AMD Mantle Interview with Oxide Games' Dan Baker

[Ian Evey]

Blah Blah, this still runs at 40% to 50% gpu usage along with cpu usage just as low, what really is the point in anyone buying a game using this gimped game engine.

I feel oxide have only hurt future sales, of course if the gpu and cpu usage was a lot higher then i could understand the increase in mantle performance, but 50%, come on.

 

[gpuerrilla]

*Cough - yeah good article.

 

[Steampunk]

Blah Blah, this still runs at 40% to 50% gpu usage along with cpu usage just as low, what really is the point in anyone buying a game using this gimped game engine.

I feel oxide have only hurt future sales, of course if the gpu and cpu usage was a lot higher then i could understand the increase in mantle performance, but 50%, come on.

So you're happy as long as your CPU is spinning it's wheels at 100 percent even though it's not achieving anything because it's stuck dealing with DX overhead instead of running the game?

You're mistaking activity for accomplishment.

 

[Ian Evey]

Erm. Gpu usage at 40%. How?

Would you honestly buy a game that ran at 30 fps and only used 40% of your CPU and gpu. I know I wouldn't.

 

[triss]

Nice article greg ty

 

[andybird123]

So you're happy as long as your CPU is spinning it's wheels at 100 percent even though it's not achieving anything because it's stuck dealing with DX overhead instead of running the game?

You're mistaking activity for accomplishment.

riddle me this... I've logged CPU usage on all cores running this demo, alongside GPU usage, using both deferred contexts on and off

the demo has no issue threading itself, I see activity on all cores, however NO core hits 100%, max 60% on a single core, so there is no CPU bottleneck... similarly my GPU NEVER hits 100%, max I saw was 70% and not at full boost either

please tell me where the bottleneck is? it isn't CPU and it isn't GPU

"it is DX" is not an answer, DX becomes bottlenecked when it can't thread itself and the primary core hits 100%, bottlenecking the GPU and every other thread it is using... in the case of star swarm this isn't happening, at all

if I enable deferred contexts I see better, more even usage across the CPU cores, but no higher GPU usage, but I do get a 50%+ increase in FPS

with every other case I've seen of dropping FPS, one or other of at least one CPU core, or GPU usage, hits 100% and then FPS drops, but in the case of starswarm I see no bottleneck, just an artificial drop in FPS for no apparent reason

 

[Calin Banc]

The main problem with our average Joe user's way of seeing things, is that we're not that good at it.
1. we lack the required tools of seeing deep intro the code;
2. we lack the required knowledge to read it and interpret it adequately.

GPU Z, MSI Afterburner or other monitoring apps, may show 99% usage on the GPU, but it's quite clear that the video card isn't working that hard by the amount of heat it puts out or just by knowing that it's running an old or not that demanding code: dx 9 or 8 vs, newer versions or even on the same - DX11.
At the same time, the card may indeed work full speed, with everything on, BUT can work inefficiently.

Just checked the situation where I've got the most gains in FPS in BF 4 (roughly 50%) and although the GPU stays at 99% and the difference in CPU usage is under 4% (between Mantle and DX11.1), performance wise, the gap is HUGE. Another fact, vRAM without Mantle was around 1,6GB, while with Mantle, around 2,8GB. 1,8 vs. 2,1 GB RAM usage (w/o vs with mantle).

Coming back to Star Swarm, using Follow, Extreme and Timed Run settings, Mantle gives 99% usage on GPU, about 2,5GB vRAM used, while dx goes between 50-70% GPU usage and about 1,5GB vRAM. Judging by the fact they are really good with d3d, I'd say it speaks for itself.

An API like this is required and we ALL benefit from it. I pretty much would like to keep and use my hardware as close as possible to it's full potential. Name it Mantle, DX12, OpenGL 5 or whatever, doesn't really matter. Even AMD said that. But a change is required and not for the sake of changing things.

 

[Hagar]

Baker: DirectX was architected in a time when two things were true. First, the hardware itself was very fixed-function. That is, there was a lot of secret sauce

Yum, what is the source for that?

 

[Silent_Scone]

Well said Calin. We need an API that isn't controlled by a competing vendor and one that is controlled as equally as OpenGL is via the Krhonos group for it to be a viable longterm solution mind you. As nice as all these new trinkets are.