Nvidia about DX11/12 CPU efficiency/No Mantle support/ Fermi&Kepler/Maxwell will have DX12 & Debunking some Mantle with DX12 assiocated myths

[Faa]

Why we need better CPU performance or ways to reduce it eg a lowlevel API like Mantle;

With the release of AMD Mantle, the idea of a “low level” API has been placed in the minds of gamers and developers.  The term “low level” can mean many things, but in general it is associated with an API that is more direct, has a thinner set of abstraction layers, and uses less translation from code to hardware.  The goal is to reduce the amount of overhead (performance hit) that APIs naturally impair for these translations.  With additional performance available, the CPU cycles can be used by the program (game) or be slept to improve battery life. In certain cases, GPU throughput can increase where the API overhead is impeding the video card's progress.



On the left we see performance scaling in terms of GFLOPS and on the right the metric is memory bandwidth. Clearly the performance of NVIDIA's graphics chips has far outpaced (as have AMD’s) what the best Intel desktop processor have been able and that gap means that the industry needs to innovate to find ways to close it.

Why no Mantle?







What NVIDIA did want to focus on with us was the significant improvements that have been made on the efficiency and performance of DirectX 11.  When NVIDIA is questioned as to why they didn’t create their Mantle-like API if Microsoft was dragging its feet, they point to the vast improvements possible and made with existing APIs like DX11 and OpenGL. The idea is that rather than spend resources on creating a completely new API that needs to be integrated in a totally unique engine port (see Frostbite, CryEngine, etc.) NVIDIA has instead improved the performance, scaling, and predictability of DirectX 11.

NVIDIA doesn’t deny that a lower level API is beneficial or needed for PC gaming. It does, however, think that the methodology of AMD’s Mantle is the wrong way to go.  Fragmenting the market into additional segments with a proprietary API does not maintain the benefits of hardware abstractions or “cross vendor support”. I realize that many readers will see some irony in this statement considering many in the industry would point to CUDA, PhysX, 3D Vision and others as NVIDIA’s own proprietary feature sets. 

Benchmarks from sources







TL;DR Mantle barely reduces more CPU bottlenecking than Nvidia's Directx - it's only doing great against their own Directx and refusing to fix this to showoff how great Mantle is. Across the first Mantle driver all the way up to the new Omega driver, there's still the same amount of overhead in their directx drivers.

Debunking DX12 myths 

I've seen people here posting that Microsoft took Mantle's main feature over. Stuff like that Mantle was the first out there that was after reducing CPU overhead or that Mantle pushed Microsoft. Or that AMD helped Microsoft for DX12. Far from true and they're only GPU vendor atm that doesn't have a DX12 driver at all, read here why;
 

NVIDIA claims that work on DX12 with Microsoft “began more than four years ago with discussions about reducing resource overhead. For the past year, NVIDIA has been working closely with the DirectX team to deliver a working design and implementation of DX12 at GDC.” This would indicate that while general ideas about what would be in the next version of DX, the specific timeline to build and prepare it started last spring.

NVIDIA is currently the only GPU vendor to have a DX12 capable driver in the hands of developers and the demo that Microsoft showed at GDC was running on a GeForce GTX TITAN BLACK card. (UPDATE: I was told that actually Intel has a DX12 capable driver available as well leaving AMD as the only major vendor without.)

Source is here; http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Talks-DX12-DX11-Efficiency-Improvements
 

[CoolaxGaming]

Sigh, NVIDIA says something. AMD says something

Both are lies, after seeing GPU's for soo long, I don't believe it until I see it

[Sidiox]

Who the hell would think that AMD made MS create DX12? They have had DX12 on the roadmap for years ofc, they might have just announced it earlier than planned because of Mantle and allocated some more money to its development, but they sure as hell didn't take the idea from AMD.

 

And I agree with the point that more proprietary stuff isn't good, and as for DX11 efficiency on Nvidia's part and Mantle on AMD's part I can't comment, because I simply don't know. 

[Enderman]

Mantle is used in so few games it almost doesnt matter.

And it only makes a significant difference when the CPU is a bottleneck, which an i5 and up is not.

[Faa]

Mantle is used in so few games it almost doesnt matter.

And it only makes a significant difference when the CPU is a bottleneck, which an i5 and up is not.

If you haven't checked the benchmarks I included, Mantle is only doing good against their OWN Directx and simply falling behind Nvidia's Directx or doing slightly better. Star Swarm is a benchmark that was specially written to advertise Mantle.

 

 

Sigh, NVIDIA says something. AMD says something

Both are lies, after seeing GPU's for soo long, I don't believe it until I see it

Which is why I included benchmarks who were done by PCPER/PClab.pl/computerbase.de testing who's removing the most CPU bottlenecks and we clearly see AMD's DirectX being nowhere in the league of Nvidia's one..

[Misanthrope]

NVIDIA doesn’t deny that a lower level API is beneficial or needed for PC gaming. It does, however, think that the methodology of AMD’s Mantle is the wrong way to go.  Fragmenting the market into additional segments with a proprietary API does not maintain the benefits of hardware abstractions or “cross vendor support”.

 

Yeah, like  Gameworks or G-Sync don't do exactly that right?

[Faa]

 Yeah, like  Gameworks or G-Sync don't do exactly that right?

Theyre bringing a public SDK out for Mantle so anyone can use it and since then the question for Nvidia has been if they will use it or not and this article answers it with "no". Reasons why check the Star Swarm benchmark which has Mantle included who's main purpose/performance boost was removing CPU bottlenecks.

[mr moose]

 

Yeah, like  Gameworks or G-Sync don't do exactly that right?

he already pointed out the irony.

 

 

Who the hell would think that AMD made MS create DX12? They have had DX12 on the roadmap for years ofc, they might have just announced it earlier than planned because of Mantle and allocated some more money to its development, but they sure as hell didn't take the idea from AMD.

 

And I agree with the point that more proprietary stuff isn't good, and as for DX11 efficiency on Nvidia's part and Mantle on AMD's part I can't comment, because I simply don't know. 

 

There are lots of people who think dx12 is mantle with a different name.  Whole threads have argued it.   It's kinda old and annoying that people don't think critically anymore.

[patrickjp93]

Nvidia and hypocrisy go hand in hand. That goes without saying. AMD might have pushed Microsoft to earlier release, but that's it. AMD's last remaining software bastion is HSA at this point. It's amazing how fleeting the company's work has been of late.

[Misanthrope]

Theyre bringing a public SDK out for Mantle so anyone can use it and since then the question for Nvidia has been if they will use it or not and this article answers it with "no". Reasons why check the Star Swarm benchmark which has Mantle included who's main purpose/performance boost was removing CPU bottlenecks.

 

Oh right, the reason they will not use a public SDK that is free for everybody is because it's a proprietary API.....the very thing we just fucking said it is not.....yeah....This is Ubi levels of incompetence here.

[Faa]

Oh right, the reason they will not use a public SDK that is free for everybody is because it's a proprietary API.....the very thing we just fucking said it is not.....yeah....This is Ubi levels of incompetence here.

I didn't say Mantle is a prop. API.

[Misanthrope]

I didn't say Mantle is a prop. API.

 

Right and to clarify my comments are not directed specifically at you but more at Nvidia of course.

[patrickjp93]

I take issue with those graphs on the grounds they don't reflect the Xeon Phi cards which are right now being converted into a socket CPU called Knight's Landing. We all knew serial compute had its limits, but it's still the backbone of heterogeneous systems today.

[worldwin]

I do not like benchmarks by either vendors. There will also be bias hence why we have individual sources like guru3d. The problem with Starswarm is the inconsistency of the drawcalls. For all we know in the Nvidia slide the drawcall can be  at 10k and AMD's @ 70k. We do not know due to how inconsistent the benchmark is.

[Faa]

I do not like benchmarks by either vendors. There will also be bias hence why we have individual sources like guru3d. The problem with Starswarm is the inconsistency of the drawcalls. For all we know in the Nvidia slide the drawcall can be  at 10k and AMD's @ 70k. We do not know due to how inconsistent the benchmark is.

Mantle was supposed to have 10x the drawcalls of DirectX. http://www.legitreviews.com/wp-content/uploads/2013/09/amd-mantle-645x384.jpg

[worldwin]

Mantle was supposed to have 10x the drawcalls of DirectX. http://www.legitreviews.com/wp-content/uploads/2013/09/amd-mantle-645x384.jpg

What does that have to do with what I wrote?

[qwertywarrior]

Date: March 22, 2014

ultra old news dude

[patrickjp93]

Mantle was supposed to have 10x the drawcalls of DirectX. http://www.legitreviews.com/wp-content/uploads/2013/09/amd-mantle-645x384.jpg

That is not what that is saying. Mantle was supposed to make it so one could more directly control what happens in a draw system. for the PS4/XBOne that meant drawing layer by layer instead of sending a stack to the GPU and hoping the scheduler made it come out plenty optimized. That's also the current problem with OpenGL. Having 50x more draw calls was how the developers chose to use the system. Instead of sending an optimal stack, they sent the layers one by one. The port of such a system to DX 11 had disastrous consequences.

[patrickjp93]

Date: March 22, 2014

ultra old news dude

Find it in one of Linus' videos or on LTT or GTFO.

[qwertywarrior]

Find it in one of Linus' videos or on LTT or GTFO.

ok ill use that logic and post a news article from 1999 then

[patrickjp93]

ok ill use that logic and post a news article from 1999 then

It's news if no one broke it and the community was unaware. Therefore, old article or not, it's news. Apparently for a month AMD has had OpenCL 2.0 FirePro graphics cards, but no one noticed until I posted about it today.

[StingBull]

he already pointed out the irony.

And said nothing in defense.

[mr moose]

And said nothing in defense.

 

Why would he need to?  It's not like it means anything other than we all know it runs counter to Nvidias MO.   I would have thought simply acknowledging it was enough,  no need to keep repeating the obvious so to speak.

[StingBull]

Why would he need to? It's not like it means anything other than we all know it runs counter to Nvidias MO. I would have thought simply acknowledging it was enough, no need to keep repeating the obvious so to speak.

They chastise AMD for segregating the market with proprietary technology while admitting they are doing exactly the same thing. They have to do more than acknowledgement there or it's like saying AMD murders innocent babies on tuesdays and should be hanged for it but we only do it on every other day of the week so we are cool.

[mr moose]

They chastise AMD for segregating the market with proprietary technology while admitting they are doing exactly the same thing. They have to do more than acknowledgement there or it's like saying AMD murders innocent babies on tuesdays and should be hanged for it but we only do it on every other day of the week so we are cool.

 

I was talking about the OP, because the OP pointed out that it was ironic of Nvidia to say that, I don't think the OP needs to defend anything, he just links us to the news and we judge on that. 

 

As far as Nvidia goes, well they like all companies should support their claims with evidence and rational arguments.  Unfortunately they are not alone in that they just saying what makes them look good and leave out the inconvenient.