What's the explanation for this Xenosaga performance?

[Secret Dragoon]

Still not seeing any real speedups in 0.9.8. I get lag in two specific occasions.

1. In battle with any AGWS units out. (This is probably due to Large 3D Models)
2. In Cutscenes where there are a lot of Models.

Weird example of when I am lagging:


Notice that it's the GS that's being stressed here. The issue however is that both my cores are at 90%+ usage when most of my lag happens however my Graphics Card is only around 30%. Shouldn't the GS be using my Graphics card? (Specs in Sig)

Also, GSdx r4000 SSSE3, DX10 Hardware Mode, Speedhacks but both sliders are at default levels (Those don't really make a difference here anyways)

Note 90% of the time, I play at 60FPS. These are the only two examples of when I lag in this game. Also Save Point Glitch only occurs at save points before you reach the Elsa. Only save point that glitches that you'd regularly use is the Elsa's EVS one (next to item shop). So far I haven't seen many Save points that boot to the Disk Launcher.

[nosisab Ken Keleh]

It's not too strange, the video card is at the mainstrean- side and being indeed stressed to the limits there. That is a game which shines at stressing either the EE or GS depending on what is happening in game.

[orchlon]

This happened to me once GS being stressed like hell. Worked fine tomorrow for some reason

[rama]

The problem is that the game creates hundreds of thousands of draw calls which have to pass the D3D stack.
It won't stress the GPU much more than usual but the CPU has to process all this.

[Secret Dragoon]

So I guess it's a Limitation of DirectX and as a Result bottlenecks the CPU?
This is a great example of a bottleneck then if I'm reading that right.

[Squall Leonhart]

You could try increasing the skipdraw hack, which reduces the draw call by a tiny amount.

This is a totally cpu bound situation, not so much D3D, even GSDX software doesn't provide much improvement.

[rama]

D3D runs on the CPU, Squall.

[Squall Leonhart]

well yeah

if only there was a way to scale draw calls across multiple cores without using gsdx software.

[rama]

Well, DX10 already supports multithreaded setup and DX11 improves on that.
For some reason enabling this is slower in high overhead games, never found out why.
From the source:

Code:

// NOTE : D3D11_CREATE_DEVICE_SINGLETHREADED
//   This flag is safe as long as the DXGI's internal message pump is disabled or is on the
//   same thread as the GS window (which the emulator makes sure of, if it utilizes a
//   multithreaded GS).  Setting the flag is a nice and easy 5% speedup on GS-intensive scenes.

uint32 flags = D3D11_CREATE_DEVICE_SINGLETHREADED;

Edit: Ah, forgot that this flag is actually a performance optimization since it disables thread safety.
The actual flag to *disable* threading optimizations is "D3D11_CREATE_DEVICE_PREVENT_INTERNAL_THREADING_OPTIMIZATIONS".

So yea, we're already using the built in multi threading.

[Squall Leonhart]

you guys will add things as you figure them out so its by no means the end just yet. Someone may find a way to get better performance from multithreaded draw calls in the future. Or cpu's will just get to the point where we don't need them.

Hmm, Rama is there a chance that overhead issue might be due to cards previously lacking support for Driver Command Lists?