Educational Post - CPU realities

[Guest Big Time Software]

Well... I wrote this up to answer a particular question about adding more terrain variations to CM (check out the "Towns and urban areas... Looks like Main St. USA" thread), but after I got done writing it I decided that it should go up as its own post. It is important to educate you guys so that you understand what your computers can and can not do (BTW, this post is not aimed at anyone, rather EVEYONE ). Many people think that today's hottest chips are capable of anything. Well... that is pure folly. People thought that of their first P5 and PPC 601 running at 60MHz

The truth is that today's computers are still quite incapable of easily dealing with even "simple" simulations of real life. Nobody knows this better than Charles. If he wasn't such a fantastic programmer (the best I have ever met, no lie), CM would be nothing more than a minor step up from the existing wargame offerings. So keep in mind that for every one thing that CM isn't delivering, you are getting about a dozen things that no other game offers. Most Grogs tend to look at the glass as half empty, but if our glass is half empty than some other game companies are pretty thirsty So, am using Ken's question about house placement as it relates to hardware realities to make a larger point (Ken, absolutely NOTHING wrong with your question, so please take no offense). Keep in mind that Charles is in "exile" from the boards until the AI is done so much of this is non-technical and second hand from him...

No, you can't place a house anywhere within a tile. The house has a terrain tile "stuck" to its bottom already. So when you put down a house the ground goes with it with the house in a preset position. There is a very good reason for this that even I did not fully understand when CM changed from hexes to tiles (oh... about 2 years ago!).

The lack of total freedom is due to hardware I am afraid to say. All the really "expensive" CPU/RAM elements of the game hate variety with a passion. Exponential hate too! The houses can already be in any one of 8 positions, which to the system is 8 unique house types (or possibly 3 for some calculations?). If you could put a house anywhere on that tile you wanted the number would go up exponentially (like many hundreds of potential positions). This means that the various systems like LOS can't make very important assumptions to lessen the calculation load. Instead the calculations would have to be made on a 2x2m basis and, as stated above, nobody's system would be happy with this. Turn computation time could take 10-15min per turn on a P500 (just an illustration, could be more or less). Rule is that each thing added makes for significant increases in CPU and RAM usage. The "assumptions" BTW do not affect quality of calculation results, only CPU cycles needed to get them.

Something that people don't realize is that ALL games have such limitations, even 2D ones that are much more simple in the simulation aspects. Close Combat, for example, has a tile structure under its maps too. We have the understanding that their tiles are 6m x 6m. Seeing as their maps are at least less than 1/2 our max size (someone know what CC3's max map size is??) it makes sense they can make the terrain tile resolution finer, even if their ability to customize height and LOS within a tile is lower (i.e. our 20x20 tiles are actually broken up into 2x2m subtiles for this). But on top of that CC's number of calculations being made pale in comparison to CM's, which is probably how it can be realtime. This smaller calculation point isn't a slam on CC, but a reality statement. They have a max of what, 30 units on the board at any one time. CM can have literally hundreds. It is the functions of the units that require all the CPU cycles. LOS, LOF, movement, etc. And the more terrain there is causes a huge increase in needed CPU cycles for EACH function. The overall impact could mean the difference between playable and unplayable.

Then there is the well established limitation of VRAM, which adds to our problems with terrain variety. Not only does more and finer terrain screw with the CPU, but it also requires more art. More art means higher VRAM needs. That could mean tossing out all the people that have 8MB cards. Having an 8MB card myself, this would be bad Seriously, even if we COULD increase the number of terrain types and permutations without CPU problems, we would have to toss out a very large chunk of potential customers. Maybe the majority. That would mean going out of business...

OK, so that is 101 in CPU realty! Now to answer the rest of your questions...

Currently we have 5 different types of buildings:

1. Small Houses (wood)

2. Small Houses (stone)

3. Large Houses (2 story)

4. Church

5. Large Building (free standing or adjacent)

As for the immersion factor, I don't miss brush or wooded trails at all. There is SO MANY other things to think about. If CM were just a "push your counter across the board and roll the a D6" kind of game I think I would. But CM's detail level is just so freaking insane that you won't miss boulders, picket fences, telephone poles, house interiors, or any number of very real, but very minor, details.

Having said all of that, we DO want to have all these details in CM at some point in time. But that isn't up to us as it is up to the pace of hardware improvements and customer adoption. We could add things like trials and fences, but we have to keep an eye on total polygon and VRAM usage. Panzer Elite might have telephone poles, for example, but they don't have 30-50 vehicles and hundreds of figures running around either. So we MUST prioritize the importance of each and everything that goes into CM. We don't want to have limitations on vehicles and troops, so we have to have limitations on terrain. We are SURE that we have the correct balance, one that gives a VERY high degree of realism and fun. But of course, we have the benefit of playing it

So yes, as time goes on CM will become better and better. It will not stagnate, I can promise you that. BTS believes 100% that nothing can every be perfect, but you should always bust your ass to try anyways But until the average customer has gigahertz computers and 64MB 6th generation video cards, we will probably always have to leave something out. So the choice is take what is possible now or wait for about 4 years for the "perfect" (right!) game.

Thanks!

Steve

[Thorsten]

"Close Combat, for example, has a tile structure under its maps too. We have the understanding that their tiles are 6m x 6m. Seeing as their maps are at least less than 1/2 our max size (someone know what CC3's max map size is??)"

The maximum map size in CC3 is 576m x 408m. Terrain elements are 2m x 2m. Megatiles are 8m x 8m, but they have nothing to do with calculations, and are merely used for troop deployment, plotting artillery barrages, etc.

Just so you know, I got this info from Cappy-R. I'm not much of a mapping expert myself.

Thorsten

[Lee]

Well said BTS. It's very simple, we will have to wait

for all the extra goodies that we'd like to see in

CM but aren't very important to the quality of the

simulation's realism. Personally, I'll take all the

detail I can get now and wait for CM to get better

and better with each new version to get all those nice

details that we'd like to see. The first gigahertz

systems are on their way next year, so the wait won't

be too long.

[Thorsten]

Ack... miscredit. I got the information from Mick(xe5), not Cappy-R.

Perhaps when you folks are all done coding CM, and adding all the things in that you want to after the initial release, you could code an experimental version of CM afterwards. Just have a small (100m x 100m, for example) area, and then see how detailed you can get. Individual soldiers instead of squads, smaller tiles, more unusual terrain and buildings. Then just put in a squad or two on each side and see what happens when they fight it out. Perhaps even release it as shareware to get other peoples' input.

Just a thought.

Thorsten

[Guest Big Time Software]

Tanks for the info Thorsten. CC's maps are roughly 1/3 what our max is. Plus, those "megatiles" are most likely used the way CM's are for many calculations behind the graphics. Also, and this could be wrong, I am under the understanding that a unit can only be in one tile at one time. So that 1/2 squad that looks so nicely laid out is actually in one 2x2 tile. Is this correct?

As for your suggestion... you are obviously not a programmer Such suggestion would take a couple of months to make. To do the AI, probably another 2 months. It isn't as if we can change 100 variables and have a different game. Doesn't work that way.

Plus, we don't WANT to make a tiny, microscopic slice of war where tanks start out 400m away from each other at best. If we did we would have already done that. We want to simulate a battalion size battle or lower, so that is what we did. All future versions of CM will be at the current scale, no ifs ands or buts about it

Steve

[Thorsten]

I don't know to what level of detail units or individual soldiers are modeled in CC3. I know that it does model individual soldiers, though. The largest Russian unit has 10 soldiers, so 14 of those teams, plus a 5 man command unit, makes a max of 145 soldiers. The largest German unit has 7 soldiers, so on their side, the max is about 103. Thus the maximum is 248 soldiers in a single battle. With this many, my P200MMX slows to a crawl. Now from visual indications, it doesn't seem like these soldiers are confined to 2m x 2m boxes, but visual indications do not always indicate what is in the code underneath.

I am a programmer, just not a very good one.

I thought it would be a fun idea to have something that was half way in between CM and Hidden & Dangerous, especially since the most common criticism I've heard of CM is the quality of the graphics, especially the soldiers. That and it's a whole lot more fun to see a squad of 10 men charging a position, than a symbolic 3 man representation.

Thorsten

[Rick]

Steve,

Thank you. I think the initial post that started this thread is one of the most important things written on this forum. Tells us what kind of things might be totally unrealistic to expect. We on this forum, have had a tendency to start whining when told something won't be in the game (I feel that all of us are at least a little guilty of this, so not pointing fingers.)

It's hard to think of the limits of CPUs when there is so much that they can do these days.

[Guest Big Time Software]

Frist, I must correct myself... CM's MAX size map is more like five to six times larger than the largest CC map. I was thinking "average" CM map vs. max CC map. Silly

Each squad in CM does simulate each individual man, but not his exact position. But if CC is behaving as we have been told it does, CC and CM simulate individual men about the same (in terms of position). So compare CC's 248 men to CM's 2000-3000 men. And in terms of graphical figures, we generally do have just as many on the map at one time as CC. And they don't have to wory about 3/4 views More isn't better, of course, but more means that there is less we can further ask of the CPU.

The thing that irks us about the Hidden and Dangerous thing is that it can NOT be compared to Combat Mission. You want CC with H&D qualtiy soldiers? Dream on and wait about 2 years We have just as 3D "figures" as Close Combat has 2D sprites. If you suddenly made CC 3D, it doubt their figures would look any better than ours. They CERTAINLY wouldn't be as detailed as H&D. How many figures does H&D have on the screen at once? And how far can you see in its world? CPUs and 3D cards can only do so much.

So again, some people are asking for something that can not be done. Thinking that it can be and actually being able to do it are two different things. Plus, wargamers are supposed to care about the game more than the graphics (which we disagree are as poor as some on the CC boards think).

Also disagree that it is a "whole lot more fun to see 10 men charging a position". I think it is a whole lot more fun to see a company acting in concert with armor and other formations in a realistic manner. The way CM simulates this graphically is very engrosing. Close Combat, as much as I liked #2, really left me laughing at its depiction of war. Click on the start button and wait to see if I kill of of the AI's tanks in the first 2 seconds so the AI will withdraw leaving my forces untouched? After playing a campaign where I did this about 15 times in a row, I got tired of it really quickly. I'll go for the realism thanks very much

Again, I liked CC2, but as an entertainment product more realistic than C&C, but not as realistic as CM (by a long shot). Lots of flaws in that game. Lots and lots of flaws. And the flaws were all in areas that I, as a Grog, care very much about. When I found out that CC3 had the same flaws as CC2, but WORSE in many ways, I decided not to bother.

Thanks Rick. The post was supposed to do just that. Make people understand that wanting something is OK, but that they must understand that computers like to argue with our wants all the time. I want a computer program that can type messages to you guys while I ride around in my Weasel (which I haven't driven in a month!), but I know not to expect anybody to give me such a program anytime soon

Wanting and expecting need to be closely linked together based on reality. Wanting more is good, but if it is not checked by realistic expectations, then everybody is going to be upset. Unrealisitic Ggamers are upset because we can't do x, we are upset because the gamer might as well ask us to cure cancer while we are at it. After all, "it should be a simple fix"

Steve

[This message has been edited by Big Time Software (edited 08-29-99).]

[Pak40]

Steve,

thanks for explaining some of the techno details, it's always helpfull to know this kind of stuff because it helps us understand and appreciate the game better.

One clarification about Close Combat's units. Close Combat tracks line of site to each individual soldier which would explain why players see only some of the soldiers in a squad at certain points in the game.

CC also tracks movement of each individual sodier as well as their phsycological state.

So while there may be a max of 30 military units in a CC game, the reality is that CC treats each soldier as a "unit". If both sides compose of nothing but infantry you could have as many as 200+ "units" on the map. I notice my P133 chugs when I play these types of games because of the sheer number of soldiers on the map that CPU has to keep track of.

Combat Mission only tracks each unit and not each soldier, correct? But I suppose that a 3D calculation for LOS and bullet tracking must be more complex for a CPU to calculate.

Any thoughts?

[Thorsten]

I certainly did not intend to start a Close Combat vs. Combat Mission debate. This fall I fully intend to shell out money for both CM and CCIV.

CC and CM are the same scale: 2m by 2m, and both have the ability to push my CPU to the limit, no arguments there. As for the level of detail and realism in CM, I understand that was a big reason for creating the game in the first place. It raises the bar for all squad level war games. I can only hope that Atomic takes it as friendly competition and rises to the challenge. If not, then I hope someone else picks up the real-time and 2d game models where Atomic left off.

One of the most positive things about CM is that there is room for improvement, and that the designers KNOW there is room for improvement. Too often, game designers don't have an idea on what needs to be improved, and their games end up becoming devoid of playability and are nothing more than pretty pictures. I see CMs place in 3d wargames the same as I saw Castle Wolfenstein's place in first-person shooters, the start of a new generation of games. Hopefully one that will have many descendants, all improving on the original.

Thorsten

[Guest Big Time Software]

Thanks Pak for the further poop on CC. Doesn't add up with some of the other stuff I have heard, but I suspect it is correct. As far as the 2D LOS vs. 3D LOS, oooooooh, you ain't kidding it makes a worse hit on CPU speed. And yes, although CM does track each man and his weapon to some degree, there is a lot of "unit" attributes that apply to all. But guess what? These are the kinds of calculations that aren't really difficult to do The tough ones are LOS, LOF, Best Path, and anything require the AI to do something with its units. So CM could technically simulate each man's stats the way CC does, but we felt with 1500 men under your control you wouldn't care to know all that. Lord knows *I* wouldn't care

Thorsten, sorry for turning this last bit of the thread into a CM vs. CC thing. I didn't mean to do it. However, it seems that the people who are MOST critical of CM are the CC folks. Even though most swear up and down at Atomic, curse every new and crazy bug or design flaw that is found, feel pain at some of the really bad gameplay that can result, and are already tearing into CCIV's design, we seem to be knocked about like a punching bag for not living up to this, uhm, high standard. Please understand that it is as frustrating to us as it is unfair. (CC folks aren't the only ones either. SP, SL/ASL, WF, EF, etc. all have their crowds punching us)

As for Atomic, Charles has known Keith for many years. We do wish CCIV well, but we don't really want to be compared to it or, worse, held to it as some sort of standard. No game system should be forced onto another. But people always feel the need to compare one game to another, so unfortunately we feel compelled to defend ourselves when the comparison is unfair or untrue. And since CM isn't actually out, we can't just hold the game up and say "judge for yourself". Right now some are happy enough to look at one screenshot and decide that CM is junk. Work on something fantastic for two years, without pay, be harshly and irrationally trashed, and then see how much (not you personally Thorsten) you like it. Trust me, it sucks

We don't claim to have a 100% realistic wargame, but we do claim to have the MOST realistic wargame ever made. The former is an ideal the latter is a reality that can always be improved upon. And CM, as good as it is right now, can be improved in the future. And that will happen! Thanks for recognizing our pledge to continually improve our games Thorsten.

Oh well... enough of my ranting!

Steve

[Ken Talley]

Hi Steve,

Don't worry about me, I definitely don't take responses personally < >. I did want to follow up on your 101 course because I am not sure if I reached the correct conclusion.

I did understand the explanation of houses needed assumed locations and unit functions in regards to CPU requirements. I did get a little lost on the exact cause of increased CPU and VRAM requirements in regards to types of terrain.

"And the more terrain there is causes a huge increase in needed CPU cycles for EACH function. The overall impact could mean the difference between playable and unplayable."

Am I assuming incorrectly that you mean the finer scale of terrain (2x2 vs 20x20)causes the increase in CPU requirements? Or do you mean the greater quantity of terrain (an entirely forested board would create larger CPU requirements)? Or do you mean the greater variety of terrain types even if at 20x20 scale increases CPU cycles? I hope you mean number one.

And basically the same question applies to VRAM.

"Not only does more and finer terrain screw with the CPU, but it also requires more art. More art means higher VRAM needs."

Do I understand properly that CPU cycles and VRAM requirements increase with the smaller scale items such as telephone poles, boulders, individual bushes, etc. that must be represented at the 2x2 scale? Also if you went from ten to twenty types of terrain (all represented at the 20x20 scale) w/i a given scenario would there be a dramatic increase in CPU and VRAM requirements?

I am trying to understand if your explanation results in a restriction of the resolution level of terrain (no less than 20x20) or a restriction in the quantity and variety of terrain even if capable of representation w/i the 20x20 scale. I don't mean to be a headache but I was intrigued by the explanation. Unfortunately, I am just not certain of the conclusion I should reach. <{>

Thanks

Ken

[This message has been edited by Ken Talley (edited 08-29-99).]

[Doug Beman]

Pak40 said:

"One clarification about Close Combat's units. Close Combat tracks line of site to each individual soldier which would explain why players see only some of the soldiers in a squad at certain points in the game."

I remember many, many occasions when one of my squads would be trying to hide behind a wall, but an enemy would see some of my soldiers and fire at them. However, when I tried to order my guys to fire back, or just get my dudes out of the line of fire, the LOS (drawn from the center of the unit or maybe the unit leader, not from individual soldiers) would be blocked by whatever wall my squad was hiding behind. This hampered my ability to order return fire, as well as just making sure my guys were hidden (the LOS would show "blocked" and my guys would still be dying from direct fire weapons)

I would be faced with two choices: try to shepherd the squad further behind the wall (not easy to do since I'm ordering a squad, and individual soldiers are moving) OR move the squad out from behind the wall so I can get them firing back.

LOS from each individual soldier should come with the ability to order each individual soldier to do stuff, like fire weapons, or move into cover, and so forth. Creating a dichotomy between information sources (individual soldiers) and order paths (only whole units) hampers control.

This is one place where I like CM's plan; it tracks individual soldiers, but only in areas where there is no conflict with giving orders to a whole unit. Weapons/ammo and casualties are individual, but I can't give orders about those things. LOS, movement, and the like are tracked "for the squad" and I can give orders about LOS etc "to the squad."

DjB

[Guest KwazyDog]

Thats a good point Doug. Im not really sure how LOS, etc worked in CC. Youd think if they did cleanly track each individual soldiers LOS, then in the situation you describe youd think the soldiers that could been seen would also fire back. I dont recall if they did or didnt, but I think not. The was one of my major peeves about CC, having to toy with your soldiers to get them into the right place so each cen fire.

I dont think this will be a problem in CM

[Guest Big Time Software]

Hi Ken, well I picked this time to do the 101 Course on CPU Reality because your feature requests seemed so "simple" and "easy" on the surface, but really aren't either. Trust me, I have had to go through this course with Charles just as you are now. I guess that is why it is usefull to point out that some things aren't as they seem (i.e. NOT easy )

<BLOCKQUOTE>quote:</font><HR>Am I assuming incorrectly that you mean the finer scale of terrain (2x2 vs 20x20)causes the increase in CPU requirements? Or do you mean the greater quantity of terrain (an entirely forested board would create larger CPU requirements)? Or do you mean the greater variety of terrain types even if at 20x20 scale increases CPU cycles? I hope you mean number one. <HR></BLOCKQUOTE>

Hehe... get this answer:

ALL OF THE ABOVE

Let's forget frame rates for a sec and only look at between turn calculations. The two are quite different.

The worst CPU killer is the number of potential tiles (any type) on a map. If you took our current game engine and made a 10k x 10k map the game would grind to a halt. Same if you took a normal sized 2000m x 1000m map and made the über tiles 10m x 10m. The more tiles you have on the map the more load on the CPU to figure out something like LOS or how to move from A to B. The 20x20m tiles are used to make computational assumptions which save loads of CPU time without sacrificing accuracy. I don't know the specifics, and don't want to , but I do understand the basic concept.

Because of the assumptions we can simulate larger maps with larger über tiles. If we made the über tiles 50m x 50m we could make even larger maps, but we would also have to bump up the subtiles from 2m x 2m to a larger size as well, since they also count against CPU cycles.

The more variety of terrain TYPES increases the load on the CPU to some extent, but not that bad. Mostly terrain types require more coding and VRAM to support rather than actual cycles to simulate. However, an exception is something like the house anywhere in the square does have a huge effect on CPU requirements. This is because it removes the ability to make assumptions.

Now, in terms of frame rate, the more terrain types you have doesn't mean a great hit on the 3D card and CPU. However, the more bumps and curves you have in the terrain mesh, the more polygons are needed, which means the slower the game will run graphically. Some terrain type's graphics also require more polygons, like woods compared to open ground, so they can cause higher hardware demands. And every doodad you add on the map requires polygons, so one telephone pole might take 22 polygons to make look good. Now multiply that by 50 or 60 telephone poles and you can understand why we left them out Sure, we could have put them in, but then something else would have to come out.

The ability for the computer to handle polygons is finite. You can do tricks to reduce the polygons being drawn (and we are doing LOTS of tricks), but you really do anything to make the computer draw more and draw faster without hacking the hardware (or so I understand). Because of this we have to be VERY careful to not put in too many things that require too many polygons.

And then we have to stick units in on top of this don't forget! And they are VERY demanding...

As for VRAM...

Every thing you can possibly see in a single scenario has a texture loaded into VRAM (exception are lines, like the LOS, which just use a SOLID color). So, if you add a boulder terrain type you have to have boulder graphics. These are stored into VRAM. The pixel size of the graphics determines how much VRAM is used up. The more graphics that are in a scenario (not just on the screen at once) the more VRAM is used.

Also the higher the resolution of the graphics, the more VRAM is used. I could make a 16x16 pixel boulder or something like a 128x128 one. The more pixels the better looking it is in the game. Unfortunately, the more VRAM it sucks up. So you have to be careful to balance quality with usage.

So the more terrain we have available, the more terrain will be in each scenario, the more VRAM is used up. Some terrain is really inexpesive (like the wall texture) while others are horrible (roads are the worst). And there has to be different tiles to add variety sometimes, like the open ground and multiple tree types.

On top of this you have units, vehicles, weapons, smoke, fire, UI graphics, etc. All of this ALSO has to go into VRAM.

The only bright side to this is that you only need to have the texture loaded in ONCE. So if you have 16 tree graphics you can have 1000000000000s of trees without hitting up VRAM for anything else. But of course each new tree hits the CPU and 3D card for drawing power. So there are practical limitations.

Having a smaller terrain scale does not necessarily mean a greater usage of VRAM. BUT, the more variety you have does. So in theory we can shrink the current über tiles down to 2m x 2m without taking up any more VRAM, but adding something like telephone poles would, no matter what the scale. But of course the CPU would raise some objections with both

The end point here is that the more stuff you put into the game, the more it causes a burdon on the hardware. Although some elements come cheaply, nothing comes for free. We have way more candidates for inclusion than the hardware has power and/or capacity to support. Therefore, somethings will have to sit and wait until hardware improves.

[Guest Lokesa]

I'm sorry but reading these CC2 posts have made me nostalgic with memories of my flammpanzer torching everything in sight on the "Shijndel arena" map and ambushing a tiger with my last piat dashing the enemys hopes for the win he thought he had. Great fun

I need to find a way to get paid for all the hours spent on these games...

oops, steve's post got in first making mine seem kind of out of place

Thanks Steve for the explanations as to how it all works.

[This message has been edited by Lokesa (edited 08-30-99).]

[CoolColJ]

Steve

How much of a speed increase would you get by coding in assembly or whatever low level language risc processors use, rather that C++. This would make the gamers smaller as well right? I remmeber in the old days, mosta AMiga games used assembly, which was significantly quicker than c coded programs, and a far bit more compact.

CCj

[Greg Deych]

I'm not a professional programmer (yet), but AFAIK assembly programming as a way of building sizable projects is dead. You may use it in programming drivers or other mission-critical apps which need to be small, but it's really not possible to do CM-sized project in assembly. Not unless we want to wait another year and half for Steve and Charles to finish it up.

To all of you who want more stuff...have faith in Moore's law. That gigahertz processor should be here in time for next summer...just about the time CM: East Front is slated to arrive.

Of course, we'd have to allow for the time for everybody to get it, say another 6-8 month, which actually makes it more in time for CM: Med. Oh well.

[Guest KwazyDog]

Yikes, the thought of having to do CM in assembly just gave me a cold shudder

[Guest Big Time Software]

Greg, very correct about the assembly thing. Very dangerous to do, especially with the fast rate of chips cycling in and out of the market. Hell, we don't even want to develop for API sets like MMX or whatever. Even these are dropped and/or redone with great frequency.

As for the gig processors... Intel also said Merced would be out how many years ago?

Steve