The AMD Mantle Thread

[blackened23]

I don't know why you're doing comparisons to OpenGL because there are so few, few games using OGL. You state that Steam box will change things. I'm 99.9% sure Steam Box changes nothing - steam box will be nothing more than a pre-built PC with a price to match; certainly the steam box will not set the world on fire with sales passing the PS4. The PS4 will outsell the Steam Box 10000:1.

The fact of the matter is OGL is nearly irrelevant as almost nothing uses it (and instead uses DX9/DX11) and steam box + linux will not sell well. Steam box could sell well if priced as a console at 400$ or less - doing that at 400$ would require a slow horrible AMD APU and we know how that would end up. Conversely, the engines that Mantle is targeting is already used by 85%+ of games. That is AMD's strategy. We'll see how it plays out. In the meantime, OGL and Steam Box are nigh irrelevant.

 

[VulgarDisplay]

Ive already explained, it does not work that way, you may stop 1 seconds and start thinking about it, or not, not my problem.

Why does it not work? I will admit it's probably not a seemless transition, but this console generation is unlike any that has come before and AMD has done something that will have a huge impact on game development in the future.

They are in the Xbone and PS4 with nearly identical hardware. The entire goal of what they are doing with Mantle is to simplify development across platforms and in the process make games developed for GCN run great on both consoles and the PC. No one can conclude how it is going to play out because this console generation is unique and there is nothing to compare it to. Anyone who is unwilling to see that AMD has some master plan revolving around leveraging their console wins is out of touch with reality. The way things have been going I'm of the opinion that it is playing out exactly how they imagined and that we are in for a treat once it all comes to fruition.

 

[3DVagabond]

Im software engineer, well not yet. But you can guess i have a good idea on how much time and cost takes to implement a new api, Mantle or OpenGL or anything else does not matter, still i can assure you than implement Mantle will be harder than implement OpenGL, its a new api, that needs training for devs, and if it has low level elements on it thats worse.
And if your game engine was not developed with mantle in mind it will be even worse.


Im not saying Mantle will not be great, but there is a lot of issues with it to implement in a real world, its not so easy as AMD makes you to belive.

If you are going to be developing for GCN consoles you are already putting in these low level GCN optimizations. Mantle allows you to much more easily use these same features in your pc port. It shouldn't be too hard as to not be worth it. Not by anyone who cares about their end product. Good, well done games, make heaps of money.

 

[3DVagabond]

Ive already explained, it does not work that way, you may stop for one moment and start thinking about it, or not, not my problem.

No, actually you have not explained anything. You have simply stated that you are of the opinion that implementing Mantle will negatively affect the bottom line, it's too hard, etc... The reason you've used, increased development cost and more work, is short sighted to the extreme.

I don't know why so many people seem to be in love with Dx. The devs have been complaining about it for years. The devs appreciate that AMD has given them an API with low level access. Haven't you been reading anything about it?

 

[BallaTheFeared]

I wonder how grateful the devs will be when they actually try to use the additional performance allowed only to discover 65% of the market hates their poorly performing product?

That's the way I see it, either you're inflating FPS on one side, or crippling performance on another. There is no in between win/win because of how this enhances performance (cpu wise).

 

[3DVagabond]

I wonder how grateful the devs will be when they actually try to use the additional performance allowed only to discover 65% of the market hates their poorly performing product?

That's the way I see it, either you're inflating FPS on one side, or crippling performance on another. There is no in between win/win because of how this enhances performance (cpu wise).

Mantle is in no way shape or form going to effect Dx11 performance. It's a separate rendering path. The lower performance using Dx11 will exist on both AMD and nVidia cards. This is not hurting nVidia's performance in respect to AMD's at all. Any performance deficit is at the feet of Dx. I understand that by it's nature Dx has to be a high level API to allow everything to run on it. You watch though, let Mantle perform as it should, be added to other OS's besides Windows, and you will see M$ get off their corporate behinds and make major improvements to Dx. OpenGL does not compete effectively against Dx. The numbers are overwhelmingly in Dx's favor. If Mantle, through appreciably higher performance (because that's the only way it will) makes inroads to Dx's dominance, we'll get a better Dx. Competition is a good thing for consumers. People need to realize this.

 

[BoFox]

http://alienbabeltech.com/abt/viewtopic.php?f=6&t=31776&start=10#p97385)[/url]] Lets say that Mantle does what it says on the tin - it improves performance significantly enough that its popular and that games developers want to use it, a big if. Were that the case it would be highly tied to the current generation, design and capabilities of today's graphics cards. Future designs would need to maintain a very low level API and hence couldn't change their behaviour much for fear of breaking old games. Porting a game to new hardware would be a considerable exercise. This would only work if GPUs had stopped evolving the sort of features they have and their general capabilities.

So in many ways Mantle does to PC graphics cards what consoles do to PC hardware. They fix it to a particular moment in time and with the low level access extend the life of the fixed hardware platform a decent time period. But what it does not do is support a strong future with multiple vendors (of which AMD is currently the smallest behind Intel and NVidia dependent on market) nor the performance gains those bring. Whatever DirectX might cost today in performance in 2 years time and one node change will complete destroy the previous generation of cards and likely wipe out whatever minimal benefit Mantle might have brought. Maybe Mantle will support one extra generation of cards, but probably not two generations. So in 4 years time it would be obsolete or gaining cruft to try and support different cards, then the developers are supporting both and so on. Its not a sustainable model to not abstract the hardware away unless the hardware is very stable in its capabilities.

A lot hinges on what they mean by low level and how the features of the card are exposed. Its either possible for NVidia to write a driver for it and thus it will also support future AMD architecture or it isn't. If it isn't possible to support Intel/NVidia hardware then its very likely future AMD hardware will be limited by it as well. In reality most games don't currently use enough draw calls for the benefits to really show through. The more exciting thing is the multithreaded support, but we need to see it to understand its limitations and we can't currently see it. Either way DirectX may very well be able to adopt more multithreading, seems likely it will, but also support an abstraction that scales with hardware.

A new API with a modern driver interface supporting true multithreaded access could really take off, a truly low level API is a really bad idea for the PC market in general, it assumes cards are not going to improve and I don't think that is very likely.

Let's say that it's like PhysX drivers for Nvidia's CUDA architecture. CUDA provides a low-level API as well. Because there was PhysX for some games a few years ago doesn't mean that Nvidia cards are not going to improve. Even if 50% of games out there started supporting PhysX from 2 years ago, Nvidia would still be able to move on to faster CUDA-based architectures.

All we know for now is that Mantle is a GCN-based API. My bet is that AMD could still move on to faster GCN-based architectures, just like NV with their CUDA cores from GT200 to GF100 to GK100 to GM100 and so on... GCN is a rather new, forward-looking flexible uarch altogether, just like CUDA was for NV, it seems.

Nevertheless, for 4-5 years down the road, AMD most likely does not have enough R&D to forfeit GCN for yet another completely new uarch altogether, after ditching VLIW5/4 that didn't last as long as CUDA.

RE: "So in many ways Mantle does to PC graphics cards what consoles do to PC hardware."
Well, you say that consoles have been doing this to PC hardware (including graphics cards)? Well - there are next-gen consoles too, mantle or not, that will be around for a while too. Xbox360 and PS3 allowed for vastly greater draw calls than PC-based DirectX allow for, still today - several years after the release of such consoles. So, in a way, something else has to be blamed.

 

[BallaTheFeared]

I don't think you understood what I said.

If they design the game to take advantage of Mantle it will tank horrifically on non AMD GCN systems, if they design it to function in DX11 all we'll see are inflated FPS scores where the cpu is limiting and no actual benefits for most users here.

Seemingly the two benefits are: Lower cost for better performance than previously available with AMD cpu based systems, and enabling AMD cpus to gain some traction over the course of the consoles lifetime by reducing CPU load at similar fps levels via the "Good enough - effect".

 

[3DVagabond]

I don't think you understood what I said.

If they design the game to take advantage of Mantle it will tank horrifically on non AMD GCN systems, if they design it to function in DX11 all we'll see are inflated FPS scores where the cpu is limiting and no actual benefits for most users here.

Seemingly the two benefits are: Lower cost for better performance than previously available with AMD cpu based systems, and enabling AMD cpus to gain some traction over the course of the consoles lifetime by reducing CPU load at similar fps levels via the "Good enough - effect".

I've read where you are under the impression that this will only benefit CPU bottlenecking. Mantle allows for direct access to all of the GPU's features bypassing Dx. That's not just going to relieve CPU bottlenecks. It's also in no way going to make any difference in Dx11 scores.

Since almost all games are already designed to run on consoles, they will already have low level GCN optimizations. Even if there was no such thing as Mantle for PC's. This simply allows the GCN GPU's in PC's to have access to the same low level optimizations.

 

[24601]

I don't think you understood what I said.

If they design the game to take advantage of Mantle it will tank horrifically on non AMD GCN systems, if they design it to function in DX11 all we'll see are inflated FPS scores where the cpu is limiting and no actual benefits for most users here.

Seemingly the two benefits are: Lower cost for better performance than previously available with AMD cpu based systems, and enabling AMD cpus to gain some traction over the course of the consoles lifetime by reducing CPU load at similar fps levels via the "Good enough - effect".

People forget that this will benefit the pre-sandy bridge intel quad and hex cores the most if it actually pans out.

 

[VulgarDisplay]

I don't think you understood what I said.

If they design the game to take advantage of Mantle it will tank horrifically on non AMD GCN systems, if they design it to function in DX11 all we'll see are inflated FPS scores where the cpu is limiting and no actual benefits for most users here.

Seemingly the two benefits are: Lower cost for better performance than previously available with AMD cpu based systems, and enabling AMD cpus to gain some traction over the course of the consoles lifetime by reducing CPU load at similar fps levels via the "Good enough - effect".

The point of Mantle is to take console optimizations and port them straight into PC's. That work has been getting done for the entire existence of consoles. If it plays out like they are saying it will not mean much additional work for the PC version. So, dev codes mantle version for ps4 and xbone, and it work son GCN cpu's on PC. Dev codes for DX11 port like always. It should be the same amount of work, or very little extra work compared to what they have always done.

 

[BallaTheFeared]

The problem with "console optimizations" is that nobody really actually knows what they are outside of draw calls which is cpu related.


Does Mantle make a 7950 operating at 99% usage in DX11 get higher fps under Mantle? I find it's highly unlikely that will be the case.

 

[3DVagabond]

The problem with "console optimizations" is that nobody really actually knows what they are outside of draw calls which is cpu related.


Does Mantle make a 7950 operating at 99% usage in DX11 get higher fps under Mantle? I find it's highly unlikely that will be the case.

It's supposed to. I guess we'll find out.

 

[redhotiron2004]

What is mantle actually supposed to do? I know it's a low level API supposed to work even with high level API. And its supposed to be similar to xbox API.
I know AMD is going to introduce Huma. Which would be a hardware version of unified architecture. Which would leverage the power of both GPU & CPU as one for improvement in performance.

So, I am just thinking of mantle as a software part to utilize that hardware. There is going to be improvement in performance due to hardware. But, why mantle such a big deal?

People here are making such a big hype about the performance being doubled. Why? Isn't the hardware supposed to be the main reason for that boost rather than the software.

Weren't there existed any low level API before mantle?
Even if mantle is that revolutionary as you mentioned. I really doubt that anything software wise could improve the performance even by 10-20% max(which is good). But, not such a huge deal.
I find there claims to be fictitious.

 

[3DVagabond]

Everyone's performance claims, whether single digit or 2x or somewhere in between, are their opinions. The ones at the extreme's typically aren't objective in their assertions.

Some people are saying it's minimal because they are only taking CPU optimizations into acct. Some say it will be huge because it's been typically put forth that Dx compared to direct to metal like on consoles is a 2.5x performance hit.

 

[BallaTheFeared]

I guess we'll finally get to end the console optimization debate once and for all then.

 

[BoFox]

The problem with "console optimizations" is that nobody really actually knows what they are outside of draw calls which is cpu related.


Does Mantle make a 7950 operating at 99% usage in DX11 get higher fps under Mantle? I find it's highly unlikely that will be the case.

It's not just the draw calls. DirectX is bloatware. Eliminating the bloatware could ideally reduce not only the CPU overhead but also the GPU overhead.

A GPU operating at 99% usage does not mean that it is running 99% efficiently.

There could be far more efficient usage of 99% of the GPU's uarch, allowing for 2x the frame rate, or perhaps some additional API/driver algorithm stuff like PhysX/CUDA for Nvidia's GPUs but something that really takes advantage of Radeon uarch (like Bitcoins, hehe, I kid..).

A poorly written game could be using the GPU at 99%, with high-level stuff maxing-out the color fillrate for example, and look like crap.

 

[BallaTheFeared]

DX is the reason we all use windows and not free OSes like Linux, because "bloatware" allows everyone to play, not just a select few that buy the product that an unreleased game ends up using.

I've never seen anyone say DX11 graphics overhead was a problem, it has always been about draw calls, even AMDs own slides mention nothing about increased graphical performance, only "enhancements" - though they quite clearly spelled out the reduction in draw call overhead...

Bitcoins do better on AMD because they can do an op a cycle whereas Nvidia cards required three.

A poorly written game on any API is still a poorly written game, just as a bad example is bad no matter what the comparison being made is.


At this point it's just words and opinions, if there were any facts there would be no debate.

Hopefully BF4 ends this debate, and I'll be in a decent spot to watch it all unfold since I can max my usage at 1125/1500 in CF for around 60 FPS without being cpu limited.

 

[NIGELG]

DX is the reason we all use windows and not free OSes like Linux, because "bloatware" allows everyone to play, not just a select few that buy the product that an unreleased game ends up using.

I've never seen anyone say DX11 graphics overhead was a problem, it has always been about draw calls, even AMDs own slides mention nothing about increased graphical performance, only "enhancements" - though they quite clearly spelled out the reduction in draw call overhead...

Bitcoins do better on AMD because they can do an op a cycle whereas Nvidia cards required three.

A poorly written game on any API is still a poorly written game, just as a bad example is bad no matter what the comparison being made is.


At this point it's just words and opinions, if there were any facts there would be no debate.

Hopefully BF4 ends this debate, and I'll be in a decent spot to watch it all unfold since I can max my usage at 1125/1500 in CF for around 60 FPS without being cpu limited.

When you get or buy your Battlebox as touted in your picture there you won't have any need for Mantle or AMD.

 

[Skurge]

I will just leave this here.

 

[BallaTheFeared]

When you get or buy your Battlebox as touted in your picture there you won't have any need for Mantle or AMD.

It's for going to war in more than one game.

 

[ShintaiDK]

I will just leave this here.

You might want to add that Glide didnt support 32bit in the game. So as resolution went up, DirectX got faster than Glide. But its a nice example on that GPU speed didnt increase, simply the CPU part. And thats with a very early DirectX, already beating Glide in GPU performance.

Not to mention this. using a 866Mhz P3 instead of Anandtechs 550Mhz P3.

 

[SiliconWars]

You also might want to add that early DirectX didn't support multitexturing and had to use totally different textures back then.

Unreal engine on Direct3D is like something out of the early 80's...

 

[Erenhardt]

I wonder how grateful the devs will be when they actually try to use the additional performance allowed only to discover 65% of the market hates their poorly performing product?

Don't hate the game...hate the hardware.

There may be parts of GCN that are not used efficiently/at all, that mantle could possibly make use of, but that would make GCN cards run games like they were power viruses.

 

[SiliconWars]

http://forums.battle.net/thread.html?topicId=17899774350&sid=3000&pageNo=1

Most people probably run (Diablo II) in Direct3D mode, but the game is better in 3dfx Glide mode, giving better framerates and much sharper ground textures.

Comparison of quality in Diablo II (remember Glide is also faster even when using a wrapper) -

Glide - http://tinypic.com/view.php?pic=4sfcjm&s=5
D3D - http://tinypic.com/view.php?pic=2qkusl4&s=5

This was a game released in 2000, after Glide was basically on the way out.