Massive FPS drop when viewing different A/C or F10 map

[warpz]

Hi,

 

I've experienced massive FPS drop on NTTR when switching  to view another planes via F2 external view or when switching to F10 map view. It goes like this. Start of the mission, in cockpit, around 60 FPS -> switch view -> drops to around 15-20 PFS, for 20-30 secs, then normalizes at around 60 FPS -> switch back to cockpit -> same FPS drop. It's Nevada specific, no problem on similar on PG.

 

Track here - https://www88.zippyshare.com/v/tqveRoSo/file.html - too big to attach.

[Furiz]

It happens to me when I click on some other aircraft on f10 map, in multiplayer, that it freezes for a sec

[Dangerzone]

I believe this is widely known about and general community consensus from what I've read is that it's a VRAM issue with poor optimisation or freeing of the VRAM memory.

 

I haven't heard any word from ED regarding this though - but I'm kinda hoping that they're aware of this but have decided to put the effort into the vulkan upgrade and are dealing with it there as opposed to putting resources into fixing code that will be obsolete shortly anyway (over them not being aware or ignoring it). 

[Worrazen]

Freeing of RAM is what I noticed is an issue, I ran an encyclopedia load test again and saw what are indications of same data being re-read from disk (and pagefile) back into RAM.

 

When you traverse through the Encyclopedia, you load up all the assets/materials/textures of that unit, all of the units it's around 25GB Commit I think ... When getting close to 32GB Commit an 8GB pagefile has around 10-20% usage, then when you end up close to hitting RAM limit it's around 35% pagefile usage, but in order to get that high I had to view and browse through same units I had already been earlier, they loaded in VRAM/RAM and they shouldn't be reloading anything from disk, all it should do is a VRAM-RAM swap when getting back to the oldest data, RAM copies keep building up, and pagefile's being read more and more, which keeps rising RAM and pagefile usage. I'll do a proper thread on this once I analyze that again and figure more things out if my theory holds true.

 

The high RAM usage of DCS never made sense to me, especially the buildup of it, this may have been a bug all the time, garbage collection is a huge and important deal. The game may think it dumped the older assets, but it may be only the reference that was removed from a list, while the stale memory wasn't actually released and freed.

 

VRAM being kept full is actually on purpose and a good thing, so there's as much texture data there where it has to be to prevent VRAM-RAM swapping or waiting for things to get into VRAM, unfortunately VRAM out there still being too low for optimal DCS's needs, but it's thankfully getting better with Radeon RX6000 series indeed. Great memory management can go a long way ofcourse and optimization in this regard is always good, ofcourse not minimizing usage too strictly, because that creates other side effects which may not be desirable.

 

As far as FPS and sutterning goes, the pure usage part, the RAM-to-VRAM transfer may not be the biggest issue, I think it's the CPU usage of the Disk I/O threads clashing and competing for resources with the main thread and/or my latest hunch is that their own (Disk I/O threads) are responsible for some activities that the main engine execution needs and thus can't proceed if they don't do a particular job at the required time.