AMD Mantle update...

[Lanek]

I think the reason you are seeing such a speed boost variance is based on what limitation a CPU bottleneck is affecting performance. While you may get 10% where no bottleneck exists, when you compare a slower CPU (such as AMD's APU) vs the fastest (Intel top of the line) you are going to see a much bigger boost with the APU % wise vs that of the Intel EE. You will also see greater improvement in a game which uses loads of CPU calculations like AI, such as the Swarm demo, Starcraft, or other RTS type games.

I will said, it should depend more of the game, at least when we speak about overhead only. , but the video from oxyde comparing DX and Mantle with an I7 980x + 290x ... show the same increase. (( DX 8-13fps , mantle 40 to 60+ ( with peak at 85+fps )... http://www.engadget.com/2014/01/14/o...y-mantle-demo/

But starswam is an extreme case in term of CPU calculation anyway. 5000 independant AI objects who run at the same time... You take starcraftII, you take the worst case in term of numbers of objects and you multiply it by 10x ( 500 max objects to 5000 ). We are speaking a lot about RTS with the Oxyde engine, but personally i can see the impact on flight simulation like DCS A10C.. the impact on the cpu for it was crazy, because every AI ( objects, radar detection, missile, etc ) is calculated inreal time and simulated based on real "system " .

[Iconyu]

No, the reason there is such a variance is that Starswarm was coded especially for Mantle. It's beyond a best case scenario. In 90% of the Mantle titles we see for the next few years, the standard DirectX + Windows driver will be the primary rendering tool while Mantle will be added in. Hence why there is a much lower gain in BF4, the poster child for an actual Mantle-supporting game.

HSA isn't active right now, nor will it come into effect within games. Games actually do a pretty good job of partitioning resources (especially those that support DirectCompute, etc.) already so HSA is mostly focused on compute-focused apps through OpenCL 2.0.

I was thinking along the lines of the direct compute already in Frostbite simply being shifted over to the APU to free up the 290x for rendering, something Mantle can do adding more GPU resources to the pool without getting involved with AFR the engine uses natively. The extra control Mantle allows means that it would be easier to fine tune especially if you knew the hardware you were going to demo on.

But it depends on how much work they'd be willing to put into it, I imagine latencies could be an issue, but people have been hacking nvidia drivers to allow AMD as primary renderer for hardware accelerated physx for years.

From my point of view, it would be a smart idea to implement, but it would make 45% rather suspect unless you knew either way if DICE were using the APU like that.

[Lanek]

I was thinking along the lines of the direct compute already in Frostbite simply being shifted over to the APU to free up the 290x for rendering, something Mantle can do adding more GPU resources to the pool without getting involved with AFR the engine uses natively. The extra control Mantle allows means that it would be easier to fine tune especially if you knew the hardware you were going to demo on.

But it depends on how much work they'd be willing to put into it, I imagine latencies could be an issue, but people have been hacking nvidia drivers to allow AMD as primary renderer for hardware accelerated physx for years.

From my point of view, it would be a smart idea to implement, but it would make 45% rather suspect unless you knew either way if DICE were using the APU like that.

Actually, i read a lot about " the works who want spend developpers of a game ", but in reality, developpers are not developping a specific game, they developp tools for a specific engine, they want to sold then.. The reason Oxyde is puting so much effort today and Dice with Frostbyte engine, is they sell this engine to other studios, So basically developpers, when they have buy the tools will see there "time " reduced drastically. " .. exactly like Crytek and the CryEngine.. ... Ofc then, you see too developpers ( i think to StarCitizen develeopers, PCcars developpers and other ) who are just seeing what can bring to their game / future games engine Mantle in pure developpers term.. ( more drawcall, finetuning, allow them to exploit better their own engine outside the limitations of DX, because they are developping the tool, the engine, the game ( ). .. This said, we can too see "small " studio, who use Nvidia closed source library ( Gamework ) for their "first titles ".. its easier for them.. (( and in the case of gameworks, or PhysX, win some visibility, contract ( with nvidia ) . )

I have just seen Tombraider comparaison between PS4 and PS3 ... What is the first thing you see outside better textures, weather effects, 1080p and the fact it look really similar of the pc version of TR ... TressFX is enabled now in the PS4 version.... And it run really really well ..

[kuroikenshi]

most game companies buy a graphics engine, and they will buy whatever makes their game work the best. Oxide's demo shines light into a new DAWN for RTS where in the past just a few dozen units would demolish system performance (Starcraft & DOW 1 & 2 specially)

Obviously the improvement will be observed overall, however RTS's are going to be one of the major winners in this new era of low level coding for pc environment.

[Lanek]

most game companies buy a graphics engine, and they will buy whatever makes their game work the best. Oxide's demo shines light into a new DAWN for RTS where in the past just a few dozen units would demolish system performance (Starcraft & DOW 1 & 2 specially)

Obviously the improvement will be observed overall, however RTS's are going to be one of the major winners in this new era of low level coding for pc environment.

Anyway, the Oxyde engine, is not pulling ahead only in the term of " overhead " or Mantle, i dont know ( not sure ) if it is due to Mantle, but they have completely rewrite the way the code and thread call is made.. if one year ago you will have say to a game developpers, someone want do this, they will have just say it will not work.. now they run after Oxyde for this.. ( and this is not necessary due to Mantle " as API" ). When i find the forums posts of another studio who have let down their own game engine for the Oxyde one and explain why, i will try find time for post it .. ( completely busy with my job this month, christmas hollidays are nice, but now im 2 weeks late in my works )

[Hornet331]

I will said, it should depend more of the game, at least when we speak about overhead only. , but the video from oxyde comparing DX and Mantle with an I7 980x + 290x ... show the same increase. (( DX 8-13fps , mantle 40 to 60+ ( with peak at 85+fps )... http://www.engadget.com/2014/01/14/o...y-mantle-demo/

But starswam is an extreme case in term of CPU calculation anyway. 5000 independant AI objects who run at the same time... You take starcraftII, you take the worst case in term of numbers of objects and you multiply it by 10x ( 500 max objects to 5000 ). We are speaking a lot about RTS with the Oxyde engine, but personally i can see the impact on flight simulation like DCS A10C.. the impact on the cpu for it was crazy, because every AI ( objects, radar detection, missile, etc ) is calculated inreal time and simulated based on real "system " .

If you actually would have looked at the demo you would realize the performance increase has nothing to do with AI or unit count but with the motion blur filter they used or better disabled when running DX.
As said this is very similar to first metro where they also used an ultra expensive motion blur algorithm witch brought down highend cards back then in sub 30fps region.

Also 5000 units is nothing, we have games right now that can handle much more, Star ruler has no unit cap at all and yes all units have AI.

most game companies buy a graphics engine, and they will buy whatever makes their game work the best. Oxide's demo shines light into a new DAWN for RTS where in the past just a few dozen units would demolish system performance (Starcraft & DOW 1 & 2 specially)

because all this games are designed to be small scale, there are games that can handle much more units without problems, SC (few thoused), star ruler (unlimited), Sins of solar empiers (with mods simple mods in the 10k region, unmoded something like 2-3k)

[kuroikenshi]

yeah well, sins of a solar empire in one of those games which i've only played a little, but in terms of real time rendering I'm not sure if those games come close to what Oxyde is doing, I dare say no.

I'd like to see a demo of a constant interaction of 5k-7k units constantly battling under the same parameters Oxide is doing it in. for those who sat through the entire Oxyde presentation a few weeks back will understand more or less what I'm referring to.

the number of units alone is not impressive, but rather the parameters in which they're doing it.

[blindbox]

No, the reason there is such a variance is that Starswarm was coded especially for Mantle. It's beyond a best case scenario. In 90% of the Mantle titles we see for the next few years, the standard DirectX + Windows driver will be the primary rendering tool while Mantle will be added in. Hence why there is a much lower gain in BF4, the poster child for an actual Mantle-supporting game.



HSA isn't active right now, nor will it come into effect within games. Games actually do a pretty good job of partitioning resources (especially those that support DirectCompute, etc.) already so HSA is mostly focused on compute-focused apps through OpenCL 2.0.

Nitrous engine's (that is, starswarm's engine) directX was done first, and is completely focused on directX, and then was ported to Mantle. It's not written solely for Mantle. They even said that they haven't done any optimization for Mantle. Interestingly, Dan Baker implied he ported it by himself, alone in 2 months.

Link where Dan Baker said it wasn't optimized (warning, long presentation but it's nice to listen to): http://www.youtube.com/watch?v=QIWyf8Hyjbg
Link where he said he ported it to mantle, and said it is a made-for-directX engine. He also said it wasn't optimized in this one. http://www.youtube.com/watch?v=yak25TKCCrc

It's more like a worst case scenario for the kind of games Oxide wishes to enable. The reasons why BF4 is less affected by Mantle is probably because it wasn't optimized for Mantle yet either, and that BF4 is less cpu-hungry compared to starswarm, which is natural for a strategy game (in this case, demo) to be CPU bound.

[Lanek]

If you actually would have looked at the demo you would realize the performance increase has nothing to do with AI or unit count but with the motion blur filter they used or better disabled when running DX.
As said this is very similar to first metro where they also used an ultra expensive motion blur algorithm witch brought down highend cards back then in sub 30fps region.

Also 5000 units is nothing, we have games right now that can handle much more, Star ruler has no unit cap at all and yes all units have AI.




because all this games are designed to be small scale, there are games that can handle much more units without problems, SC (few thoused), star ruler (unlimited), Sins of solar empiers (with mods simple mods in the 10k region, unmoded something like 2-3k)

I ask me where you pull off this ? you think the drop in fps is due to motion blur ? in dx it so much laggy you will not even see motion blur, aobject are then static ... i can tell you this have nothing to do with DOF. ( who is not removed in DX phase anyway ..)

If i understand you well, the Mantle phase is DOF, but not the DX one, and this is why DirectX is so slow ? ... huuum... you really want to find something no ?

Im sorry hornet, but i will let you speak about games developpers for understand what this demo is really involving ....

[blindbox]

I ask me where you pull off this ? you think the drop in fps is due to motion blur ? in dx it so much laggy you will not even see motion blur, aobject are then static ... i can tell you this have nothing to do with DOF. ( who is not removed in DX phase anyway ..)

If i understand you well, the Mantle phase is DOF, but not the DX one, and this is why DirectX is so slow ? ... huuum... you really want to find something no ?

Im sorry hornet, but i will let you speak about games developpers for understand what this demo is really involving ....

He's not pulling it out of his ass. It's from here. Also, what hornet said isn't really attacking Mantle, so back off with the aggressiveness will ya? All he said is that the way Oxide decided to show the difference with Mantle is to disable or enable motion blur. I quite agree with him, since I'd rather see the demo bump the unit count to 10000 and show the difference between Mantle and DirectX. (EDIT: I know I wrote Oxide here instead of DirectX.. my bad)

I do disagree with the rest of his post though. I've checked Total war, I've checked supreme commander, I've checked all the other high unit count games. NONE of them could do it with the performance the Nitrous engine could. NONE of them could have 5000 units all at once, on a single screen, with all the weapons going off and still run at 30 fps.

Though, as we all know, the Star Swarm demo is coming pretty soon. I'd much rather wait for that and test it for myself.

[Lanek]

He's not pulling it out of his ass. It's from here. Also, what hornet said isn't really attacking Mantle, so back off with the aggressiveness will ya? All he said is that the way Oxide decided to show the difference with Mantle is to disable or enable motion blur. I quite agree with him, since I'd rather see the demo bump the unit count to 10000 and show the difference between Mantle and Oxide.

I do disagree with the rest of his post though. I've checked Total war, I've checked supreme commander, I've checked all the other high unit count games. NONE of them could do it with the performance the Nitrous engine could. NONE of them could have 5000 units all at once, on a single screen, with all the weapons going off and still run at 30 fps.

Though, as we all know, the Star Swarm demo is coming pretty soon. I'd much rather wait for that and test it for myself.

Im sorry for the aggresivness, but if you enable DOF ( or motion blur ) you dont decrease the work of the gpu, in fact you increase the work of the gpu.. so this why i was think it is funny to say DX version got is fps plumbed because it dont use motionblur or DOF. ( im not even sure DOF is enabled, disabled, just with 8fps on directX motion blur is going to bug like crazy )..
[QUOTE]

increase has nothing to do with AI or unit count but with the motion blur filter they used or better disabled when running DX

Motion blur or DOF is a computing effect added after the phase of rendering... it change nothing in term of gpu/computing calculation, in fact the effect is added over the other calculation.. ( its only on console games it was used for hide the extreme LOD used on far object and hide clipping ).. The initial computing is the same... > scene > Motion blur or depth of field > render .... scene rendering > DOF ..

Oxyde was nice to disable Blur motion or DOF in this demo for the DX parts.. it will have surely run only at 2 fps with it enabled... not the invert... ( i dont think they have disable it, but a motion " blur " effect is buggy when you run at only 8frames per second )

[Hornet331]

srsly look at the damn video

I just give you a little break down:

0:00-0:46 showing of mantel running with motion blur (20-60fps)
0:46-0:58 running DX with motion blur (10-20fps)
0:58-1:21 running DX without motion blur (50-100fps)
1:12-1:36 reanable motion blur (7fps)
1:36-end back to mantle (15-38fps)

The only thing they changed with the whole scene was enabling or disabling motion blur.
In the end they showed nothing more then how crappy there chosen motion blur effect runs on DX and how "superior" it is on mantle.

For me that seems awful similar to how devs have implemented physx in many games.

I do disagree with the rest of his post though. I've checked Total war, I've checked supreme commander, I've checked all the other high unit count games. NONE of them could do it with the performance the Nitrous engine could. NONE of them could have 5000 units all at once, on a single screen, with all the weapons going off and still run at 30 fps.

Look at star ruler.

[blindbox]

Look at star ruler.

I just did that.

http://www.youtube.com/watch?v=ExvDTQT8Lts

2D LODs? There's not that many 3D objects there. Got a better video, or can they be disabled?

[Hornet331]

I just did that.

http://www.youtube.com/watch?v=ExvDTQT8Lts

2D LODs? There's not that many 3D objects there. Got a better video, or can they be disabled?

nah its 3d, what you look here is the zoomed out state (it has seamless zooming from ship level to galaxy level) where ships are represented as an icon (because they are to small to see otherwise)

http://www.youtube.com/watch?v=44ZIN9BVkqY
http://www.youtube.com/watch?v=0rjwZPRnv08

its also not like the demo of the oxide engine drops to sub 20fps (even with mantle) when there are many ships on the screen at once.

[blindbox]

nah its 3d, what you look here is the zoomed out state (it has seamless zooming from ship level to galaxy level) where ships are represented as an icon (because they are to small to see otherwise)

http://www.youtube.com/watch?v=44ZIN9BVkqY
http://www.youtube.com/watch?v=0rjwZPRnv08

its also not like the demo of the oxide engine drops to sub 20fps (even with mantle) when there are many ships on the screen at once.

The videos you are showing is running at less than 20 fps, with just the ships. The first demo is 15000 units, with less than 10 fps when the game is running at 1.00x speed. The second demo lags just because of the capital ships alone. Oxide's demo is running at 20 fps minimum, worst case, with motion blur on.

EDIT: All I'm saying is, I haven't seen a scene like this, with that high of an FPS, in any strategy games. http://i.imgur.com/p8ARa78.png

[Stukov]

No, the reason there is such a variance is that Starswarm was coded especially for Mantle. It's beyond a best case scenario. In 90% of the Mantle titles we see for the next few years, the standard DirectX + Windows driver will be the primary rendering tool while Mantle will be added in. Hence why there is a much lower gain in BF4, the poster child for an actual Mantle-supporting game.

So you postulate that in BF4 when Mantle is finished you still will see a large variance in performance based on the capability of the CPU? For example if the 4770k is 10% faster than the 8350 right now, it is going to remain 10% faster once Mantle is up and running?

[Mechanical Man]

I find it odd, that nobody is speaking about minimum fps at all. I expect Mantle to up minimum fps on high end machines, even tought maximum to be about the same. Basically, what i expect is. Mantle will enable more games to be playable with AMD APU's and games have more smoother fps profile for high end cpus & cards (less drops and better minimum fps). I dont see anyone losing there. I do hope nVidia dont be too arrogant and atleast really tries to negotiate with AMD to enable Mantle on nVidia GPU's too. That would benefit us all.

[SKYMTL]

So you postulate that in BF4 when Mantle is finished you still will see a large variance in performance based on the capability of the CPU? For example if the 4770k is 10% faster than the 8350 right now, it is going to remain 10% faster once Mantle is up and running?

I honestly have no idea. It all depends on how many functions they can actively carry over to Mantle.

[SKYMTL]

I find it odd, that nobody is speaking about minimum fps at all. I expect Mantle to up minimum fps on high end machines, even tought maximum to be about the same. Basically, what i expect is. Mantle will enable more games to be playable with AMD APU's and games have more smoother fps profile for high end cpus & cards (less drops and better minimum fps). I dont see anyone losing there. I do hope nVidia dont be too arrogant and atleast really tries to negotiate with AMD to enable Mantle on nVidia GPU's too. That would benefit us all.

I actually want to know if AMD's frame pacing can be easily carried over into Mantle apps. If that doesn't happen, minimum FPS or not, the experience will be a disaster at higher framerates.

[Mechanical Man]

I actually want to know if AMD's frame pacing can be easily carried over into Mantle apps. If that doesn't happen, minimum FPS or not, the experience will be a disaster at higher framerates.

Based on what?

[Hornet331]

EDIT: All I'm saying is, I haven't seen a scene like this, with that high of an FPS, in any strategy games. http://i.imgur.com/p8ARa78.png

that not high to begin with, 35fps with what 2-3k untis?

the examples i have given you have 4-5 times the unit count. :P

Dont be surprised to see the same crawling when they actually have to show the same 10k+ count.

[blindbox]

that not high to begin with, 35fps with what 2-3k untis?

the examples i have given you have 4-5 times the unit count. :P

Dont be surprised to see the same crawling when they actually have to show the same 10k+ count.

That's no more than a 2~3k on the second video. The first video is very, extremely, LODed, and is having crap FPS just from the units being shown.. The only thing we've have proven is that, you can't compare the Star Swarm demo with the current games out there. None of your videos managed to show what Oxide showed, and neither did Oxide show anything that is remotely related to the video you showed. Sure, 5000 units is nothing, but come back when I see that in a third person view of a spacecraft in the middle of the action, 30 fps minimum.

[Stukov]

I honestly have no idea. It all depends on how many functions they can actively carry over to Mantle.

I'm not saying you can't be correct about the "native Mantle" idea you presented, but what I was saying based on the CPU bottleneck, the situation I presented the slower CPU is going to benefit more % wise from Mantle, including games where they are more CPU bound than GPU (such as RTS's).

[Iconyu]

[/B]
Based on what?

Right now in crossfire AMD cards can sometimes throw frames at a monitor so fast that it chokes on them. And when I say choke, I mean the monitor isn't always getting a chance to put a full frame on screen as two frames are being rendered way too close together, then there's a big gap and again two frames too close together. The card can produce 60+ FPS, but it sends them out in a way that make the monitor only show 30fps and a load of junk frames.

[Mechanical Man]

Right now in crossfire AMD cards can sometimes throw frames at a monitor so fast that it chokes on them. And when I say choke, I mean the monitor isn't always getting a chance to put a full frame on screen as two frames are being rendered way too close together, then there's a big gap and again two frames too close together. The card can produce 60+ FPS, but it sends them out in a way that make the monitor only show 30fps and a load of junk frames.

Wtf, nobody was talking about crossfire. You cant just quote someone talking about oranges and start talking about shovels ffs.