64b SB Pro PE and Multicore

[Werewolf]

CAVEAT: Been using PC's since 1978. Most folks I know at both work and personally believe me to be a computer whiz. They're wrong. I just know more than them. I'm far from being a hardwar guru and don't even come close to being an engineer so if my supposition about SB PRO and multicore as described below is somewhere out in left field don't be shy about correcting me.

Been running some tests using the 64b version of SB Pro. 3.028 I believe is the version.

My system is i7 5820K 3.6Ghz 6 Core, 16GB ram, nVidia GTX970.

Been using MSI AfterBurner, HWMonitor, win7 performance monitor, MS process explorer and task manager to monitor temps, how SB uses memory, the GPU and CPU.

I started monitoring because when I cranked up the graphics levels in Pro PE I saw frame rates drop down into the low 30's which is only about 10% better than I got on my old PC that had an i7 2600 series 4 core processor, a GTX 560 and 8GB of RAM. I thought that FPS to be rather low considering my current rig's capabilities. EX: With GTA 5 settings all maxed out except those setting that would make GPU memory usage go over 3.5GB I average low 50's. For Witcher 3 set on highest every thing I average high 40's to low 50's and those two games are known to stress the hell out of even the likes of a GTX980. With most everything else I get 60+ FPS.

On Pro PE the max GPU usage I've ever seen with all the bells and whistles on is 53% and it averages 47%. Card memory usage never goes over 31% and the Temp maxes out at 50+C and that's the lowest of any game I play. Most push 70C and some move the temp up to 80C.

So I started monitoring CPU and it doesn't appear that the load across the 6 cores and the other 6 hyperthreaded virtual cores is balanced at all. Usually just one Core is carrying almost all the load. Average CPU usage with SB PRO PE running is just 22%. On most games the CPU temp hits 48C but with SB PRO it runs between 42 and 45C. NOTE: liquid cooled CPU

Point is that the system isn't even coming close to being stressed out.

Framerates - even with everything maxed - out should - in my not so educated opinion - be pushing 60 at the very least but I'm lucky to get into the 50's even on a small scenario and into the mid 30's to low 40's on large scenarios. Let's not even mention Terrastan/Terrasim - run anything on it and FPS drops into the 20's.

I don't believe that the bottleneck is graphics card. I think the bottleneck is the CPU and it's not really that either. I'm gonna just toss it out there and guess that the real performance bottleneck is that SB PRO isn't efficiently using all 6 cores and the other 6 virtual cores to do the intensive calculations necessary in a sim like SB. I'm gonna guess it's using just one and any core switching going on is being done by Win7 because I never see more than 1 or 2 cores loaded down when SB is running.

Which begs the question: if that supposition is correct will the coming upgrade include a change that implements full usage of all cores available? Wouldn't that go a long way towards improving SB PRO PE performance.

[Ssnake]

a) No

b) Yes, it would.

We're dealing with legacy code. Switching over to multithreadedness is incredibly risky. Modern computer games use modern engines, and one benefit of these is that they utilize modern hardware much better. We know all this, but as we have to change engines while we're driving the car in a mountain road race we can't implement radical changes without the very real risk of breaking more than we can fix before careening off the cliff behind the next road bend.

That's not to say that we will never switch over to a truly multithreaded application. We recognize the benefits. Still, we must balance them against the risks of such a move when making the decision and as you can tell, so far the benefit doesn't clearly outweigh the disadvantages. We rather recommend that you'd cut back a few of the graphics and detail settings instead.

[OnAlienware]

Is the path forward incremental change, or a bold rewrite in a future update?

[Ssnake]

From what I'm hearing from experienced software veterans, bold rewrites end in a fiasco more often than not.

[Zaphod]

For those creating scenarios , what do they have to watch out for that chews a lot of processing power.

Is there a priority list of some sort?
I assume other than size and how many units

[Ssnake]

Cutting the visibility range both optical and thermal is a rather effective step. Note that the thermal range is up to 3x the optical range (if no precipitation is involved), so 3,000m optical range plus some rain or snow is a good way to keep framerates up.

 

Of course that's a workaround that cannot be applied to every possible scenario, but where it's justified, it's a good idea.

[ashdivay]

I've seen massive lag in games with extensive scripting of brigade+ level as units on map means more complex processing load on cpu and gpu load. 

I am theorizing a possibility of running sb in a vm with vcpu being balanced across multi cpus. 

[Ssnake]

Are you talking about single player (then it's not lag), or multiplayer events (then lag is probably a bandwidth/server/internet infrastructure issue)?

[ashdivay]

10 hours ago, Ssnake said:

Are you talking about single player (then it's not lag), or multiplayer events (then lag is probably a bandwidth/server/internet infrastructure issue)?

 

I had lag in single player mission.