Dolphin, the GameCube and Wii emulator - Forums

I'm not blaming you for your first post, since we decided on what to use (3.0 + Starscream's settings) afterwards. If we're going to compare HLE and LLE, the settings for each need to be as similar as possible. The only reasonable exception should be LLE on Thread since it's pretty useless for HLE audio. To make sure VBeam is taken into account, I've went ahead and retested with Accurate VBeam enabled for both. The speed boosts were pretty meager with LLE audio, and it did allow SMG2 to perform slightly better as well on HLE. I don't think it's that Accurate VBeam emulation doesn't affect HLE, it's just that it tends to affect LLE more so than HLE.

Super Mario Galaxy 2 - Starship - LLE Audio + VBeam
Result: 75 FPS average


Super Mario Galaxy 2 - Starship - HLE Audio + VBeam
Result: 85 FPS average


Percentage Difference: + 12.5%

Accurate VBeam emulation only bumped it up 3 FPS on each. Dunno if anyone can reproduce the HLE + VBeam speed increase, but that's what happened to me. With LLE audio + VBeam, the difference shrinks to about 9% so that the result is borderline significant/insignificant, but the gap returned with VBeam enabled for both.

rpglord Wrote:Shonumi,your test proves that running SMG2 doesnt require 4.4 ghz Sandy Bridge

This is just one area, I haven't even played through most of the levels like the one admin89 suggested. It'd be great if I didn't have to overclock, however

EDIT: Retesting on NSMB gave me similar results. LLE + VBeam closed the gap between HLE audio by itself. But HLE + VBeam returned the gap, and it was even larger than HLE vs LLE (nearly 18% O__o). Gotta go to work, so no time to post pics, will do so when I come home.

Of course when I tested with vbeam,I used it both for HLE and LLE. Otherwise,I would consider these results pointless
I just saw you are using linux..did you use dx9 backend ? Can you even use it on linux ?

Nope, no DX9 or DX11. OGL is the only way I can test. I don't think the backend should be an issue since it's performance is going to vary depending on one's drivers, regardless if it's set to OGL or DX9/11. You'd need everyone on the same OS and same machine if you wanted completely strict results. The gap differences between HLE and LLE audio are what count here. You can test it on OGL if you want to see if the results are interesting or not.

I found many times gpu backends affect things they shouldn't,things which are related to cpu speed !
This is my results with opengl and vbeam enabled for both hle and lle:

OpenGL is definitely much slower for Super Mario Galaxy 2.
The difference between HLE and LLE has shrunken though.
Pretty interesting results.

Super Mario Galaxy 2 HLE + VBEAM


Super Mario Galaxy 2 LLE + VBEAM

And it has increased for me ! Pretty interesting results indeed.

Quote:OpenGL is definitely much slower for Super Mario Galaxy 2.

For you drivers sure, but as I'm sure you can see others have different results.

Quote:The difference between HLE and LLE has shrunken though.

Maybe the GPU became the bottleneck for when you ran HLE? Though with LLE the CPU is still the bottleneck.

(10-12-2012, 07:29 AM)Axxer Wrote: [ -> ]Maybe the GPU became the bottleneck for when you ran HLE? Though with LLE the CPU is still the bottleneck.

We are using starscream's speedsettings. Gpu definitely shouldn't be bottleneck.

The GPU thread might become the bottleneck instead of the audio or CPU threads is what I mean.

All 3 threads have to sync up. Normally the thread with LLE (whether LLE on thread is used or not) becomes the bottleneck, but maybe with OpenGL on that computer the GPU thread becomes the bottleneck.

Quote:I found many times gpu backends affect things they shouldn't,things which are related to cpu speed !

Not all of the video backend functions are shaders.