mingus

64 bit and you soldier!

109 posts in this topic

well, to be fair 64bit does nothing really in relation to multi-threading. we'd need to move to Vulkan/Metal graphics APIs to get significantly better multi-threading for the client renderer. I'm sure there are other places that can be threaded, but the big one "the client 3d graphics portion" from what i've read even openGL 4 doesn't do a great job with threading.

 

for those that don't know, Vulkan is essential OpenGL Next, its a logical extension to OpenGL and has a wrapper allowing Vulkan to run over top of Metal which is a MacOS specific graphics API.

 

now the work for 64bit can't hurt a future move to Vulkan, however, the two are not directly related nor is threading necessarily connected either. 64bit allows for more memory and some slight speed improvements to specific types of operations - it won't double your FPS nor extend the life of older hardware though.

Share this post


Link to post
Share on other sites

Viz limit (not raw distance, amount of people) is largely a communications thing, not video.

If i make players a simple cube and put 5000 players in the same place, that is only 60k polys, no big deal video wise, laughable.
But that is 5000 network data streams that must be all tied together so they are closely in sync with each other.
Large organizational and management task in the networking area, and packet size increases and many clients themselves may find themselves
unable to keep up

That is the biggest reason for the viz limits, some networking guru could probably add a lot more useful detail to it

as for distances, i am not sure exactly those are decided upon

 

Share this post


Link to post
Share on other sites

its more packet rate than size or bandwidth. that is solvable via the viz lists and new priority.

meaning, planes should essentially cull all non planes and maybe ships at maybe 5K so its outside the reasonable ranges for ~40mm AA. by cull i mean player objects. you can know about them, but there is no reason to send data to/from infantry if a plane is 5KM away.

but what about proper level bombers from altitude you ask ... 'STOs' need to take over here and likely a one way data path from plane to infantry so they can see the circle at 5KM but this doesn't need to be a two way communication path.

so, example

5KM:

allied plane at 5KM and axis plane at 5KM have a two way communication

allied plane and axis infantry at 5KM have a one way communication path with the plane saying "im flying here and in this direction"

=>6km

allied plane and axis plane have two way communication

allied plane and axis infantry have zero communication

 

there are ways around the technical issues - likely needs a thread dedicated to tracking in game objects and with 64bit this thread can have access to lots of RAMs.

Share this post


Link to post
Share on other sites

Wow, where to begin...?

 

I'd love to see video evidence of folks getting 250-300 fps during any battle at 1440p *with all graphics set to maximum* (that is, shadows on high, etc). I have the same specs as asimov (16gb of high performance RAM and a Core i5), however my CPU is overclocked to 4.5 ghz, and I have a GTX 1080 Ti (also overclocked). As far as I know, I was the first person in the world to upload 4k videos of WWIIOL gameplay to Youtube. In this video, everything is set to maximum, but importantly, normal maps are disabled (this was before normal maps were fixed).

 

At 4k, my fps can go into ~30's (or less) in intense urban battles.

You do *NOT* have 120+ fps at 1440p with all graphics at their maximum setting *in situations in which the number of players within 2 km is close to the game's vis limit*.

 

....and that's exactly my point.

There is no amount of optimization for multi-threading/multi-core on the CPU that will yield the supposed performance gains being discussed here.

Hopefully the Rats will post here and state the plain truth to dampen down these wild expectations before things really get out of control.

Client-side performance gains from the switch to a 64-bit executable will be modest in the best of circumstances or even essentially non-existent in terms of framerate.

Let's just be crystal clear here:
The visible game world you see (that is, the "vis limits" madrebel is talking about) is rendered by your GPU. Period. If you increase those limits, rendering objects at further distances, or rendering *more* objects at the same distance, you will increase the render load on the GPU. All else being constant, your framerate will decrease. A 64-bit executable has nothing to do with this. The #1 reason CRS is switching to a 64-bit executable is because they have to due to an impending macOS restriction; any performance gains or future enhancements with possible graphics engine updates is a distant #2 reason for the switch. Again, the 64-bit executable will yield little (if any) fps improvements. Future work to optimize utilization of multiple cores would likewise yield only modest performance benefits (if you doubt this, google benchmarks for, say, 8-core vs 4-core performance on big-budget games specifically designed to utilize those cores; the fps differences for even *twice* as many cores is typically paltry).

 

 

 

If you think I'm wrong, here's a simple test you can do offline or wherever:

Spawn in with all graphics settings set to the lowest possible setting; every effects checkbox unchecked, infantry detail set to minimum, no corpses, lowest shadows...and try a resolution of, say, 1024x768. Note your fps.

Now set all graphics settings to the highest possible setting; all effects enabled, shadows on high, inf detail max, radial clutter max radius and density, high shadows, etc. Set your resolution to the highest possible. Reload the game if necessary. Spawn in and note your fps.

 

Heck, an even simpler test:

Try shadows low vs high to see an immediate difference.

Edited by xanthus

Share this post


Link to post
Share on other sites
1 hour ago, Merlin51 said:

If i make players a simple cube and put 5000 players in the same place, that is only 60k polys, no big deal video wise, laughable.

 

Sorry Merlin but this is completely wrong.

Go ahead and set all of your graphics to max, spawn offline and *with your fps visible*, record a video of you putting down clones until you reach some arbitrarily high number (I'm not even sure what the limit is). My claim is that your fps will go down. At 5000 clones (if it were possible), your fps will probably be in the single digits, a slideshow, or a CTD.

Obviously, players are not a simple cube, *and that really matters* because the number of polys increases dramatically. And even if they *were* cubes, please feel free to tell us how the new city buildings we've seen in recent WIP screenshots are effecting fps in cities like Antwerp?

My claim is simple:
More polys => lower framerate.

This is just as true for inanimate objects like buildings as it is for inf or vehicle models. Again, if you doubt what I'm saying, then show me that adding a couple of new buildings to cities and towns, increasing poly counts by what should be "no big deal" has no impact on fps.

Edited by xanthus

Share this post


Link to post
Share on other sites
3 minutes ago, xanthus said:

Sorry Merlin but this is completely wrong.

Er i dont think you are following what i said, different discussion that has nothing to do with FPS

Share this post


Link to post
Share on other sites

orly4k_zpszuoknh9f.jpg

 

note, the same GPU is also running rogan's live podcast which does in fact increase total load. That shot was from just now looking into haybes which just fell to the axis. 79fps, 42% GPU load.

i have a WORSE GPU than you by a significant amount but a better CPU in both capability and GHZ. That's why I have better FPS than you, at 4K.

for comparison, here is a partial shot from rocket league, note the significantly higher GPU percentage utilization.

rl4k_zps1pi7ayv9.png

 

you're understanding of the bottlenecks is flawed.

 

Share this post


Link to post
Share on other sites
4 minutes ago, Merlin51 said:

Er i dont think you are following what i said, different discussion that has nothing to do with FPS

 

1 hour ago, Merlin51 said:

Viz limit (not raw distance, amount of people) is largely a communications thing, not video.

I'm saying that this is completely wrong, and I posted why.

Share this post


Link to post
Share on other sites
11 minutes ago, xanthus said:

My claim is simple:
More polys => lower framerate.

nobody is disagreeing with you. however, you're of the mind that the GPU is the bottleneck, it isn't, its the CPU.

 

Would more people 'visible' slow down FPS, yes. The CPU is that bottleneck though and, further, to ground units it won't change much of anything as two AOs are rarely close enough to cause any additional load. further, this could be managed via limiting how/when players are even drawn based on priority lists and or unit type.

 

meaning, infantry rarely need to know about anything past 2KM - so why draw it or send packets about it?

Share this post


Link to post
Share on other sites
7 minutes ago, madrebel said:

That shot was from just now looking into haybes which just fell to the axis. 79fps, 42% GPU load.

 

Why are you posting GPU load percentage? And my CPU is overclocked to 4.5 ghz (clock speed is all that matters for WWIIOL); how about your CPU?

Yes, I get 79 fps looking in the direction of a distant town with one blue nametag above it. I don't understand what your post has to do with what I'm talking about?...

 

Show me your FPS with all graphics settings at minimum and low resolution (1024x768).

Now show me your FPS with all graphics settings at maximum and highest resolution (1440p or 4k or whatever your max is).

 

My claim is that your FPS will not be the same. GPU load % is irrelevant here. Your fps will be different but in fact your GPU load % should be roughly the same.

 

2 minutes ago, madrebel said:

Would more people 'visible' slow down FPS, yes. The CPU is that bottleneck though

 

No, it is not. This can be shown with offline clones as well.

Edited by xanthus

Share this post


Link to post
Share on other sites

tell me what cause fps different by look at around  suppose  looking at east  i get 124fps , north starting dropping to 50  then west hit below 20 where battle going on

Share this post


Link to post
Share on other sites

4.8GHZ and generally faster as I have more CPU cache than you being i7 vs i5.

you said more view distance can't be handled by the GPU - which isn't correct. GPU% proves that. CPU has always been the limiting factor, not the GPU.

further both of us at 4K is 4x the average so you and I can't be counted when talking about increasing view distance as we're niche. 1080p is the gross general standard, according to steam stats iirc by about 80%. can typical CPUs at typical 1080p sizes handle more information? easily.

and again, there are ways to manage keeping the info down if the ranges double.

Share this post


Link to post
Share on other sites
15 minutes ago, madrebel said:

you said more view distance can't be handled by the GPU - which isn't correct. GPU% proves that. CPU has always been the limiting factor, not the GPU.

...can typical CPUs at typical 1080p sizes handle more information? easily.

 

GPU % does *not* prove that. In fact, it proves nothing. Tell me the GPU load % for, say, Battlefield V at 4k or some other big-budget, bleeding-edge title. BFV is pretty gross, but it was the first title that came to mind...

1080p vs 4k has more to do with GPU than CPU.

 

Like I said, if you doubt what I'm saying, just try the tests I mentioned:


Max vs min graphics. Show me your FPS.

Max vs min resolution. Show me your FPS.

Max offlines clones vs no offline clones. Show me your FPS.

Is your FPS different in the middle of nowhere and in the middle of, say, Antwerp? Show me your FPS.


All of these are GPU, not CPU. Your % GPU load doesn't change that fact.

 

CRS knows that what I'm saying is true. CRS knows that more polys => lower FPS, and they know that better CPU optimization won't change that fact. They are grappling with these facts right now as they test out things like new buildings (for which they have tight performance requirements). They know full well that new buildings and anything else that increases poly count (even a little) will potentially decrease FPS.

Edited by xanthus

Share this post


Link to post
Share on other sites
2 minutes ago, xanthus said:

All of these are GPU, not CPU. Your % GPU load doesn't change that fact.

if this were true, you're GTX 1080 should get double the FPS i get. yet you get fewer FPS than I do on average, at 4K.

 

why?

Share this post


Link to post
Share on other sites
6 minutes ago, madrebel said:

if this were true, you're GTX 1080 should get double the FPS i get. yet you get fewer FPS than I do on average, at 4K.

 

Why should I get double fps?

And I don't agree that I get fewer fps than you do at 4k; I think I get more. That's one reason why I'd like to see your FPS in the situations I described.

 

Another thing for anyone who doubts what I'm saying:

How long have you been playing? In all that time, did you ever upgrade your video card while keeping the same CPU? Did you see a performance increase or not?

In fact, I claim that the single most reliable piece of hardware to invest in for performance gains in this game is GPU not CPU (just like it is for every other game). WWIIOL was CPU-bound in 2002. It is not CPU-bound in 2019 (just like pretty much every other game).

Edited by xanthus

Share this post


Link to post
Share on other sites

KFS1 discussed the vis distance / interaction list limit problem at some length several times in the old Design Forum. His explanations were consistent with Merlin's comments, i.e. the limiting factor in vis distance / interaction radius is not the GPU, but the quadrupled calculation/communications load to determine which units in that area will be included in the interaction list. 

The calculation workload is both at your end, to implement the prioritization rules, and at the server end to determine who is relevant to you. The communications load is the packet stream of course. 

We were told that the packet size and rate would not be increased because it of course drives a key CRS expense item. Maybe that's no longer current info, of course.

Share this post


Link to post
Share on other sites
7 minutes ago, jwilly said:

KFS1 discussed the vis distance / interaction list limit problem at some length several times in the old Design Forum. His explanations were consistent with Merlin's comments, i.e. the limiting factor in vis distance / interaction radius is not the GPU, but the quadrupled calculation/communications load to determine which units in that area will be included in the interaction list. 

The calculation workload is both at your end, to implement the prioritization rules, and at the server end to determine who is relevant to you. The communications load is the packet stream of course. 

We were told that the packet size and rate would not be increased because it of course drives a key CRS expense item. Maybe that's no longer current info, of course.

 

Nothing KFS1 could say can refute plain reality:
Go offline and see what happens to your FPS as the number of clones you spawn increases.

Whatever *else* is true, more polys => lower FPS. It really is that simple. If it weren't, our buildings and terrain would like *VERY* different; nothing KFS1 said can change that fact. If any Rat disagrees, then it's fair to ask why we don't have higher poly models in the game like, *right now*.

If this seems obvious, it's because it is.

If CRS introduces higher poly models for game objects, higher resolution terrain, etc etc FPS goes down. This is what they see when they load the game with these new objects replacing current ones.

The reason they're able to introduce e.g. better-looking buildings like the ones in development now is because of better GPUs.

Edited by xanthus

Share this post


Link to post
Share on other sites
6 minutes ago, xanthus said:

 

Why should I get double fps?

https://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-Ti-vs-Nvidia-GTX-1060-6GB/3918vs3639

i have a GTX1060 stock you have a GTX 1080 overclocked. If this game gave a crap about GPU, you'd have significantly higher FPS than I do, but you don't. why?

you have i5 running 4.5ghz

i have i7 running 4.8ghz

thats a hint ...

 

Share this post


Link to post
Share on other sites
2 minutes ago, xanthus said:

 

Nothing KFS1 could say can refute plain reality:
Go offline and see what happens to your FPS as the number of clones you spawn increases.

Whatever *else* is true, more polys => lower FPS. It really is that simple.

I don't think that anyone is arguing that more poly = lower FPS. I think it more that performance and graphics is more nuanced.

Now I'm no expert but I do build my own PCs, so I always look at things that can bottleneck performance. The classic example is getting a GTX 1080ti and pairing it with a 2-core Intel i3. Thats because you need the CPU to send data to your GPU for it to process it.

JayzTwoCents did a nice video on this:

 

Share this post


Link to post
Share on other sites
11 minutes ago, jwilly said:

We were told that the packet size and rate would not be increased because it of course drives a key CRS expense item. Maybe that's no longer current info, of course.

its not anymore. bandwidth is really inexpensive at COLOs and isn't metered like it used to be. I mean, there is still a technical upper end but we're nowhere near bumping into that. also, we officially dropped dial up support in ... 2012? those comments from KFS1 as I recall were still restricted by dial up support.

Share this post


Link to post
Share on other sites
17 minutes ago, madrebel said:

https://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-Ti-vs-Nvidia-GTX-1060-6GB/3918vs3639

i have a GTX1060 stock you have a GTX 1080 overclocked. If this game gave a crap about GPU, you'd have significantly higher FPS than I do, but you don't. why?

you have i5 running 4.5ghz

i have i7 running 4.8ghz

thats a hint ...

 

 

Again, my claim is that I have higher FPS than you at the same resolution and graphics settings. Probably much higher. I've said how we can test this.

How about we just upload screenshots or video from an offline spawn?

 

And again, just to be clear on what I am stating about this game:

- The switch to 64-bit executable will result in only modest (or no) performance gains on average.

- Above some threshold, polygon number and FPS are inversely correlated.

- To the extent that anyone in this game has "low FPS," and they are using a modern, midrange CPU, the cause of their low FPS is their GPU, not CPU. They will see performance gains by upgrading their GPU.

- All else being equal, core i5 @ 4.5 ghz vs core i7 @ 4.8 ghz will yield no reliable FPS difference; but GPU differences most certainly will.

- If vis limits are increased (either visible player limits or visible distance limits or both), fps will go down, 64-bit executable or not.

- Vis limits beyond 2 km are likely infeasible and probably always will be, at least in *this* game.

- Vis player limits above 128 are likely infeasible and probably always be, at least in *this* game.

 

 

Nearly 20 years ago, CPU was what mattered most for performance in this game. Given today's run-of-the-mill CPUs and motherboards, that is no longer the case. Today, GPU is what matters most for performance in this game (just like every other game). In fact, the posted video *actually corroborates* what I'm saying: modern CPUs DO NOT constitute bottlenecks on big-budget titles for anything but the most high-end, expensive GPUs.

Edited by xanthus

Share this post


Link to post
Share on other sites

This thread changed topics. Merlin and I both were responding to this post of Mad's, which pertains to vis distance / interaction list, not frame rate or display resolution:

 

Share this post


Link to post
Share on other sites
4 minutes ago, jwilly said:

This thread changed topics. Merlin and I both were responding to this post of Mad's, which pertains to vis distance / interaction list, not frame rate or display resolution:

 

Vis distance is tied to frame rate. Of course...?

Interaction list is tied to GPU insofar as the objects with which you're interacting are actually represented as rendered polygons and textures. I can perfectly accept that none of that would matter for GPU if these objects were invisible, but they're not, and that's my point.

Edited by xanthus

Share this post


Link to post
Share on other sites

 

25 minutes ago, xanthus said:

 

Again, my claim is that I have higher FPS than you at the same resolution and graphics settings. Probably much higher. I've said how we can test this.

How about we just upload screenshots or video from an offline spawn? 

go for it, i already provided a SS from in game at what was a pretty active battle. run rivatuner, post away.

which i5 do you have? there is only 1 on this list faster than my 4790k which sits at #13 still.

https://cpu.userbenchmark.com/

again, if GPU were a factor, you'd crush my FPS - you don't though.

7 minutes ago, xanthus said:

 

Interaction list is tied to GPU insofar as the objects with which you're interacting are actually represented as rendered polygons and textures. I can perfectly accept that none of that would matter for GPU if these objects were invisible, but they're not, and that's my point.

no, the GPU in this game is only really interacting with the pixel fill rate and texture fill rate of the GPU. none of the shaders, and none of the pre-frame setup/GPU offload that has occurred in client render engines since 2001 is relevant here. this is the part you're not getting and also why GPU utilization in this game is so low.

all of the scene setup, all the lighting, everything is setup VIA the CPU and handed off to the GPU to just draw - which was standard practice circa late 90s/early 2000s ... when the client render engine was written.

having 256 infantry near you wont really change the load if we push from 6KM max view range to 12KM, as infantry aren't in fact drawn past Xm anyway. you won't all the sudden start actively drawing (meaning having to leverage frame buffers or anything related to 'drawing more pixels/polys/textures') infantry in a battle 8KM from you - even if the distance is doubled.

 

understand?

Share this post


Link to post
Share on other sites

here, these SSs explain it better. i'll see if you can figure out the why.

vsync off graphics maxed, all at 4K. note the GPU % utilized in these first two shots as well as the FPS.

offline4k01_zpshkfvzfzu.png

offline4kwithclones180away_zpsogwc8bas.j

note the GPU% and the FPS. now, 180 degrees behind me are a bunch of clones

offline4kwithclones01_zpsjjlole4p.jpg

note the GPU% and the FPS.

explain why in the first two I have significantly higher GPU utilization % and higher FPS yet when i turn around 180 degrees with the clones in view my FPS plummets AND so does the GPU%. logically, if the game were GPU bound even remotely, the last SS should show 100% GPU utilization. yet it doesn't - if you understand why here, you'll realize why doubling the view range will have very little impact on local AO ground performance.

 

*edit* hang on fixing links

Share this post


Link to post
Share on other sites

  • Recently Browsing   0 members

    No registered users viewing this page.