I initially wanted to post all benchmarks simultaneously, but I haven't had time to run them all yet, so I'll just add to this post as I get results. I am testing a set of scenarios that are hard to emulate (well, relatively speaking), with a version just before the VideoCommon rewrite, and a very recent one after most of the changes have been merged.
The purpose is simply for fun and to share information, in case it helps in any way to identify potential .
Your results are very different than mine, which makes me wonder what's going on. I know there's a bug with recording performance with Immediately Present XFB on in older builds, and then I know there's a bug with performance with Immediately Present XFB copies on.
I'm guessing it's likely a difference in enhancements. When I get performance results, I try to use good settings that I know work in the games... while a lot of users love shit like MSAA/AF even though it tends to break little things. It's important to know this stuff, but, if you wanna take it to the next step (and perhaps get stenzek to optimize the cases you ran into) it'd be worth seeing if you could find out what is slower.
For what it's worth, in general, maximum GPU performance will be slightly lower after the rewrite due to some restructuring, but it shouldn't actually matter outside of "way too high MSAA" type settings.
This is one of the things that makes performance testing so hard: there are so many settings that can be affected, and now I think you found one that got slower. It's just which one.
Thanks. There are a few other cases that I'm going to test but if you can think of anything I can do to turn this exercise into something useful for stenzek and the dev team, I'll be happy to run it. I've been doing this with the settings that I personally use, which are geared towards getting the best picture possible in my 4K TV, but to the extent that I can keep full emulation speed.
By the way, I just tried Skyward Sword without Immediately Present XFB, using DX11 and the result was exactly the same (24 FPS).
Immediately Present XFB (off) is bugged mostly in Vulkan as far as I know. It's a weird bug that hasn't really been looked into much. Check out Twilight Princess Faron woods and watch the performance act crazy.
If you can isolate what enhancement is slower, it can probably be optimized or at least noted so that someone else can look into it in the future.
Does anyone know why april blog article says vulkan does not work with intel HD gpus??? I have win 10 and it has regular driver updates (not as old win systems) and vulkan looks ok for other applications. Is that mean that intel driver have bugs or not fully implemented all vulkan features?
Forgot to say, dolphin works on my intel HD620 and vulkan backend. No crashes.
I couldn't get it to run on my IntelHD 630. I tried updating drivers. Stenzek couldn't get it to run on his haswell either. If you have an exact driver I can install and it'll work, then I'll reconsider/edit the article.
Hi. I cannot think of any special driver version. I have latest driver (it has now all opengl 4.6 extensions!, only Shader language missing) from intel site. Always install manually not from win update. I have laptop with nvidia discrete graphics. Nvidia installer might install all needed vulkan files in system folders what intel might not. So to prevent a situation with missing files (or even missing registry entries), you might install official vulkan runtime (vulkan RT) from this site
https://vulkan.lunarg.com/sdk/home#windows I cannot think of what kind of else problem you may encounter.
(04-07-2019, 03:00 PM)sirdaniel Wrote: [ -> ]I have laptop with nvidia discrete graphics.
Vulkan is running on your nvidia GPU.
Oh c'mon im not so dumb. I choose and use intel gpu processor specifically. Im doing this all the time when testing other emulators.
I updated the attachment in the first post to show a few more tests.
In general:
- OGL took a huge haircut in some scenarios
- Vulkan and D3D gained a little in some cases and lost a little in others
- 6xIR is a huge performance hog vs. 4xIR in the couple of cases I ran both
I am not surprised by the fact that 6xIR is such a taxing feature but I was surprised by the magnitude of the difference vs. 4xIR. It should be noted that I am using that in conjunction with 4xMSAA.