BTW I have a 7900 xtx. When running the program I can get to 5.5 million particles before I drop down to 30 fps. Hopefully that could be helpful in some way. I'd assume you get more useful info from a lower end system. At 7.5 million I have 15 fps average.
Thanks! Btw renderdoc might just be missing what’s going on in the async compute queue. There is some very inefficient lighting volume stuff going on at 512x512x128. I need to apply some sparse and adaptive concepts to it. I should be able to get it performing much better at your specs and handhelds.
@@dottedboxguy Also, GPU Upload Heaps (aka Resizable BAR) was added to the DX12 Agility SDK but currently it requires a Windows 11 Insider build. I hope it makes its way to Windows 10 too. This particle transfer does benefit a lot from ReBAR.
hmmm it only runs 1 million particles at 10 fps on my computer, when your old liquid heating benchmark test ran at 20 fps, ive noticed that your fps dropped quite a lot since you added the multigrid, have you thought of removing that? or do you think the problem could be something else?
The total particles is technically actually 1M higher because the count doesn’t include the particles belonging to the worms. But also it can still get faster. I didn’t optimize my lighting volume code to become sparse so for some people that is a bottleneck, if you see GPU usage very high and CPU low. The multigrid can be optimized further, it was originally done one way and is being switched to another way and currently has some single threaded parts.
@@GrantKot Nope, it runs a bit and then crashes, it even ran for about half a minute once. The CPU is an i9-13900K and the GPU a RTX 1080 with Windows 11. Let me know if you need more info.
@@mcccxxxvii Hm, it should like that CPU, I do most testing on a 13900HX. I'm going to try to add a crash minidump logging thing or Nvidia Aftermath soon. It's a bit more difficult to figure out if it's crashing after running. I'm also wondering if the CPU is on stock or tuned? This uses all cores including efficiency cores and AVX2 so can generate a lot of heat.
@@GrantKot Yeah the CPU is stock, no OC or anything like that. And yeah, it is basically instantly and constantly at 100°C. Also, I just did another test and this time it didn't crash anymore. So it might be some other program that caused this, I'm not sure. Thanks for your help and the awesome demo!
@@GrantKot works on my i5 8300H and 1050 Ti mobile laptop, I also have 32GB ram, that could be the trick :D I get around 15 FPS after a few minutes of playtime. With the amount of simulated particles it's impressive.
Very impressive. The performance is crazy considering the amount of calculations in play, especially on a handheld.
BTW I have a 7900 xtx. When running the program I can get to 5.5 million particles before I drop down to 30 fps. Hopefully that could be helpful in some way. I'd assume you get more useful info from a lower end system.
At 7.5 million I have 15 fps average.
@@Snoozy96 Thanks, that is helpful. And I am always happy to hear when someone can run my demo 30fps or above :)
that's so cool, this simulation is so nuts, what a project !
Thanks! Btw renderdoc might just be missing what’s going on in the async compute queue. There is some very inefficient lighting volume stuff going on at 512x512x128. I need to apply some sparse and adaptive concepts to it. I should be able to get it performing much better at your specs and handhelds.
@@GrantKotthat's very nice to hear ! i can't wait !!
@@dottedboxguy Also, GPU Upload Heaps (aka Resizable BAR) was added to the DX12 Agility SDK but currently it requires a Windows 11 Insider build. I hope it makes its way to Windows 10 too. This particle transfer does benefit a lot from ReBAR.
@@GrantKotwhat the hell, this makes literally no sense 😔
Really cool stuff! Wanted to chime in and say it runs at above 30fps for me until about 50 million particles :) I am using a RTX 4070
hmmm it only runs 1 million particles at 10 fps on my computer, when your old liquid heating benchmark test ran at 20 fps, ive noticed that your fps dropped quite a lot since you added the multigrid, have you thought of removing that? or do you think the problem could be something else?
The total particles is technically actually 1M higher because the count doesn’t include the particles belonging to the worms. But also it can still get faster. I didn’t optimize my lighting volume code to become sparse so for some people that is a bottleneck, if you see GPU usage very high and CPU low. The multigrid can be optimized further, it was originally done one way and is being switched to another way and currently has some single threaded parts.
Hello, sadly the demo crashes after a few seconds for me. Is there any way to debug this?
Does it crash before anything shows up on screen? What are your system specs?
@@GrantKot Nope, it runs a bit and then crashes, it even ran for about half a minute once. The CPU is an i9-13900K and the GPU a RTX 1080 with Windows 11. Let me know if you need more info.
@@mcccxxxvii Hm, it should like that CPU, I do most testing on a 13900HX. I'm going to try to add a crash minidump logging thing or Nvidia Aftermath soon. It's a bit more difficult to figure out if it's crashing after running. I'm also wondering if the CPU is on stock or tuned? This uses all cores including efficiency cores and AVX2 so can generate a lot of heat.
@@GrantKot Yeah the CPU is stock, no OC or anything like that. And yeah, it is basically instantly and constantly at 100°C. Also, I just did another test and this time it didn't crash anymore. So it might be some other program that caused this, I'm not sure. Thanks for your help and the awesome demo!
@@GrantKot works on my i5 8300H and 1050 Ti mobile laptop, I also have 32GB ram, that could be the trick :D
I get around 15 FPS after a few minutes of playtime. With the amount of simulated particles it's impressive.