The AMD Mantle Thread

[Paul98]

I don't see where it says that...?

"About half an hour ago, AMD sent us the drivers for our own performance evaluations."

Looks like we will be getting mantle tomorrow

 

[jackstar7]

"About half an hour ago, AMD sent us the drivers for our own performance evaluations."

Looks like we will be getting mantle tomorrow

If not tomorrow, then possibly Monday. We'll have to see.

 

[Paul98]

I wonder if these trolls are being paid, I don't know why anyone would spend so much time talking about something that they aren't interested in.

Warning issued for thread crapping.
-- stahlhart

 

[tonyfreak215]

I wonder if these trolls are being paid, I don't know why anyone would spend so much time talking about something that they aren't interested in.

I agree. It's crazy. It's like AMD stole their lunch money or something.

Warning issued for thread crapping.
-- stahlhart

 

[frozentundra123456]

I have typed this same response to this backwards compatibility nonsense and no one ever responds. Please tell me why mantle needs to be backwards compatible?

AMD cards are not dropping support for direct x. They are backwards compatible. This is honestly the worst thought out fud in a thread full of fud. The important thing for AMD and Mantle is going to be forward compatibility not backwards.

Mantle will do nothing to change the performance of older titles where AMD has performance parity across price ranges with nvidia already because they support direct x and will continue to. Just stop posting about backwards compatibility. It has no place in this debate because it is a non issue.

I will post whatever I feel like, thank you, unless the mods tell me it is inappropriate. As to why it needs to be backwards compatible, the answer is obvious: because to become a truly wide ranging API it needs to be able to provide improved performance to *all* games, not just a select few on certain hardware, under certain settings. As long as that is all it can do it is a very limited benefit. It can be a "non-issue" to you if you want it to be, but it certainly isnt to me.

 

[james1701]

Is there any press live streaming their tests of the beta driver?

 

[Paul98]

I will post whatever I feel like, thank you, unless the mods tell me it is inappropriate. As to why it needs to be backwards compatible, the answer is obvious: because to become a truly wide ranging API it needs to be able to provide improved performance to *all* games, not just a select few on certain hardware, under certain settings. As long as that is all it can do it is a very limited benefit. It can be a "non-issue" to you if you want it to be, but it certainly isnt to me.

You can't make a new API and have it magically applied to old games. If you create a new API no matter what hardware compatibility games will need to be written with that API.

 

[VulgarDisplay]

I will post whatever I feel like, thank you, unless the mods tell me it is inappropriate. As to why it needs to be backwards compatible, the answer is obvious: because to become a truly wide ranging API it needs to be able to provide improved performance to *all* games, not just a select few on certain hardware, under certain settings. As long as that is all it can do it is a very limited benefit. It can be a "non-issue" to you if you want it to be, but it certainly isnt to me.

Which is why it needs to be forward compatible with upcoming games and GPU generations. Stating that it needs to work with every game ever made is nonsense. Did Microsoft update every game prior to direct x to use direct x? No they didn't and it became the industry standard even though it wasn't backwards compatible.

Direct x instead steadily found its way into more and more games and eventually became the graphics api of choice. AMD needs to focus on getting mantle into more and more engines and games moving forward. The past is the past and their direct x performance is ample enough to cover the need for backwards compatibility.

Direct X won an uphill battle at a time where every game had its own engine and there were far more GPU makers all with their own api's even without backwards compatibility. AMD has 2 vendors and probably less than 10 engines to deal with in their quest to become an industry standard. I like their chances of becoming widespread more than direct x in its infancy. The only factor that remains in my opinion is how the other of the GPU duopoly decides to deal with mantle.

 

[Venomous]

Mantle surely isn't backwards compatible for 7xxx owners because AMD already stated gcn 1.0 showed degraded performance and it will be addressed later. So, tell me about backwards compatibility again?

 

[sushiwarrior]

Mantle surely isn't backwards compatible for 7xxx owners because AMD already stated gcn 1.0 showed degraded performance and it will be addressed later. So, tell me about backwards compatibility again?

GCN1.0 still shows gains but not the same as GCN1.1/2.0 gains, it's just a matter of driver tuning as far as I know. Just some of the initial kinks really, drivers went from having negative scaling to having positive scaling within a single day so give it another update and it should be OK.

 

[Adampa1006]

I think what's being missed is that mantle addresses the fact that CPU performance has stagnated, especially in comparison to GPU performance. GPUs are 100-150% faster than they were 3-4 years ago, while CPUs are only 10-20% better max. And AMD has given up on the high end CPUs, while Intel is pretty much doing the same by focusing on power saving and on board GPUs, which no one cares about on high end.

Mantle plays right into this.

 

[bystander36]

I think what's being missed is that mantle addresses the fact that CPU performance has stagnated, especially in comparison to GPU performance. GPUs are 100-150% faster than they were 3-4 years ago, while CPUs are only 10-20% better max. And AMD has given up on the high end CPUs, while Intel is pretty much doing the same by focusing on power saving and on board GPUs, which no one cares about on high end.

Mantle plays right into this.

If Mantle can gain major performance improvements by being multi-threaded, then CPU's can make rapid improvements in the same way that GPU's have, by adding more cores. The problem is, many CPU tasks are not well suited to parallel processing. Mantle does give a way for the CPU to keep up, though it would be best if all games started doing this. Getting DX to do similar things would go a long ways to giving CPU's a way to improve speeds much like GPU's do.

 

[Carfax83]

I think what's being missed is that mantle addresses the fact that CPU performance has stagnated, especially in comparison to GPU performance. GPUs are 100-150% faster than they were 3-4 years ago, while CPUs are only 10-20% better max. And AMD has given up on the high end CPUs, while Intel is pretty much doing the same by focusing on power saving and on board GPUs, which no one cares about on high end.

Mantle plays right into this.

I think you're crisscrossing things a bit. Mantle does not address CPU performance, it merely reduces the API overhead and makes things more streamlined, thus enabling faster communication with the GPU.

The CPU was never the problem. The problem was the overhead from the API.

Anyway, the CPU's workload is totally different from the GPU. With a GPU, you can nearly double your resources with a die shrink, and get nearly double the performance because the GPU's workload lends itself quite well to parallelism.

You can't do that with a CPU though, as their workload is more serial in nature.

 

[ASM-coder]

As long as that is all it can do it is a very limited benefit. It can be a "non-issue" to you if you want it to be, but it certainly isnt to me.

And you won't buy a CGN card, and continue to be very happy without Mantle. I got no problem with that. I am interested in the possibilities of Mantle, and apparently a few others range from curious to interested to excited to starry-eyed. I have no problem with whatever anyone wants to believe. FWIW... I am somewhere between interested and excited.

 

[3DVagabond]

Reads like a philanthropist mission statement. I dont think any company who developes technology and hardware is not thinking the same thing, move the industry forward. I do think that a lot of it has to do with offering an incentive to purchase a product first and to be a savior second.

The two are not necessarily exclusive. It just takes looking beyond the present tense to realize it.

 

[ASM-coder]

So I am reading another new online story about Mantle which references the still only available BF4 benchmarks from Dice. Talk about bias... "AMD has decided to cobble a FX-8350 CPU with the high-end R9 290X to demonstrate the benefits Mantle would have."

First of all, these are NOT AMD benchmarks. These are from Dice, from the lead developer at Dice.

Secondly, cobbled? 'To make or put together roughly or hastily' So Johan
dug around in his rubbish bin and said... "what can we cobble together to make Mantle look really good?"

I guess nobody uses such a cruddy $200 CPU, and would be looking for a serious upgrade by buying a new $$$ GPU. Better to spend $$ on a new mobo and $$ on a high end CPU but stay with your current cruddy GPU. There must be an instruction manual somewhere that spells out proper CPU/GPU configurations and would keep me from making an embarrassing configuration faux pas. And no white GPUs past labor day.

 

[Abwx]

It is interesting how the narrative from Mantle went from outrageous performance gains from 1000000 draw calls

Can you point us a link where and when they claim 1000000
draw calls ; i m quite sure that not only you could not provide
any source but still worse , this number doesnt come from AMD
but from here :

http://forums.anandtech.com/showthread.php?t=2366071

Actualy it s you who is making outrageous claims...

 

[Venomous]

So I am reading another new online story about Mantle which references the still only available BF4 benchmarks from Dice. Talk about bias... "AMD has decided to cobble a FX-8350 CPU with the high-end R9 290X to demonstrate the benefits Mantle would have."

First of all, these are NOT AMD benchmarks. These are from Dice, from the lead developer at Dice.

Secondly, cobbled? 'To make or put together roughly or hastily' So Johan
dug around in his rubbish bin and said... "what can we cobble together to make Mantle look really good?"

I guess nobody uses such a cruddy $200 CPU, and would be looking for a serious upgrade by buying a new $$$ GPU. Better to spend $$ on a new mobo and $$ on a high end CPU but stay with your current cruddy GPU. There must be an instruction manual somewhere that spells out proper CPU/GPU configurations and would keep me from making an embarrassing configuration faux pas. And no white GPUs past labor day.

I think when they begin building a new game , thought about who will be playing them, age group they reside in and variables of hardware configurations, play an important role. You want to sell as many titles you can to maximize profit due to the investment.

With saying that, it was probably a surprise moment seeing how well this API did on low level hardware. Meaning, just about anyone with a modern 4-6 yr old system could enjoy the game.

I think that is what really boiled it down for them to say, we have something here. I don't think one legit developer would sit at his desk and scheme up ways to make his masterpiece API look like best under false pretense .

These guys are smart and know where the issues are when creating these games. Having that relationship and being able to fix the issue both hardware wise and software wise is what mantle is. We've needed multithreaded games since the dual core scene when AMD ruled the world.

I seriously doubt that guy had to test crappy hardware just to make his API look top notch.

 

[Stewox]

Can you point us a link where and when they claim 1000000
draw calls ; i m quite sure that not only you could not provide
any source but still worse , this number doesnt come from AMD
but from here :

http://forums.anandtech.com/showthread.php?t=2366071

Actualy it s you who is making outrageous claims...

Actually it's you who comes here late and thinks he knows better

http://www.redgamingtech.com/amd-ma...100000-draw-calls-similar-to-ps4-api-library/

 

[Abwx]

Actually it's you who comes here late and thinks he knows better

http://www.redgamingtech.com/amd-ma...100000-draw-calls-similar-to-ps4-api-library/

From 100 000 to 1 000 000 there s quite a margin, actualy AMD mentioned 9x the draw call count of current CPUs wich are in the 10k-15k bracket , that s
the number they target for current games , speculations from sites that
try to have an oversight of the evolution is another matter wich has
nothing to do with AMD current expectations.

 

[Stewox]

From 100 000 to 1 000 000 there s quite a margin, actualy AMD mentioned 9x the draw call count of current CPUs wich are in the 10k-15k bracket , that s
the number they target for current games , speculations from sites that
try to have an oversight of the evolution is another matter wich has
nothing to do with AMD current expectations.

Nobody in the core claimed these will be the numbers from current games, any experienced mantle follower never expected that. You have been fed misinformation, like many other people. Mantle has been extremely controversial topic, not because of the no-nvidia concerns, but because many inexperienced and casual people have once again proven them selfs whow good they are at spreading BS and making stuff up.

Speculations from mainstream gaming sites are irrelevant to any worthwhile discussions, those sites are usually poorly technically experienced and should not even be mentioned in any debate. The site what I pasted was a reference to a quote from Oxide games, nothing more.

Plus, AMDs expectations and claims are basically irrelevant, what is relevant is developers. The majority of the work is now done by developers, when the core API is finished, most of AMD's work will be on the SDK, validation, tools, etc. AMD does not "optimize" any games in mantle mode like it does DirectX games, there will be no GPU Driver wars anymore in Mantle, on DX, there is no GPU war, it's all a driver war and it always was, all the benchmarking communites duped from top to bottom, there was never a genuine hardware benchmark. Continuing, AMDs claims are a PR-forward from the developers, or, AMD's claims are from internal non-game tests which don't really represent the the full capabilities of what will be possible to do with mantle. The developers with Mantle can easily optimize, update, overhaul after, so the initial releases of existing and also future Mantle games are extremely variable; because the developers now have much lower level access with Mantle as well as transfer of responsibility, they don't have to rely on the driver and they have the access to do deep optimizations not possible with other APIs such as OGL/DX, which makes patching a lot more easier and a lot more faster, the performance updates will come from the developer of the game like they should suppose to, Mantle updates will only be to the core Mantle API and even that is not yet known whether or not the DLLs are shipped with Games or inside Catalyst. The mantle driver updates on the AMD's side will mostly be basic and with no game-specific code whatsoever, that's why the driver is called "thin".

The only way AMD's claims would be AMDs, if they are internally developing a game themselfs, which they do not.

The point is, why do you hump into AMDs claims at all, who the hell cares, I don't need them, I knew everything I needed to know weeks ago, so that's just my opinion at the end, because I pretty much researched this up and down so basically I don't have much to talk about as everything has already been discussed last year.

The thread on TPU is the one I spent quite some time participating (and I'm the OP) and elsewhere such as here and a few other forums. I have not participated in any discussions since new year as everything was clear to me.

The development of Mantle was directly tied to development of BF4 Mantle renderer, Johan Anderson was deeply involved, the initial tests were made on BF4 as well as other artificial tests from which they got 9x draw calls from, which doesn't have anything to do with frames per second, this is the one of the main goals of mantle, to relieve the CPU dependency, so these numbers aren't much of a surprise and have been expected.


Therefore, because the natrue of what mantle enables, all these "AMD numbers" are extremely loose, they will be obsolete in a matter of weeks when new game updates come, when Thief comes, etc etc.

Take some popcorn and enjoy the show, that's the revolution we were waiting for years, it has come and it can only get better like it supposes to, it's a mathematical certainty.

 

[Imouto]

I think you're crisscrossing things a bit. Mantle does not address CPU performance, it merely reduces the API overhead and makes things more streamlined, thus enabling faster communication with the GPU.

The CPU was never the problem. The problem was the overhead from the API.

Anyway, the CPU's workload is totally different from the GPU. With a GPU, you can nearly double your resources with a die shrink, and get nearly double the performance because the GPU's workload lends itself quite well to parallelism.

You can't do that with a CPU though, as their workload is more serial in nature.

And you're twisting it like there's no tomorrow. If we go by DICE's numbers a 3970X cut the CPU frame times in half using Mantle in the third example.

Then again Mantle is about parallel rendering too as explained several times already. That's why it likes moar cores. Looks like this is why every Mantlephobe was focusing on draw calls and neglecting every other API improvement, if I don't talk about it, it doesn't exist. Everything is in the slides from Mantle presentation so please read and understand it plox.

Adampa1006 is right and I agree and have said the same before. If we see a further +60% perf increase with 20nm GPUs the CPU side of things won't get such an increase with newer uarchs or die shrinks. That said I don't think we will see much higher CPU utilization in games since everything is stuck at DX.

 

[Carfax83]

And you're twisting it like there's no tomorrow. If we go by DICE's numbers a 3970X cut the CPU frame times in half using Mantle in the third example.

Yes, it did, which is because the Mantle driver and API is more efficient at using the CPU than AMD's Direct3D driver . But when you have such a poor starting point as AMD had, then it looks more impressive. AMD's drivers have much higher frame latency than NVidia's it seems.

The screenshot I posted of my stock run with my 3930K @ 3.5ghz in the same area had just slightly higher CPU frame times than AMD's 3970x run with Mantle turned on.

So basically my stock run was right between AMD's Mantle turned off, and Mantle turned on.. At 4.5ghz, it was nearly equal with AMD's Mantle path.

Then again Mantle is about parallel rendering too as explained several times already. That's why it likes moar cores. Looks like this is why every Mantlephobe was focusing on draw calls and neglecting every other API improvement, if I don't talk about it, it doesn't exist. Everything is in the slides from Mantle presentation so please read and understand it plox.

I don't really disagree with anything you say here..

Adampa1006 is right and I agree and have said the same before. If we see a further +60% perf increase with 20nm GPUs the CPU side of things won't get such an increase with newer uarchs or die shrinks. That said I don't think we will see much higher CPU utilization in games since everything is stuck at DX.

That's why parallelization is so important. I was actually kind of shocked when I saw that my performance in those screenshots I posted a few pages back didn't significantly decrease running my CPU at stock, compared to when it was overclocked to 4.5ghz, which implies that BF4 uses the CPU very well already.. I think I was only about 10 FPS off, and considering I was getting triple digit frame rates, that's not a whole lot.

So I guess the Frostbite 3 engine combined with NVidia's drivers and an Intel hex core processor make for a great team

 

[sontin]

This should be the first independent test:
http://www.hardcoreware.net/battlefield-4-mantle-kaveri-performance/

 

[OCGuy]

This should be the first independent test:
http://www.hardcoreware.net/battlefield-4-mantle-kaveri-performance/

Refreshing honesty.

If AMD had just sold it like this in the beginning, I think it would have been better for everyone.