The AMD Mantle Thread

[BrightCandle]

If your expectation is that you will be able to have higher polygon or more textures on your models then I am afraid to say that is unlikely something Mantle will help with. Those are core GPU limited functions (fixed function no less) and wont benefit from additional draw calls at all, nor will they benefit from the parallel command queues for shaders.

If AMD and the demo are representative then it will be possible to have 10x as many things in the world, or more complicated interrelated shader programs (which might have some performance benefit in some cases) and by and large that is about it. You don't get more geometry budget, but you can split that budget up amongst a lot of smaller pieces. I guess that is why I am less enthusiastic, most games don't hit the limit on DX draw calls and hence aren't likely to gain anything with a Mantle layer. Some games do hit up against the limit and those games will be able to achieve good performance with less limitations in their approach. But its not going to change the fundamentals of how good the graphics are really, its unlikely to do much there at all.

The Arma 3 engine as I said is mostly dominated by a single threaded simulation. A piece of simulation code that is likely a million lines of code, it will take them years to make it more parallel, these things don't just happen magically because we want them too. BI have been working on multithreading for 3 years and the current state of play is the result of those years of effort. Its going to take a long time to make it work as described and that has nothing to do with Mantle.

 

[3DVagabond]

If your expectation is that you will be able to have higher polygon or more textures on your models then I am afraid to say that is unlikely something Mantle will help with. Those are core GPU limited functions (fixed function no less) and wont benefit from additional draw calls at all, nor will they benefit from the parallel command queues for shaders.

If AMD and the demo are representative then it will be possible to have 10x as many things in the world, or more complicated interrelated shader programs (which might have some performance benefit in some cases) and by and large that is about it. You don't get more geometry budget, but you can split that budget up amongst a lot of smaller pieces. I guess that is why I am less enthusiastic, most games don't hit the limit on DX draw calls and hence aren't likely to gain anything with a Mantle layer. Some games do hit up against the limit and those games will be able to achieve good performance with less limitations in their approach. But its not going to change the fundamentals of how good the graphics are really, its unlikely to do much there at all.

The Arma 3 engine as I said is mostly dominated by a single threaded simulation. A piece of simulation code that is likely a million lines of code, it will take them years to make it more parallel, these things don't just happen magically because we want them too. BI have been working on multithreading for 3 years and the current state of play is the result of those years of effort. Its going to take a long time to make it work as described and that has nothing to do with Mantle.

See, I don't know for certain, but I think you are wrong. There's a lot more to models than simply geometry.

 

[BrightCandle]

See, I don't know for certain, but I think you are wrong. There's a lot more to models than simply geometry.

It was you that mentioned the actual geometry and the desire to have more of it, that is an aspect that wont be changed by Mantle. You can have more additional information associated with a model, more shader passes to process them for lighting and other effects and you can have more detailed LOD algoriothms and whole range of other post processing aspects. You can have more models that are simpler and you can probably have more less detailed textures. But raw basic model capabilities like the textures and poly count are not going to improve as they are completely limited by the GPUs fixed function capabilities.

 

[Lonbjerg]

Arma 3 is mostly dominated by the game world updates (what they call the SimuA process). The DX calls hover around 10ms most of the time, so there is certainly draw call overhead but from my tests the time is dependent on a variety of things. Mostly the problem with Arma 3 is its single threaded, it would achieve 90 fps or so if it wasn't for the simulation being both single threaded and on the same thread as everything else.

Mantle would help it but there are other more important things A3 needs to make it run than reducing the draw calls.

Did you just call ARMA3 for singlethreaded?

Would you stop posting false info, the ARMA series have been multicore for a long time:
http://www.bistudio.com/english/company/developers-blog/91-real-virtuality-going-multicore

Real Virtuality 4 is even more multicore.
Why does Mantle needs false statements to seem attractive?

Edit:
ARMA dev chiming in:
http://forums.bistudio.com/showthread.php?100519-exThreads&p=1797139#post1797139

Edit 2:
No wonder ARMA3 is disliekd in AMD country:
http://www.techspot.com/review/712-arma-3-benchmarks/page4.html

 

[Lonbjerg]

Oh...i KNEW it.
Looked in my saved links folder...found the SAME poster, posting the same FUD about ARMA before...who would have guessed?
http://forums.anandtech.com/showpost.php?p=34730531&postcount=87

 

[3DVagabond]

Oh...i KNEW it.
Looked in my saved links folder...found the SAME poster, posting the same FUD about ARMA before...who would have guessed?
http://forums.anandtech.com/showpost.php?p=34730531&postcount=87

You'd save that? :\

 

[AtenRa]

Edit 2:
No wonder ARMA3 is disliekd in AMD country:
http://www.techspot.com/review/712-arma-3-benchmarks/page4.html

Am I missing something here or what ???

Core i7 4770K at 3.5GHz produces 49 fps, but at the next slide it jumps to 53fps.

Also, FX8350 at default 4GHz produces one more fps than Core i7 4770k at 3.5GHz.

Not to mention that 5 year old Nehalem Core i7 920 at 2.66GHz is as fast as Haswell Core i7 4770K at 3.5GHz ???

http://www.techspot.com/review/712-arma-3-benchmarks/page5.html

 

[Stewox]

Oh yawn mawn bawn ...

That's some pretty rampant denial going on there Lonbjerg. You can't actually believe this is somehow faked? If you watched it you'd know that he says at the start that it would be impossible to fake it even with a full film crew, there's just too much going on screen.

http://www.youtube.com/watch?feature=player_detailpage&v=QIWyf8Hyjbg#t=1451

When you see a movie trailer, do you say.... thats just a movie trailer!
that ll NEEEEVER... be made into a REAL movie!

Thats akin to what your saying.

Why is it so hard for you to just take someones word for it? no trust in your fellow man?

If a game dev shows you a working demo of their game engine, and say its from a game their developing, why is it you automatically assume its fake?

You love to spread FUD, thats the most likely explaination I can come up with.
(and you really love nvidia, and reeeaally hate amd, we get it already, but enough with all the fud)

They said specs like it was marketing? What does that even mean. They tell you exactly what CPU and GPU it runs on. Considering the demo is GPU bottlenecked they CPU clock is not relevant, and since it's a demonstration from a month ago, you can assume it's a reference Hawaii board, so it's not going to clock that well.

They tell you that with DX they run into problems past 15k batches. While in Mantle you can see the demo reach 100k batches at decent FPS. The only problem with that number is that we don't know if it's a GPU or CPU bottleneck. But considering how limited DX core scaling is, even if it was a CPU bottleneck, they still couldn't reach the numbers Mantle is having.

For those that actually want to try the demo out, they said they will release it in Q1.

I am in denial because I remember this?
http://www.youtube.com/watch?v=h1E6vLsX2UI
AMD GPU physics demo...

Never came to life...

Like so many other broken promises.

And my ARMA 3 runs just fine...but since renders a MUCH lager world than BF4...you might want to upgrade your hardware...like I said...I run s socket 1366 PC....max settings...I even have the performce overhead to record video's...at max settings
EDIT:
Again i7 990X + Titan running ARMa3 at max settings (view distance 7400 meters)...show me where I am draw call limited?
http://www.youtube.com/watch?v=SkHQlId8GfI&feature=c4-overview&list=UUb2qvg7CbWkPLE4wc9ULT4g

Why all the false "arguments"...in defence for PR?

Is there a way to bulk report off topic posts? Thread is begging for it.

Here I come for the rescue!

Here guys make your selfs happy.


Proudly Presents:


Mantle Tribute:

http://www.youtube.com/watch?v=gO-P4t8yGtY

 

[SiliconWars]

Good stuff Stewox!

 

[Lonbjerg]

Good stuff Stewox!

I got something for you...as promised...same settings
http://www.youtube.com/watch?v=qR6qDGrlCSg

Edit:
Forgot to turn of my mike...so the "strange" sounds are IRL ambient stuff...good thing I wasn't humming ^^

 

[BrightCandle]

Here is a profile of Arma 3 taken just a couple of weeks ago. This shows the extent of the multithreading available in Arma 3 using the games inbuilt profiler:

As you can see its mostly single thread dominated. There is some offload during the rendering process but by and large the top bar is dominated by two things, the simulation and the rendering (which is likely draw calls limited). I stand by my assertion, I have been profiling Arma 2 and 3 for many years, have helped track down many bugs in both games and I do know what I am talking about when it comes to this game, I have well over 1000 hours of gameplay in A3 alone.

 

[Lonbjerg]

Here is a profile of Arma 3 taken just a couple of weeks ago. This shows the extent of the multithreading available in Arma 3 using the games inbuilt profiler:

As you can see its mostly single thread dominated. There is some offload during the rendering process but by and large the top bar is dominated by two things, the simulation and the rendering (which is likely draw calls limited). I stand by my assertion, I have been profiling Arma 2 and 3 for many years, have helped track down many bugs in both games and I do know what I am talking about when it comes to this game, I have well over 1000 hours of gameplay in A3 alone.

So you against ARMA devs:

Try adding A.I to the mix...nice try

It's like you ignored my link on purpose...just like the last time

But why don't you try and run ARMA3 on a singlecore and post your results?

 

[Noctifer616]

So you against ARMA devs:

Try adding A.I to the mix...nice try

It's like you ignored my link on purpose...just like the last time

But why don't you try and run ARMA3 on a singlecore and post your results?

Just like you, I am calling the "BS PR" card till a source that isn't a ARMA 3 dev can confirm this to be true.

 

[Lonbjerg]

Just like you, I am calling the "BS PR" card till a source that isn't a ARMA 3 dev can confirm this to be true.

It's real easy...ask BrightCandle to run ARMA3 on a single core ^^

 

[Grooveriding]

Why is this thread being derailed with Arma 3 ? That unoptimized pile of turd has nothing to do with Mantle.

 

[Lonbjerg]

Why is this thread being derailed with Arma 3 ? That unoptimized pile of turd has nothing to do with Mantle.

Because some posters claims this title would benefit from Mantle.
Try reading the thread

 

[BrightCandle]

Do you see the difference in the two charts because there is a significant difference in what my chart measures compared to the one the devs provided? The dev one is focussing exclusively on the rendr part of the process, which you can see in my chart also has some amount of parallel work as well, but its only 1/3 of the time. The dev chart specifically excludes in their picture the SimuA section, which is clearly misleading. They are showing just a third of the chart I am, I am showing the entire frame from end to end, every single contributing aspect to the frame time. The devs, for marketing reasons have decided to focus on the bit they have made parallel, not because it was the biggest problem or impact but because it was what they have done. By only showing the rendr and its parallelism they have intentionally mislead people into thinking the game handles multiple cores and threads well, which is clearly not the case when you look at the entire frame time. I don't dispute the rendr part of the process is parallel, infact it scales to 6 cores for about 30% utilisation in all, but its only 1/3 of the frame.

When you look at the big picture the simulation is a bigger part of the frame time, and its 100% single threaded only. Even if Mantle removed all the rendering time and reduced the rendering to 0ms we still wouldn't be getting more than about 60 fps in the example I have shown. We have multiplayer games that go down to 25 fps and mantle might make that 35 fps, but its not going to fix the underlying issue because for that the simulation part of the thread needs fixing.

To me that BI article is a classic example of marketing over substance. They make bold claims of parallelism and then specifically exclude large parts of important information which intentionally misleads their customers. I have provided these profiles to BI and they don't dispute the issue, we are working with them to trace these problems so they can fix them.

Almost no one benchmarks these sorts of multiplayer scenarios like we do, we bench real games in progress. We don't do single player tests to show how great performance is because a) single player runs a lot better and b) its simply not representative of multiplayer performance at all.

So Mantle would help a little, but not a lot. A3 is actually a pretty typical game in how it does this. Up until very recently almost all games came with sim followed by render and the multiple core use was mostly drivers and DX multithreaded support. Games designed for next gen are showing more parallelism but quite a few PC exclusives are going to have to go through a similar big development effort to fix their simulation engine to work in parallel before they can noticeably improve their performance with future CPUs and APIs like Mantle.

 

[Lonbjerg]

Do you see the difference in the two charts because there is a significant difference in what my chart measures compared to the one the devs provided? The dev one is focussing exclusively on the rendr part of the process, which you can see in my chart also has some amount of parallel work as well, but its only 1/3 of the time. The dev chart specifically excludes in their picture the SimuA section, which is clearly misleading. They are showing just a third of the chart I am, I am showing the entire frame from end to end, every single contributing aspect to the frame time. The devs, for marketing reasons have decided to focus on the bit they have made parallel, not because it was the biggest problem or impact but because it was what they have done. By only showing the rendr and its parallelism they have intentionally mislead people into thinking the game handles multiple cores and threads well, which is clearly not the case when you look at the entire frame time. I don't dispute the rendr part of the process is parallel, infact it scales to 6 cores for about 30% utilisation in all, but its only 1/3 of the frame.

When you look at the big picture the simulation is a bigger part of the frame time, and its 100% single threaded only. Even if Mantle removed all the rendering time and reduced the rendering to 0ms we still wouldn't be getting more than about 60 fps in the example I have shown. We have multiplayer games that go down to 25 fps and mantle might make that 35 fps, but its not going to fix the underlying issue because for that the simulation part of the thread needs fixing.

To me that BI article is a classic example of marketing over substance. They make bold claims of parallelism and then specifically exclude large parts of important information which intentionally misleads their customers. I have provided these profiles to BI and they don't dispute the issue, we are working with them to trace these problems so they can fix them.

Almost no one benchmarks these sorts of multiplayer scenarios like we do, we bench real games in progress. We don't do single player tests to show how great performance is because a) single player runs a lot better and b) its simply not representative of multiplayer performance at all.

So Mantle would help a little, but not a lot. A3 is actually a pretty typical game in how it does this. Up until very recently almost all games came with sim followed by render and the multiple core use was mostly drivers and DX multithreaded support. Games designed for next gen are showing more parallelism but quite a few PC exclusives are going to have to go through a similar big development effort to fix their simulation engine to work in parallel before they can noticeably improve their performance with future CPUs and APIs like Mantle.

So you didn't run ARMA3 on a singlecore, gothca.

 

[frozentundra123456]

Am I missing something here or what ???

Core i7 4770K at 3.5GHz produces 49 fps, but at the next slide it jumps to 53fps.

Also, FX8350 at default 4GHz produces one more fps than Core i7 4770k at 3.5GHz.

Not to mention that 5 year old Nehalem Core i7 920 at 2.66GHz is as fast as Haswell Core i7 4770K at 3.5GHz ???

http://www.techspot.com/review/712-arma-3-benchmarks/page5.html

Not sure what the point is of this post, but in any case, the 4770k at 3.5 in the first graph must have turbo disabled, thus the slower performance relative to the 4770k in the second graph. In the valid comparison graph, the 4770k is 10% faster than the 8350, although the entire line up of top end cpus is becoming gpu limited, so it is not really a valid comparison of cpu performance anyway.

Edit: Game.gpu testing shows 2600K almost 40% faster than stock 8350 in a less gpu limited scenario.

 

[Stuka87]

I do think the Arma talk is going off topic from Mantle.

The game probably could benefit from mantle, but the game would benefit the most from an engine that used more than two cores.

 

[Tenderness]

I do think the Arma talk is going off topic from Mantle.

The game probably could benefit from mantle, but the game would benefit the most from an engine that used more than two cores.

Considering the incompetence of the developers of that game, I doubt it could benefit from anything.








I warned you that creating another account under vacation will earn you a permaban.

Now go away.


esquared
Anandtech Forum Director

 

[Lonbjerg]

I do think the Arma talk is going off topic from Mantle.

The game probably could benefit from mantle, but the game would benefit the most from an engine that used more than two cores.

So you claim dualcores run ARMA just as well as quad-cores, against the presented data?
Why?

 

[Lonbjerg]

Considering the incompetence of the developers of that game, I doubt it could benefit from anything.

Oh look..a new poster again, again, again, again, again, again trying do "debunk" anything not praising Mantle...without any data.

It funny how everything not supporting AMD's PR about Mantle...is causing such pain and get trashtlaked...with no documentation.

 

[chimaxi83]

The massive derailing of this thread coinciding with one frequently vacationed member coming back...... what a coincidence. This thread is garbage.

 

[Tenderness]

Oh look..a new poster again, again, again, again, again, again trying do "debunk" anything not praising Mantle...without any data.

It funny how everything not supporting AMD's PR about Mantle...is causing such pain and get trashtlaked...with no documentation.

Yes new here, hello to you too.





No you aren't. I warned you that creating another account under vacation will earn you a permaban.


esquared
Anandtech Forum Director