Post by thePalindrome in Update 22 - Rendering at scale & Options comments

Update 22 - Rendering at scale & Options » Comments

Viewing post in Update 22 - Rendering at scale & Options comments

Heh, my itch client just noticed it while I was updating some of my other games!

From what I can assume (I don't work with C# myself), you could examine the exception object to grab the hresult out of it, although a general catch would work fine enough

The sounds do seem to be okay, so this build works great ^v^
I'd be a *bit* surprised if it's mp3, since I swear I have other things that work fine... Then again, there are patents and licenses and junk that cause problems with these things!

Thanks for looking into the issue, I'm just glad we're getting closer to being able to make it "just work" through wine!

Elorth1 year ago

Great I'll push a new version 2.71 then!

Could I ask you if you could run the benchmark with the Raymarching rendering enabled and give your result ?

thePalindrome1 year ago(+1)

Well, the first two are much smoother and better on my system with raymarching than with rasterization!

Sadly after the second one it hits an OOM exception during the AssessAndMoveToNextBenchmark function. staring at "free", I'm not seeing it use almost any memory, which would seem to suggest it either tries to allocate a massive amount in one sitting, or it's leaking so fast the 0.5 second refresh rate I have set isn't detecting it!

Either way, I'm going to have some friends give it a try too, see if it's just my system acting up

Elorth1 year ago

I wouldn't have thought that you would get an OOM. This last test is quite large but it should have fit into ram without OOM (takes 1.5Gb of memory). Maybe the .net runtime set a lower threshold of memory pressure for your config, or a memory fragmentation issue (I have a 32GB ram here).

Thank you to have taken the time to do the test!

thePalindrome1 year ago (1 edit) (+1)

I have 32 GB of memory as well and even with 17 of it free, it fails to allocate. Makes me wonder if some calculation is just going haywire and overallocating as a singular massive blob! Sorry I can't help triage it further, I'm just not sure what tools I have to figure out exactly what allocation is failing!

EDIT: ran it from a fresh boot, with a full 29.8 GB of memory having never been touched, and it still hit OOM. So something's asking for an absurd amount of memory up front!

Elorth1 year ago

For sure, that removes the memory pressure or fragmentation hypothesis. I will wait for others' results to have more insight on this. I have too many clues about where it could come from at the moment. Everything I could do now would be a shot in the dark.

Thank you again, you are very helpful man!

thePalindrome1 year ago(+1)

Okay it *might* be a "me" issue or something, since a friend of mine got it to work just fine (although it used 5 GB, which makes me think something is being 32-bit in the code somewhere on my end)

He's running an RX 570 with the mesa drivers

On the old renderer, he got 1323/52.9/15.2, 128/5.1/0.5, and 0/NaN/? (It wouldn't respond to click or anything, but closed immediately when he clicked the x of the window)

On the new renderer, he got 1308/52.3/22.7, 996/39.8/21.7, 1051/42.0/20.4, so it seems this new render helps a lot!

Elorth1 year ago

If you are able to formulate a hypothesis I could do a special build for, don't hesitate. Also I'm quite intrigued by the fact that you have such a large memory consumption though. Is it possible that your number includes Wine or some other layers? (and if this is the case, 32bit and 5GB won't work as you mentioned)

About the results. That's some solid numbers on a representative of the market GPU tier! So close to that sweet, sweet 60fps. I'm sure I could squeeze some more fps out of that renderer in the future.

This is tremendously valuable data for me. Once again, thank you sir!

thePalindrome1 year ago(+1)

The best I could suggest is more debug symbols or something, I'll work with my friend to figure out the deep debugging (gdb-type stuff) as well, since if we could even figure out the offending allocation, that would probably give us some insight!

The large memory allocation could be in part due to wine, but I'd be surprised if it were over 500 MB, and even that number is probably too high for what wine's injecting (A lot of it is handled by "wineserver", which is a seperate process and therefore not in this number)

~~for some informal numbers, I'm getting 9 or 10 fps in the first benchmark when things are mostly in view, and 5-9 fps in the second (measured using mangohud).~~

I managed to get the game running with wine-ge 8.13, which yielded the following results:
508/20.3/7.6

331/13.2/4.7

286/11.4/2.4

I would like to state my graphics system is a bit of a mess from switching drivers a few times and not really being bothered to do it properly, so a bit of a caveat to that!

Elorth1 year ago

Ok quite interesting, let me know you find something related to that allocation! I think I should install a Linux somewhere as a test machine, but I'd rather focus on the dev of the game itself. It has rough edges as of today and there is a lot to do!

You benchmark scores are underwhelming, what is your graphic card?

View more in thread

itch.io

Viewing post in Update 22 - Rendering at scale & Options comments