Better multi-core CPU Optimization for 2.5 maybe ?

[StandingCow]

While it has been said DCS won't support multithread... what about the upcoming dedicated server support? That would have to support it I would think.

[shagrat]

While it has been said DCS won't support multithread... what about the upcoming dedicated server support? That would have to support it I would think.

Why? It still needs to calculate all the above in a synchronized manner.

It just offloads the ASM, EFM, graphics and rendering to the client, as it doesn't need to render any graphics nor handle the complex flight model.

But the rest definitely should be kept in one synchronized thread. You don't want a desynchronization between threads to find your target at a different place for calculations, than the spot your bullets/rockets aimed at...?

 

Complex and detailed aircraft modelling happens on the client.

As well does the rendering for the world and screen.

Currently the host in addition computes the whole mission environment and AI forces, like in single player, and the clients get synced to the host through SFM like vectors and info.

 

Now with a dedicated server each client still needs to render the World, and sync the other clients plus the environment.

 

Yet, the dedicated server would have the same restrictions on tasks, that need to be in sync.

 

Maybe this old example helps to understand the limitations of multicore through synchronisation problems:

 

https://forums.eagle.ru/showpost.php?p=1728877&postcount=36

Edited November 20, 2016 by shagrat

[Worrazen]

Textures and objects need to be in RAM when you pass them to direct x for rendering.

 

But they're not staying there (long enoug?). It has to fetch the same stuff off HDD frequently. That's what I'm seeing.

 

I cycle enough so I get back to the same airport and if it stayed in RAM it obviously wouldn't need to read from HDD again right.

 

 

 

Just have a look at the overall textures and .edm object models. That all together in uncompressed form would be the RAM requirement.

 

I'll have a look - but I haven't seen any .edm files being read in these tests in Process Monitor, a map with several cargo planes, 2 su-27, 2 thunderbolts, some tankers and awacs. I wasn't testing spawning ... yet.

 

EDIT: I've recorded .EDM files for the pilot ejection sequence - it makes one big hang when the pilot comes out. This is while in F2 view. Okay for some other far out pilot, but you'd surely like to see a smooth ejection of at least your self.

 

So, heh. Here's an argument you can use now, maybe the low VRAM makes the dynamic loading thing act a bit out of whack and may properly load things when I present the game with 8 GBs.

 

But nah, surely we won't want to waste time on the game trying to predict when the ejection will happen, right, got to load this from the beginning.

 

 

EDIT2: As I have expected the pilot HDD loading didn't happen in 3 more tries I did yesterday, I'll do it today again when I have time, I did reboot the PC between tries and emptied Standby Memory to help prefent DCS files from being cached in RAM.

 

 

Now the new render engine (EDGE) supports dynamic loading. It is by design to preload terrain and objects around the player aircraft and drop textures, objects etc. when not needed for rendering.

The idea is to get optimum performance with the scene around the player when he moves over the scenario.

 

Right, but what's wrong with extra textures just sitting in RAM doing nothing most of the time, would that really affect performance for the active play area?

 

 

 

Compared to the limitations of the old DCS before EDGE this is simply fantastic...

 

Well, I think that it could be something to do with my low VRAM that makes these things more pronounced than for you and others, so I made another video as well. This time with HDD activity LED. The WPF trace was done separately, it's not running in the background on this video.

 

 

 

 

Fancy external views are nice and good for screenshots, but to jump hundreds of miles from one plane to another in real-time, well isn't the most important task...

 

I agree, but then this will be my wish, if this is so in general and not only for my current test case, I will try to make a case for this texture loading thing to max out RAM, and hopefully it would not be the hardest thing to implement and could maybe be done for 2.5 (in perfect scenario) or to at least improve. I am reporting findings as I go in this thread in more detail https://forums.eagle.ru/showthread.php?t=177026

 

I just figured out "video mem" reported in DCS own diagnostics is a lot more ... at the start it's 3900MB and 5444MB max (1.5.5 caucasus) - as a reminder my current GPU only has 2GB VRAM.

 

EDIT: "Video mem" could be the total that's on VRAM and shared on RAM, but not sure unless someone confirms this is the case with this value.

So If I had enough VRAM, all that would be used.

Edited November 21, 2016 by Worrazen

[JDarksword]

In my case better multi core optimization would significantly boost my fps considering I'm on a 3.2 GHz machine, fairly weak compared to the 4.6 GHz monsters I've seen people using. This is most likely why I only get around 30-50 fps where as in most games I get 60 fps at ultra. I don't know, maybe it would be too expensive, but I certainly toss $50 over if it meant the potential for greatly improved performace.

[BitMaster]

I will try a RAM drive once my new stuff is here, just asking myself if 48gb is enuff to make it happen.

 

How have other dev teams solved desync when they moved to SMP ? What was their approach in deviding tasks ?

 

PFM and Bullets would not nec. need to be done by one core, same to ground warfare I guess !??

 

Driving 50+ vehicles around also should not need to share core with airborne systems.

 

There are many things that come to my mind that by my novice understanding would not need to share core to prevent bad things.

 

Mind you, open up any hostets site and check cpu. Most are low clock low core xeons.

Using those cores would save BIG money as you would not need to upgrade the package.

 

No Xeon runs fast enough btw imho, and DC admins dont oc ;)

 

 

Gesendet von iPhone mit Tapatalk

[Worrazen]

I'm beginning to uncover the HDD-loading issue kinda a more bigger deal than anything else for me right now.

 

I completely agree it's not meant to be smooth viewing those distant airfield cameras, especially when I wasn't maxing out the PRELOAD RADIUS setting :doh:

 

But ejection, crash, and respawn should kinda not be part of the dynamic loading thing, okay this doesn't happen that often in campaign, but what about testing, setting up controls, free flying, etc

 

The view movement stuttering (it feels like mouse sensitivity drops, just doesn't want to move sometimes) may be the VRAM swapping that would disappear once I get a new GPU

 

 

But now that I finally have NTTR & A-10C, I don't get below 2GB VRAM even when lowering all settings, ... I'm not sure whether or not this is worthwhile to keep testing until the merge, it's complicated because the behavior isn't consistent, sometimes it's smoother, sometimes not.

 

I'll probably have to wait til atleast when I get the new GPU, ofcourse continuing in a proper thread not here anymore.