AMD Mantle API [updating]

[DavidGraham]

Mantle is a curse, and a grindstone around the neck of PC gaming. By fragmenting the userbase and sucking up development time and money the mantle path unjustly enrichens the experience for a few at the expense of the many. Fuck that.

In the light of the Mantle patch/driver fiasco. I think I need to mention that supporting a proprietary feature in an online game has the potential to be a great source of problems. Online games are prone to constant patching and bug fixing,optimizing.. etc. which results in treating the proprietary feature as a second rate citizen. maybe even neglecting it completely for extended periods of times.

Take for example NVIDIA's PhysX implementation in PlanetSide 2, Hawken and Warframe .. all online games .. the effects have been constantly reworked, had some of them removed, added. the extreme case was PlanetSide 2 which disabled PhysX at least 3 or 4 times to allow for extensive patching and engine optimization, the PhysX effects would stay disabled sometimes for months.. then they would come back slightly changed.

The issue here is that developers clearly lack time and resources to handle two different parts of their game simultaneously. so they prioritize their game first. everything else comes later. As such, I believe single player games could stand to make the greatest benefit of those kind of features. there is usually little to change about them.

The CPU times are the same things FRAPS measures as frame times (and converts to FPS) and sites like techreport.com use in their analysis. This is exactly the right thing to be measuring and great to have it built into the engine now since obviously FRAPS won't work with the Mantle path.

Got it, Thanks.

 

[Tamlin]

Battlefield 4 Multiplayer-Benchmarks
http://www.golem.de/news/amds-mantle-api-im-test-der-prozessor-katalysator-1402-104261-3.html

Star Swarm
http://www.golem.de/news/amds-mantle-api-im-test-der-prozessor-katalysator-1402-104261-4.html

Nice!

Google translated:

Are most marked differences at a resolution of 1280 x 720 pixels: Video card must be here throughout on data from the processor to wait, so the speed is increased by 78 percent. Particularly noteworthy is the minimum frame rate, which is significantly higher among Mantle as under Direct3D. This also means that users of Crossfire and 120-Hz screens, the Radeon-teams have been limited by the CPU, its performance can now take full advantage.

The frametimes also demonstrate that the images are output not only faster but also more evenly. In the same nominal frame rate is thus the feel somewhat improved. Particularly impressive are the values ​​taking account of Windows 8.1 and Direct3D 11.1, since 7 and D3D11.0 benefited Battlefield 4 of both compared to Windows.

 

[Jawed]

I've seen something desirable - Free MSAA

Core i7-3770K, 16 GB DDR3-1333, Radeon R9 290X @ 1.000/2.500 MHz, Windows 8.1 x64, free MSAA:

Siege Of Shanghai (1080p) D3D11: 63.8 avg / 49 min
Siege Of Shanghai (1080p) 4xMSAA mantle: 78.2 avg / 63 min

http://www.golem.de/news/amds-mantle-api-im-test-der-prozessor-katalysator-1402-104261-3.html

Is that a CPU bottleneck being relieved?

 

[fellix]

Is that a CPU bottleneck being relieved?

Indeed. I would rather see benchmarks with some older generation CPUs, like Phenom II. It would be amusing if such config edges, say an i7-3770 system.

 

[Kaotik]

Mantle is a curse, and a grindstone around the neck of PC gaming. By fragmenting the userbase and sucking up development time and money the mantle path unjustly enrichens the experience for a few at the expense of the many. Fuck that.

Same goes for nvidia physx, and creative's proprietary sound in the past. Difference is, physx only ever made sense for games that leveraged physics to any greater extent, while mantle has the potential for much broader reach.

PC gaming has prospered under unified 3D APIs. I don't want to see that ending, it can only lead to bad in the end.

Good thing that AMD thinks like this, then:

Q: Is Mantle a proprietary AMD technology?
A: Mantle was conceived and developed by AMD in partnership with leading game developers.
This enabled the fast and agile development required to validate the concepts and bring such the technology to life in a relatively short period of time. However, Mantle was designed in a way that makes it applicable to a range of modern GPU architectures. In the months ahead, we will be inviting more partners to participate in the development program, leading up to a public release of the specifications later in 2014. Our intention is for Mantle, or something that looks very much like it, to eventually become an industry standard applicable to multiple graphics architectures and platforms.

 

[Lightman]

Any of the review site owners willing to release this press driver to our small but willing to test community?

I could run Mantle on Athlon 64 X2 5000+ and R9 290X just for the sheer laugh

 

[Kaotik]

Any of the review site owners willing to release this press driver to our small but willing to test community?

I could run Mantle on Athlon 64 X2 5000+ and R9 290X just for the sheer laugh

The public release should happen within 24h anyway

 

[Svensk Viking]

In the case of the Star Swarm demo, it can get better performance in DX11 by turning on "Deferred contexts" manually. It says the very latest drivers are needed for an improvement though.

http://www.oxidegames.com/2014/01/31/star-swarm-faq/

 

[Tamlin]

Core i7-3770K, 16 GB DDR3-1333, Radeon R9 290X @ 1.000/2.500 MHz, Windows 8.1 x64, free MSAA:

Siege Of Shanghai (1080p) D3D11: 63.8 avg / 49 min
Siege Of Shanghai (1080p) 4xMSAA mantle: 78.2 avg / 63 min

http://www.golem.de/news/amds-mantle-api-im-test-der-prozessor-katalysator-1402-104261-3.html

Is that a CPU bottleneck being relieved?

Seems that way. Look at the minimum framerates:

 

[snc]

http://pclab.pl/art55953-3.html very good test

 

[repi]

http://pclab.pl/art55953-3.html very good test

Not really, they didn't specify how they tested multiplayer and most importantly: if the multiplayer was full 64 player server or empty. Affects CPU load massively.
Update: looks like they did specify that it was 64-player MP server which is good assuming it was full the entire time for all tests and they played for a quite long time as the workload is quite variable

 

[xDxD]

Core i7-3770K, 16 GB DDR3-1333, Radeon R9 290X @ 1.000/2.500 MHz, Windows 8.1 x64, free MSAA:

Siege Of Shanghai (1080p) D3D11: 63.8 avg / 49 min
Siege Of Shanghai (1080p) 4xMSAA mantle: 78.2 avg / 63 min

http://www.golem.de/news/amds-mantle-api-im-test-der-prozessor-katalysator-1402-104261-3.html

Is that a CPU bottleneck being relieved?

of course...it seems clear that most benefit of mantle is on the cpu side

 

[willardjuice]

Repi do you make use of the msaa compressed data with mantle? I was curious to see what could be achieved in that area.

 

[heli]

http://pclab.pl/art55953-3.html very good test

I found it's hard to believe that with i7-4770K and i5-4670K in DX, the differences between GTX 780 and R9 290X are that great. I know it's multiplayer so things are hard to control, but something isn't right, the condition must differ greatly.

 

[repi]

Repi do you make use of the msaa compressed data with mantle? I was curious to see what could be achieved in that area.

Yes we do, but the DX drivers have optimization that does it as well at least for some cases. But with Mantle we have explicit control and can go further. Don't think I can go into details about it though as it is very AMD specific and not sure what is public.

 

[gongo]

The perf gains for a regular Intel gaming PC (sorry AMD folks) are impressive..that i'm re-impressed and re-excited...so we know AMD just played the quiet game with the APU/FX charts...nothing is better than an under-hyped to over-delivered feelings of joy...lolz!!

Btw any of you remembered just days ago, AMD was sort of pushing the blame to Dice for mantle delay, yet BF4 mantle patch is up and downloaded, while the mantle catalyst remains missing...AlwaysMakingDelays

 

[snc]

Not really, they didn't specify how they tested multiplayer and most importantly: if the multiplayer was full 64 player server or empty. Affects CPU load massively.
Update: looks like they did specify that it was 64-player MP server which is good assuming it was full the entire time for all tests and they played for a quite long time as the workload is quite variable

Ok, but it's obvious its cpu demanding scene otherwise different between mantle vs dx isn't more than 10%(see single player results)

 

[mc6809e]

http://pclab.pl/art55953-3.html very good test

The most impressive thing about this test is the Pentium G3220 result.

33.1 FPS to 50.4 FPS is an amazing leap for a system with a cheap processor.

 

[Lightman]

Ok, but it's obvious its cpu demanding scene otherwise different between mantle vs dx isn't more than 10%(see single player results)

And what do you expect?
Do you think DX11 is so bad that more than few % are possible in a well optimized scene where game dev. worked around API limitations?

I don't and I'm surprised that 10% was possible at all with fast CPU's where 1 core is approaching speeds of whole next get console 6 cores.

 

[Dave Glue]

I mentioned earlier that a good litmus test for Mantle would be to see how it performs on lower-end Intel CPU's, to see if its optimizations are largely targetted at low IPC CPU's but have many cores.

Welp:

That answers that. 70-80% boost - and running at quality settings you would expect with a high-end GPU.

"Mantle: Turn your dual into a Quad" <- there's your marketing tagline.

I've seen several comments in threads on this on other sites, including Patrick Norton had this critique on the Twit Hardware podcast with PCper; basically all are along the lines of "Well of course this is silly as a lot of these tests pair a low-end CPU with a Radeon 290x...".

The ridiculous doge/litecoin fueled-prices of the 290 aside with make this pairing more ridiculous than it seems right now, why is pairing a high-end GPU with a Kaveri/Trinity, or even a i3 "silly"? Well, it is now - because we expect that we need a quad-core Ivy/Haswell CPU to drive these GPU's.

There are certainly cases where Mantle will not be able to help and of course Ivy/Haswell CPU's have more advantages outside of games with their increased performance, but what if we "need" high-end Intel CPU's to run modern games at ~60fps because of DX11's overhead? This is the exact problem Mantle is trying to solve, so using that "unbalanced" critique seems to largely miss the point.