Flaws on Oxide

[Imouto]

It's not my fault that you dont believe the Mantle-hype-benchmark-tool.

Or do you mean that i lie?

I mean I don't trust you and it's fine if you don't trust me, that's why we have review sites. Details, details.

 

[sontin]

So, you need review sites for using an benchmark released on Steam?

Sure, sure.

 

[Noctifer616]

So, you need review sites for using an benchmark released on Steam?

Sure, sure.

It's not a benchmark. You need to set it up correctly to run it as a benchmark, otherwise every run is going to give you different results.

 

[sontin]

Yes, say it to the reviewer who are using the templates.

 

[Enigmoid]

You post a thread with the title "Flaws on Oxide" - (And i guess you mean "Star Swarm" that is the engine, not Oxide that is the name of the company)

Yet you dont know the most basic about this test and engine, as this question shows. Next time take some time to try to understand and read about the subject, or post it as questions.

No I mean Oxide the company as they are the ones who designed the engine and made the (what I believe) outrageous claims.

Star Swarm is broken IMO but when you design a broken engine (like NFS rivals 30 fps cap) then something is wrong with the developers (when they make their claims).

 

[krumme]

No I mean Oxide the company as they are the ones who designed the engine and made the (what I believe) outrageous claims.

Star Swarm is broken IMO but when you design a broken engine (like NFS rivals 30 fps cap) then something is wrong with the developers (when they make their claims).

They have worked for a year with Swarm to get this performance on DX, and they probably got the basic technology beforehand. Its build for DX not mantle and imho it shows in spades when you understand what is actually happening.

Its an engine without a lead thread. Nobody seems to grasp the consequences of that. Its perhaps an even bigger innovation than Mantle imho. Nobody eg. even Johan thought that was possible. Go see the APU13 Q&A when Johan gets corrected by one of the Oxide guys. Its that innovative and hopefully in the comming years this methology can spread.

Its not the slightest broken. An RTS is just typically drawcall limited. It have always been like that and it defines how RTS games can be. The difference you get here of 300% is not outrageous. There can be situations when mantle is 1000% faster. Lower the gfx quality but get a 10 times as big army and dx just falls like a rock. It have always been like that, and this i no execption. But this engine just handles far more than usual dx engines because there is no lead thread.

 

[Shivansps]

This engine just do dumb things, rendering things you cant see you are hitting the GPU wall on there FAST, for what?

The RTS test is a JOKE, im 100% gpu by seeing dots and a static background! plus 2D explosions...
Nothing of that can by applied to a real game, its intencionally unoptimised, some times you guys are forgeting that the gpu also have to render the useless stuff.

HELL, Sins of Solar Empire Rebellion handles large fights WAY better by reemplacing stuff you cant really see on the distance. Im seeing Star Swarm and i just want to bash my head to the wall.

 

[sontin]

Its not the slightest broken. An RTS is just typically drawcall limited. It have always been like that and it defines how RTS games can be. The difference you get here of 300% is not outrageous. There can be situations when mantle is 1000% faster. Lower the gfx quality but get a 10 times as big army and dx just falls like a rock. It have always been like that, and this i no execption. But this engine just handles far more than usual dx engines because there is no lead thread.

With an nVidia card and DC the difference comes down to 20% or so.
It looks more and more that Mantle is a fix for AMD's poor DX driver quality.

 

[DaZeeMan]

As I shared two pages ago, Brad from Oxide/Stardock mentioned that the point of this demo is to stress test hardware. Hence all of the 'unnecessary' rendering.

He also mentioned that games using the Nitrous Engine would of course be much better optimized for current hardware, and be much more efficient with system resources.
My post sharing Brad's comments here:
http://forums.anandtech.com/showpost.php?p=36019816&postcount=114

He also suggested that Nvidia and Intel may release their own hardware optimizations in the coming year, hence improving their own results with this benchmark.


Also, these guys have some DX11 versus Mantle numbers with Star Swarm:
http://translate.google.com/transla...ssor-katalysator-1402-104261-4.html&sandbox=1

Essentially a 2x to 3x improvement by switching to Mantle. i7-3770K and R9-290x.

One of the themes of Mantle is to take DX11 out of the equation when possible, and this stress test helps highlight what bypassing DX11 can mean for system performance.. There is absolutely nothing stopping Intel and NVidia from doing the same, should they so choose.

The Oxide guys have their own goals, and their own opinions and ideas as to what is currently holding game engines back. They see DirectX as unnecessary overhead in the long run - and (from what I've gathered) would absolutely love to see Intel and NVidia follow AMD's lead, with their own 'lower level' solutions. Whether they are right or wrong of course will be borne out in the coming years...

 

[el etro]

Just one more reason why I don't trust TPU, and I never take their reviews seriously. Their drivers are always out of date.

Anyway, why do their benchmarks differ so much from PClabs.pl's benchmarks?

I bashed TPU tests every single day here.

All PClabs tests i see never get any positive conclusion pro-AMD.


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



BTW, one way to see Oxide Benchmark is as a benchmark that is fully intended to test the habilty of Mantle to reduce CPU bottlenecks. Not that i think that the 300% increase in performance would apply to any game.... we wait to wait a more professional article about it to take out conclusions.

 

[Erenhardt]

I bashed TPU tests every single day here.

All PClabs tests i see never get any positive conclusion pro-AMD.

As a side note to put some light on the pclab reviews. Some time ago (2010 or so) they didn't get their review samples from AMD. They went all out war, publicly complaining about it (originPC style). Since then not a single amd review had positive conclusion.

We call it PCLamb here for a reason.

 

[Enigmoid]

As I shared two pages ago, Brad from Oxide/Stardock mentioned that the point of this demo is to stress test hardware. Hence all of the 'unnecessary' rendering.

If the purpose is to stress hardware, rather than show a realistic senario then any kind of DX vs. Mantle comparison simply can't be made.

 

[Paul98]

This engine just do dumb things, rendering things you cant see you are hitting the GPU wall on there FAST, for what?

The RTS test is a JOKE, im 100% gpu by seeing dots and a static background! plus 2D explosions...
Nothing of that can by applied to a real game, its intencionally unoptimised, some times you guys are forgeting that the gpu also have to render the useless stuff.

HELL, Sins of Solar Empire Rebellion handles large fights WAY better by reemplacing stuff you cant really see on the distance. Im seeing Star Swarm and i just want to bash my head to the wall.

What sort of GPU usage would you expect rendering a single cube to the screen in directX?

 

[Erenhardt]

These complains about full simulation of objects that end up being a dot on the screen made me thing about utilizing all those draw calls.

This may improve CPU physX performance quite a lot. As I understand, particle physx requires highly parallel (...-ized...?) calculations. Looks like mantle is changing CPU capabilities from narrow sequential to wide parallel simulation.

 

[sontin]

These complains about full simulation of objects that end up being a dot on the screen made me thing about utilizing all those draw calls.

This may improve CPU physX performance quite a lot. As I understand, particle physx requires highly parallel (...-ized...?) calculations. Looks like mantle is changing CPU capabilities from narrow sequential to wide parallel simulation.

CPU-PhysX doesn't use D3D. So there is no DX limitation.

BTW: I found some numbers from someone with a7970, the press driver and Star Swarm:

Core i7-920 @ 3,6 GHz, 16 GiB DDR3-1600; HD 7970 @ 1125/1600 MHz

http://www.forum-3dcenter.org/vbulletin/showpost.php?p=10093981&postcount=2793

API:                Mantle
Scenario:            ScenarioAttract.csv
User Input:            Disabled
Resolution:            1920x1080
Fullscreen:            True
GameCore Update:        16.6 ms
Bloom Quality:            High
PointLight Quality:        High
ToneCurve Quality:        High
Glare Overdraw:            16
Shading Samples:         64
Shade Quality:            Mid
Deferred Contexts:        Disabled
Temporal AA Duration:        16
Temporal AA Time Slice:        2
Detailed Frame Info:        Off
===========================================================


== Results ================================================
Test Duration:            360 Seconds
Total Frames:            18023

Average FPS:            50.06
Average Unit Count:        4529
Maximum Unit Count:        5693
Average Batches/MS:        837.92
Maximum Batches/MS:        2266.09
Average Batch Count:        21014
Maximum Batch Count:        201229
===========================================================

And here my numbers with a nVidia card and DC:

== Results ================================================
Test Duration:            360 Seconds
Total Frames:            15847

Average FPS:            44.01
Average Unit Count:        4385
Maximum Unit Count:        5813
Average Batches/MS:        921.37
Maximum Batches/MS:        2839.25
Average Batch Count:        21846
Maximum Batch Count:        155057
===========================================================

Except Maximum Batch Count every other numbers are nearly equal.

So Mantle brings in a Mantle showcase test ~14% more performance.

Call me unimpressed.

 

[el etro]

As a side note to put some light on the pclab reviews. Some time ago (2010 or so) they didn't get their review samples from AMD. They went all out war, publicly complaining about it (originPC style). Since then not a single amd review had positive conclusion.

We call it PCLamb here for a reason.

where there's smoke, there's fire!

 

[Paul98]

For those using DC and getting good results, did your CPU and GPU usage change?

 

[sontin]

CPU usage is higher with more load on all my 4 Sandy Bridge cores.

 

[rtsurfer]

As a side note to put some light on the pclab reviews. Some time ago (2010 or so) they didn't get their review samples from AMD. They went all out war, publicly complaining about it (originPC style). Since then not a single amd review had positive conclusion.

We call it PCLamb here for a reason.

Please cross post in the Mantle thread.

So we can open eyes of some people.

 

[sontin]

For those using DC and getting good results, did your CPU and GPU usage change?

And here with a comparision between boths - top is with DC, bottom without:

It looks much better with DC. I haven't experienced any problems with it.

 

[Imouto]

computerbase.de did some benching.

http://www.computerbase.de/artikel/grafikkarten/2014/erste-eindruecke-zu-amds-mantle/3/

I asked them already to rerun the GTX 780 Ti with deferred contexts enabled. If you have an account you should ask too as they seem to interact in english in their Twitter.

 

[Carfax83]

Now seriously, you're the most clueless guy I've ever talked to. What's the novelty on it? We had this 15 years ago with Legacy of Kain: Soul Reaver on the PSX.

And how many chapters does LoK Soul Reaver have? The Witcher 3 will have no chapters or artificial breaks or transitions..

Also, to even bring up a game like LoK Soul Reaver in the context of the discussion we're having (which is about the worth of DX11) is ridiculous, as the Witcher 3 has orders of magnitude more graphical fidelity and complexity

Much maths? 50% ahead in a 7970Ghz vs GTX 780 which is already 20% faster overall? If we push the 7970Ghz a 20% to normalize things the GTX 780 would be 22% faster.

Yeah, cause testing stock vs overclocked makes a lot of sense.

Anyway the fair comparison is the 7970Ghz vs the GTX 770 which is 15% faster.

It doesn't matter. The whole point of me bringing that up, was to show you how DX11 multithreading reduces CPU bottlenecks. The best card to show you that was the fastest card of the bunch, as it was the most CPU bottlenecked..

I guess that flew over your head though..

If we compare the GTX 780 to the R9 290(nonX) the advantage shrinks to 9%.

As you can see, the benchmark itself is limited to a certain frame rate. The Titan had a lower frame rate than the GTX 780..

That's probably why Anandtech retired the benchmark, because contemporary cards are just too fast.

So much for that glorious DX MT.

That's OK. NVidia pummeled the living daylights out of AMD in Civilization 5 for how many years now thanks to DX11 multithreading :whiste:

You're talking like that game is finished or was finished almost a year ago when that vid was taken. Again no biggie, those Splinter Cell games prove it.

Who cares if it's finished or not. The game supports DCL, something which AMD does not have, and likely will never have at the rate they're going since they're dumping all of their resources into Mantle..

The sole reason for that 55% deficit in performance was DCL.. If anything, the gap will likely widen when the game is finished due to more powerful hardware, and more optimized drivers.

I'd wait for Mantle reviews, then we talk about who's living in the past.

Well Mantle reviews are here, and according to PClabs.pl, NVidia was still faster :awe:

 

[Carfax83]

It shows a bigger die, and much more expensive overclocked card winning by 28%.

You guys just don't get it. The reason why I cited the GTX 780 Ti was because it's the most CPU bottlenecked card in that lineup. It has nothing to do with being fair, or unfair. It's just making a point using the best example.

Also, who cares about a bigger die. The GTX 780 Ti still uses less power than the 290x, while providing greater performance.

The 290x when prices are not silly is more in league with the gtx780.

You're really missing the entire point of this conversation..

 

[DarkKnightDude]

With mantle.

Star Swarm: DX11 28 | Mantle 56

Not that it means much.

 

[el etro]

With mantle.

Star Swarm: DX11 28 | Mantle 56

Not that it means much.

Could mean that Star Swarm is a benchmark that only relies on draw calls and its purpose is to show how much Mantle can be better than DirectX in situations with a insane number of characters in the screen.