AMD Mantle update...

[Jethro]

I hope people are patient with Mantle. It seems to have much potential.

[ilkkahy]

I hope people are patient with Mantle. It seems to have much potential.

Amen to that.

[phantomferrari]

I hope people are patient with Mantle. It seems to have much potential.

i agree, people tend to forget that mantle is being tested against the 11th version of directx. Considering that mantel looks very promising.

[kuroikenshi]

its not even that....
something that's just come out of the developers den is already competing and beating something which has been refined for well over a decade.

when you put things into perspective it becomes apparent quite quickly the potential of Mantle.

only if the "I want it now and I want it FREE" generation stopped complaining.....

[MaddMuppet]

its not even that....
something that's just come out of the developers den is already competing and beating something which has been refined for well over a decade.

when you put things into perspective it becomes apparent quite quickly the potential of Mantle.

only if the "I want it now and I want it FREE" generation stopped complaining.....

But they expect everything now and don't want to work hard to get it.

[kromosto]

still the question is why we didn't see performance gains in bf4 with 7xxx series. optimizations ok but will all the overhead reduced by mantle we should have seen some. we need a good explanation from amd about this. i believe this answer will make us understand mantle better.

[blindbox]

still the question is why we didn't see performance gains in bf4 with 7xxx series. optimizations ok but will all the overhead reduced by mantle we should have seen some. we need a good explanation from amd about this. i believe this answer will make us understand mantle better.

Simple, DICE optimized it more towards GCN 1.1 than GCN 1.0. Heck, DICE themselves said it's not optimized for 7xxx. So you're questioning something that has been answered.

[mutantmagnet]

Simple, DICE optimized it more towards GCN 1.1 than GCN 1.0. Heck, DICE themselves said it's not optimized for 7xxx. So you're questioning something that has been answered.

You misunderstood his point. At a fundamental level mantle should enable CPU overhead reductions before optimizations (he even used that word). He is hoping that if AMD provides an explanation we'll have a better understanding of why architecturally speaking GCN 1.0 can't (as in possibly unable to under any circumstances) leverage mantle like GCN 1.1.

[blindbox]

You misunderstood his point. At a fundamental level mantle should enable CPU overhead reductions before optimizations (he even used that word). He is hoping that if AMD provides an explanation we'll have a better understanding of why architecturally speaking GCN 1.0 can't (as in possibly unable to under any circumstances) leverage mantle like GCN 1.1.

Perhaps, but do you have any numbers for GCN 1.0 coupled with a weak GPU and compared direct3d and Mantle against each other to back that up? All the reviews I see on the Internet are on 290x. In my defense, this sentence -> "optimizations ok but will all the overhead reduced by mantle we should have seen some" is pretty damn hard to understand.

EDIT: Actually, here you go. http://battlelog.battlefield.com/bf4...4-mantle-live/ . GCN 1.0 is seeing performance increase with CPU limited scenarios. I'm pretty sure he meant GPU-limited scenarios, then.

Also, someone needs to ask what is Johan's driver's smoking for it to be able to run CFX. Many are saying that CFX isn't working with Mantle.

[kromosto]

Thank you mutantmagnet believe me i can't express it anywhere near that.

[blindbox]

I answered you as well as of above (sorry for being edit-happy). GCN 1.0 works well with Mantle for reducing CPU overhead, contrary to your belief.

[DefStar]

From blindbox's link, "Level was tested with 64 ?pseudo players? that we have for our own internal testing that simulates heavy game workload that we have in multiplayer in order to get more deterministic results compared to full real multiplayer. "

We need that 'test', a deterministic performance analysis tool for MP would be great for comparing system configuration results.

I mean obviously AMD if you want us to be able to see how Mantle is improving performance we need to use the Scientific Method which requires a deterministic test...I would have totally took the time to bench my system and publish/post the results for everyone one to see = free advertising...AMD, you listening?

The test would require a server sending pre-determined updates/pseudo player actions to a client were you would bench from...so we would probably need a server software running to do this.

[Fottemberg]

its not even that....
something that's just come out of the developers den is already competing and beating something which has been refined for well over a decade.

when you put things into perspective it becomes apparent quite quickly the potential of Mantle.

only if the "I want it now and I want it FREE" generation stopped complaining.....

So true.

Microsoft DirectX became competitive against 3DFx Glide only with the 5th revision.

[behrouz]

Why does Mantle use so much Memory ? it's Terrible!!

[KiSUAN]

Why does Mantle use so much Memory ? it's Terrible!!

If it doesn’t something else, who cares?? Besides, now average Joe will be making the right choice when choosing a VGA with 3483484334 of RAM!!!!

[DefStar]

It was just to keep an eye on temps and compare memory usage and GPU usage compared to DX11.

Yup, I always use afterburner's OSD to watch my temps to...i'm always paranoid my water pump will give out since it is so quite I can't here it.


@STEvil, yeah...14.2 drivers will hopefully take care of us GCN 1.0/7000 series...I'm pretty confident they will as AMD stated they will, its just takes time to optimize and bug test. Ever since they released that Frame Pacing driver I've been pretty impressed with AMDs driver team.

blindbox's link shows that they have optimized 7000 series some, but no one else (I've heard of) has been able to use that test to verify those results.

[Lanek]

Why does Mantle use so much Memory ? it's Terrible!!

Hopefully it use the right memory usage, linked to cores / threads use.. lol....

[Stukov]

From blindbox's link, "Level was tested with 64 ?pseudo players? that we have for our own internal testing that simulates heavy game workload that we have in multiplayer in order to get more deterministic results compared to full real multiplayer. "

We need that 'test', a deterministic performance analysis tool for MP would be great for comparing system configuration results.

I mean obviously AMD if you want us to be able to see how Mantle is improving performance we need to use the Scientific Method which requires a deterministic test...I would have totally took the time to bench my system and publish/post the results for everyone one to see = free advertising...AMD, you listening?

The test would require a server sending pre-determined updates/pseudo player actions to a client were you would bench from...so we would probably need a server software running to do this.

I believe you would need multiple systems to create such a repeatable test. I don't know how many you would need, but essentially you would have to have multiple copies of BF4 running, run scripts to make bots run around in the same way each time, then run a script for the system you are benchmarking. It would be rather complex and expensive. The less precise way, but much easier, is to do multiple long tests of someone playing BF4 and doing an average of these sessions to give you a number. Over a long enough period of time you should be able to iron out most non-repeatable elements and get a ballpark figure for performance comparison.

[DefStar]

I believe you would need multiple systems to create such a repeatable test. I don't know how many you would need, but essentially you would have to have multiple copies of BF4 running, run scripts to make bots run around in the same way each time, then run a script for the system you are benchmarking. It would be rather complex and expensive. The less precise way, but much easier, is to do multiple long tests of someone playing BF4 and doing an average of these sessions to give you a number. Over a long enough period of time you should be able to iron out most non-repeatable elements and get a ballpark figure for performance comparison.

No, if you wrote a program that 'mimicked' a server and that program just followed a script and always sent the same information to the client...then you would have a deterministic benchmark. You wouldn't need multiple clients logged into the server. Think of how Sacred 2 runs a server and a client on your pc at the same time. On the client end, the client would not even have to interact with anything...just receive server messages and benchmark the session. How Dice does it I have no Idea.

I was just thinking of doing a few DX11 server matches and then a few Mantle server matches and then taking the average of those...or within the same match, a few 3 minutes periods of DX11 and Mantle alternating between the two and then taking the averaGE of those. But I shouldn't need to go through that much trouble to bench mark Mantle when it is AMD who needs to promote Mantle and thus needs to provide us with deterministic analysis tools...AMD providing a test allowing the community to promote and advertise would not be expensive, it would be beneficial to AMD.

[kromosto]

I answered you as well as of above (sorry for being edit-happy). GCN 1.0 works well with Mantle for reducing CPU overhead, contrary to your belief.

i don't believe so. Most of the users sees no difference (like me), for some problems begin (as expected for a beta) and a little reported performance gains but they can be wrong (because of testing methods) or just exceptions. also at the begining ea excluded 7970 and 280x from mantle support but then amd came and said no we will it will support them but we are going to make optimizations for them seems like a confession to me about 7xxx series. but actually all these are not that important for general mantle impression on mantle but i want to know what is happening to understand mantle better (as much as i can of course).

[Hornet331]

http://www.youtube.com/watch?v=-bCeNzgiJ8I

Its a presentation from the steamworks devs days talking about driver overhead reduction on ogl, its by nvidia but its pretty general about ogl.

If its get things done as nice as the presentation suggests mantel is pretty much meaningless, since ogl is multi platform, supported by pretty much every major graphics vendor and much on the instructions are already in current hardware and only needs to be exposed by the driver.

[mongoled]

http://www.youtube.com/watch?v=-bCeNzgiJ8I

Its a presentation from the steamworks devs days talking about driver overhead reduction on ogl, its by nvidia but its pretty general about ogl.

If its get things done as nice as the presentation suggests mantel is pretty much meaningless, since ogl is multi platform, supported by pretty much every major graphics vendor and much on the instructions are already in current hardware and only needs to be exposed by the driver.

Thanks for bringing tht up but it has ABSOLUTELY NOTHING to do with the thread title.

If you want to discuss nvidia ogl then I strongly suggest you start a new thread.

Bringing this point up in this thread is simply PROVOKING trouble.

[Hornet331]

yeah it has nothing to do with except that this is exactly the same thing as mantle does, reducing driver overhead and giving the developers easier times in optimize there games/applications.

Hey and its not like OGL hasn't already been mentioned and compared to mantle over a dozen times in this thread.

[vario]

yeah it has nothing to do with except that this is exactly the same thing as mantle does, reducing driver overhead and giving the developers easier times in optimize there games/applications.

Hey and its not like OGL hasn't already been mentioned and compared to mantle over a dozen times in this thread.

You REALLY think that, if OpenGL allowed such low level to the metal coding AMD would even bother with this whole mantle thing ?

If someone thinks logically, only real answer is, no.So, it must mean that AMD could not get same results with OpenGL.

Also as for GCN 1.0 optimizations, to me it looks its right there, not as stellar as GCN 1.1 but still , with a weak cpu you will see gains.

[MaddMuppet]

You REALLY think that, if OpenGL allowed such low level to the metal coding AMD would even bother with this whole mantle thing ?

If someone thinks logically, only real answer is, no.So, it must mean that AMD could not get same results with OpenGL.

Also as for GCN 1.0 optimizations, to me it looks its right there, not as stellar as GCN 1.1 but still , with a weak cpu you will see gains.

I think you missed his whole point.