GPU to do draw calls so frequently I think, I got this but I forgot the details my self, I just know the CPU doesn't have to spoonfeed the GPU, and it removes the need for the driver threads to be in the CPU all the time.

Imagine a giant man that is 50 meters tall, this giant man has his arms hurt and cant move them, but he needs to eat, so there's one little man that takes care of him by climbing up a series of ladders to spoonfeed him the food, the little man has to go down for more food from a big big big pile of food, the little man can only carry 20 KG, while the giant man's mouth can hold 200 KG at once, the time it takes for the little man to climb down, grab more food, and climb up, is the API overhead on DX.

With Mantle, you make the giant mans arms work again, he can carry 190KG of food in one hand, and he's using both to eat all day long, by the time he eats one load, another load is ready, which means a very short delay between. That mans the giant man doesn't need the little man anymore. And then the little man can only focus on farms, has more time to operate even more farms than previously, to keep producing more food (ex. geometry, number of units, AI), the little man doesn't have to directly assist the big man.

Thanks for the reply, but I don't think you understand my question, unless you are telling me that the DirectX overhead causes the GPU to perform useless calculations while it is waiting for the next command. My concern is that even when a GPU like the 290/X is bottlenecked by the CPU right now, it is already hitting the power and heat limits of the silicon. So even if the GPU has as much data as it wants to process, it can do any more because heat and power are constraining it already. In addition, having reduced CPU load is great, but games still don't use all the cores available on modern CPUs so I don't see as much benefit in reduced CPU load.

To use your metaphor, it would be as if the giant still has a 200kg capacity mouth but can only eat 18kg of food at a time before he gets a stomach ache. The little man could satisfy him completely in the past. The only difference is that now, the giant can feed himself even though he can't consume food any faster. In addition, the little man is free to do whatever he wants; however, the town has more workers than jobs and now the little man is unemployed.

It doesn't fit completely because reduced CPU load means less power usage (which is good) versus unemployment (which is bad), but there still aren't enough jobs (threads) to satisfy the workers (CPU cores).

 

[RoutedScripter]

Thanks for the reply, but I don't think you understand my question. My concern is that even when a GPU like the 290/X is bottlenecked by the CPU right now, it is already hitting the power and heat limits of the silicon. So even if the GPU has as much data as it wants to process, it can do any more because heat and power are constraining it already. In addition, having reduced CPU load is great, but games still don't use all the cores available on modern CPUs so I don't see as much benefit in reduced CPU load.

To use your metaphor, it would be as if the giant still has a 200kg capacity mouth but can only eat 18kg of food at a time before he gets a stomach ache. The little man could satisfy him completely in the past. The only difference is that now, the giant can feed himself even though he can't consume food any faster. In addition, the little man is free to do whatever he wants; however, the town has more workers than jobs and now the little man is unemployed.

It doesn't fit completely because reduced CPU load means less power usage (which is good) versus unemployment (which is bad), but there still aren't enough jobs (threads) to satisfy the workers (CPU cores).

I've answered it in the end I think. I was explaining the difference in beginning.


1. The API overhead is not "untouched potential", it still is there producing heat and taking power, with mantle that heat won't go to waste and you'll see much better FPS or better visuals, that's it.
2. If the GPU will be pushed to that high limits it wasn't designed for (previously untouched potential by making ALUs not idling), you will have to enable manual fan control, a total non-issue.

You have that "can't consume as fast or anymore" completely wrong, you don't suppose to rely on automatic fan controls to make your GPU cool enough to work. Remember Starcraft 2 frying GPUs, it's not the game's problem, it's the GPU Fan control problem, most of the avreage PC gamers have automatic settings, enthusiasts and experienced users never use automatic anything, I'm using manutal fan control since as far as I know back, in the old days of XP and Radeon 9600XT, I used Rivatuner, but then manual-fan control came with Catalyst control center and been using from there ever since.

If I think this is how it will work, with the idle ALUs issue, this is all speculation, but if I think it works like how I think currently, there could be a dangerous situation of many people trying the new Mantle Benchmarks and frying their GPUs because they had auto-fan settings.

But the only way you can be worried is, if you manage to heat up the card that even 100% constant fan speed won't keep it cool, then that's a problem, would at least create more interesting GPU cooler aftermarket.

 

[The Von Matrices]

I've answered it in the end I think. I was explaining the difference in beginning.


1. The API overhead is not unsused space, it still is there producing heat and taking power, with mantle that heat won't go to waste and you'll see much better FPS or better visuals, that's it.
2. If the GPU will be pushed to that high limits it wasn't designed for (previously untouched potential), you will have to enable manual fan control, a total non-issue.

Okay, thanks for the clarification. I still am curious if processing frames in Mantle requires more energy than waiting for the next DirectX call. If so, then the performance improvements might not be as large because the GPU would throttle more. I also don't agree that enabling manual fan control is a non-issue; from the 290/X reviews it is clear that there are a lot of us (including me) have moderate limits on how much noise we find acceptable. I'm sure this would still be a net gain in performance and efficiency, but the overall performance gain might be diminished by power and noise constraints.

The question that comes up for me is that if the CPU is such a bottleneck then why does graphics overclocking have any effect today? Sure, if you want 1000 frames per second the CPU will become a bottleneck preventing any more performance gains, but if you're shooting for something more moderate like 60 or even 120fps, then graphics overclocking can allow you to turn up the detail settings to use that extra performance. Why is the CPU not a bottleneck there?

 

[xenocide]

I don't buy it. AMD started drumming up the "down to the metal" parade a few years back and several companies immediatly said they were wrong, including Crytek (they know a thing or two about high level PC Graphics). They have said many times (and many ways) that what AMD was advocating was nice in theory, but wasn't feasible.

I'm also ashamed that not one person while mentioning the bloat of DirectX mentioned that it primarily applies to DirectX 9.0c and earlier--but not DirectX 10/11. With DirectX 10 they completely gutted the API and reworked it to be more efficient. But you know what the huge problem was? Nobody made games for DirectX 11 because the last generation of consoles were DX9 only. It's like pointing out that games don't optimize for the large amount of RAM these days, when it's mostly due to the fact that despite most people having x86-64 compatible CPU's nobody makes games that support that. Hell, if game devs were really so hindered by the evils of DirectX, how come nobody is using OpenGL? Sure Sony does a bit and iD, but what about Crytek, and DICE, and Rockstar? The reason is because DirectX isn't the huge bucket of crap it's being made out as.

How has nobody pointed out his absurd notion that somehow a software API is converting heat and energy into processing power? That just doesn't happen (well, in Graphene it apparently does but these are Silicon-based). If that were the case someone would have already done it back in the 90's or even earlier when they started focusing on things like heat-generation and power usage. I'll point out that in the world of engineering nothing is 100% efficient. You can get damn close, but there are always inhibiting factors.

Show me a video of side-by-side gameplay with a game running on DirectX and on Mantle, using the same hardware and same settings, and maybe we'll talk. Otherwise, it's all bullshit. In case you forgot AMD Marketting is really really good at that. Remember when Bulldozer was supposed to use less power than anything Intel offered and perform on par or 10-15%+ better than the highest end i7? I sure do, and I'll let you look up what actually happened because it was nothing of the sort. I'm not being duped by more marketting crap, and until AMD shows more than slides or some reps for companies they have "reimbursed" handily, I could care less about this tech--which by the way, won't be used for Xbox One or Playstation 4.

 

[RoutedScripter]

Okay, thanks for the clarification. I still am curious if processing frames in Mantle requires more energy than waiting for the next DirectX call. If so, then the performance improvements might not be as large because the GPU would throttle more. I also don't agree that enabling manual fan control is a non-issue; from the 290/X reviews it is clear that there are a lot of us (including me) have moderate limits on how much noise we find acceptable. I'm sure this would still be a net gain in performance and efficiency, but the overall performance gain might be diminished by power and noise constraints.

Okay, fair points that technically are correct. But I don't think you understand what enthusiast PC market is, we don't care about little issues such as fan noise, and even if it would be a big issue, it would be a temporary one once GPU vendors would be forced to add better coolers by default,

Nobody cares about this, seriously, I'm sure it won't be as bad as you are afraid, but just please don't make a fool of your self and whine about this to AMD or DICE, it will just create all the other noobs and G-Sync fanboys to join the failtrain about complaining something that is not an issue in the hardcore space, even if it gets super loud, I don't care, I can sacrifice that for the performance glory

As the presentation said, Mantle isn't for everyone, if you are not

These kind of things are petty issues, we call them petty because

Most PC enthusiasts always disable all throttling and power-saving features anyways because it interfeers with overclocking, so there you go
But I don't overclock and I have it disabled anyways because it's annoying and sometimes makes stuttering in various games, this is just experience, but I don't have time to test if these things improved in years, I just built a new PC

I can explain more if you really want, although I think there is nothing more to say really, but I just advise you to not start making this into a big deal, you won't be taken seriously with this, I don't know of a PC enthusiasts that would be a noise-freak at the same time, maybe there is someone but they most probably have water cooling in that case if they're so annoyed, but these kind of issues never get discussed, you don't see overclockers talking about oh my "fan noise is loud". It's all a set of tradeoffs, we trade off every other thing for TOP PERFORMANCE, everything else is ACCEPTABLE SACRIFICE.


Acceptable Sacrifices:

• Fan Noise
• Compatability
• Wide support
• A bit longer development time (more code)
• Requires more dev skills
• etc.

That looks like right doesn't it. No, these are all temporary sacrifices, Mantle is made for AAA market, which means, it only takes a few of those AAA developers to join the wintrain, you don't suppose to look at the gaming as a whole and just compare mantle to all kinds of other markets, because you need to understand we don't really care about them, they're IRRELEVANT, a game that doesn't require much GFX or any kind of hardcore performance doesn't have to use mantle and I don't care if they whine or not, I care about the stuff that would really SHINE with Mantle like the original CRYSIS, that's what I want to see super-optimized and improved.

This is a very interesting talk here, I need to break all this down, it's important to distinguish groups and interests, all these people's opinions who attack Mantle outright, can have an agenda behind their opinion which isn't compatible with actual consumers in the PC enthusiast space, you need to keep an eye on those and expose them, the biggest of them all are bloggers and twitter junkies who self-proclaim them selfs as market anaylzers to be the policemans that decide what everyone in the market does is right or wrong, they care about stuff like "development costs" - Would would a random stupid casual game programmer care about development costs at DICE or Crytek ? That's not his problem, and not even his job to care about and judge Mantle and it's potential around that silly argument. It's so funny I though back then in the initial release it's going to be like 50% or more, that's the highest I went speculating at the time, no numbers no nothing, just LOGIC and BRAINS.


This is quite deep stuff, it's an attack from the mainstream to the idea of PC enthusiast market spinning off, It is my duty to defend the position! And those people whine because they're either business people who always look to make easy money as fast as possible, OFCOURSE they don't want to have many APIs in the market, but it's not my problem, it's not our problem who want to have a few great games that you can finally use 4 GPUs in and have 4K at 120FPS or something, at this point Carmack's opinion really is irrelevant, he isn't even working on any game at all, I don't know what the hell is going on with Doom4. This is what PC needed all the time, spin off the mainstream, get off that failtrain, why do we have to keep limiting our selfs because 10 million LoL fanboys are okay with DX, finally, I might be strongly yelling here but please this is all with a reason.

What you people should be doing right now is yelling to the game developers to support mantle, benchmark makers, that will fix the compatability and wide support issue faster, that will establish who's on the side of hardcore PC gaming, and de-associate them with the mainstream hopefully, that's all you people out there need to do, no point to keep wasting time attacking mantle and questioning AMD about the numbers, AMD doesn't have anything, unless provided demos provided by the developer, but I already cleared that up it's called Nitrous and it's releasing in Q1 2014, and it's NOT made by AMD.

It's not like the sound is damaging your ears, I would understand if it was some big machine and you wanted to protect your ears.

Also keep in mind I'm generally speaking, I'm not replying to anyone directly in this thread particular when referring to the mantle-attackers, I'm referring to how it is around the webs. So don't take any of this personally.

The question that comes up for me is that if the CPU is such a bottleneck then why does graphics overclocking have any effect today? Sure, if you want 1000 frames per second the CPU will become a bottleneck preventing any more performance gains, but if you're shooting for something more moderate like 60 or even 120fps, then graphics overclocking can allow you to turn up the detail settings to use that extra performance. Why is the CPU not a bottleneck there?

That's because CPU doesn't have to worry about the GPU, the CPU only focuses on AI, geometry and all the scripts, the application requires besides graphics. And those extra details don't add that much to the CPU that it would cause the same slowdown, with the proper multithreading the CPU utilization goes even higher now.



Now that I've also seen these numbers, a bit of a sneak peak, I can't stop to think the 50% is the minimal this thing can improve GPU performance initially if they manage to do it good, ofcourse that 50% or more on the CPU side is basically guaranteed.

I don't like to speculate, but this might super-charge the PC industry right in time against the consoles and steambox, I don't mind steambox being PC hardware and linux, but I don't want steambox to steal PC glory with the living room and controller , basically I don't really know what's the point to play PC games with a controller, I am not fond of multi-platform games that simply throw in whatever controls they can without the actual , and then the whole game has to be designed with the worst I/O in mind, it's basically a clusterfuck and to be honest, I don't really like valve that much, they just keep piling cash and they could have do better, while they keep doing dota games to reap the moneys because they're too lazy to come up with something original.

Even with the visions that take off in the mainstream, it's made for mainstream, it has nothing that enthusiasts can really take, so how does Valve support PCs, yes it did with steam, but are they going to take it away now by making a console that's not really a console, something thrown together.

But I have faith and confidence that the established PC people won't be swayed so by trendy fads so easily, it will mostly catch the LoL noobs, the piles of them which we don't need anyways, they didn't do much of a visible impact for the PC space other than having numbers to offset and make it look like PC has the upper hand, but those aren't permanent numbers, they don't build DIY PCs and buy the hardware, they're just avreage gamers who are easily manipulated, so if LoL ends up on steambox which I don't think it will but let's say Valve goes apeshit and pays them to do it or something, , there are my subjective opinions, since I'm quite strongly anti-Pay2Win and all the other cheap F2P crap.

These large companies don't look for game design consistency, they don't really care about keeping some self-established rules that you can't forcefully port some game to a fucking controller if it wasn't designed, the gaming industry isn't as evolved yet, it behaves like a child, you'd never go tear up a good classic car and just throw something in and call it the best car of the year, that's bullshit, if you find a great old

Valve has been going down the slippery slope in social degeneracy for years, all their hype now is stupid TF2 hats for zombies and other childish bullshit. Whatever their HL3 is or something, if it doesn't , I'm sure everyone else can then rule Valve out of any relevance to PC gaming and consider them compromised. Absolute power corrupts absolutely, I think the boss is just sitting on his lazy fat ass all day long.

The PC enthusiast market needs new grown-up without PR bullshit business model that wouldnt require millions spent on PR and on screwing people around by witholding information because of their carefully crafted PR plan and using stuff like "FOV slider" as a PR hype material and other kinds of cases that make me absolutely sick, a good example is Star Citizen taking at least some form close to what I have in mind with the kickstarter thing and promising no pay 2 win among other things.

Okay that's enough ...

 

[xenocide]

...What on Earth are you even TALKING about at this point?

 

[RoutedScripter]

New Video: Q&A with Mantle Developers: AMD Mantle Q&A with Developers at APU13 - YouTube

Summary:

• Mantle can reduce memory footprint in some cases by inteligent and efficent use of memory management, but this will not be so noticable in games because the developer will use that extra memory for even more stuff.
• Mantle will enable developers to forsee resource constraints (how much a texture takes mem for example) no more guessing.
• AMD is taking Mantle debug tools very seriously, profiling, etc, PerfStudio is supported!
• The bottleneck with mantle could raise up very close to physical bottlenecks of the hardware, where it should be, but they aren't yet sure them selfs with initial tests to give numbers out.
• Optimizations and techniques will vary between developers (they pointed this out!)
• AMD Driver Guy admitted the driver team can only optimize "mediocre" they don't really know! (exactly what I was saying all the time, GPU Vendor driver people never worked on games, they don't know, their optimization is only game-specific driver hacks, not how it should be)
• Dice: Most consumers who will use Mantle will be GPU bound because they want to run on highest settings with top CPUs. (CPU Bottleneck should we way up there, WAY!)
• AMD:For example Extra CPU Peformance could be used for increasing "VIEW DISTANCE" which goes back to better eye-candy. (Crysis, GTA, etc)
• Fine grain tuning of Bindless _-_ will allow flexibility at which point to choose GPU or CPU performance priority.
• Initial Co-Supporting developers such as dice and oxide were instrumental in Mantle's design and making a proof-point for the industry, they didn't want the pool of initial developers to be too big.
• AMD: Other APIs were designed without the counseling and testing of game developers, for Mantle the design was a direct cooperation with many physical meetings and weeks of groundwork. paradigm shift
• Oxide: Mantle can bring back old game genres from 90' when you could do "all sorts of things", the games that didn't survive the 3D transition for example.
• Oxide: Some of the services stuff in DX API were never used because of extreme unpredictability the drivers will crash, Mantle should remove the PC customers the need to thinker with driver updates and Catalyst settings to make their game running.
• They cannot imagine what small developers are going through when doing PC stuff, all the driver bugs, etc, Mantle will make it easier to develop on PC for everyone.
• The whole current idea of driver hacks, new driver for every new game, etc is troublesome for not just the consumers but also developers, drivers are also getting very big because of the app-specific code they have for all kinds of games, getting to ridicolous sizes like 300MB (catalyst)

I hope you all just realize here, this is a big historic moment happening, all those years back AMD was saying to these same developers "it's not our fault - it works fine" - Finally the Driver teams have been awakened they actually don't know how to optimize games and all they were doing was mediocre hacking the driver to try to make something work under DirectX.