Beta build 71275 is up!

[MavericK96 0]

Can someone chime in and help explain something? I have a full-page spreadsheet of beta data from the last month and I need to make sure I'm not missing something that may be of value to BI.

1: As someone else asked, is "-exThreads=7" supposed to be the premium option for quad-core computers?

2: Is it necessary for quad-core users to comparison test with "exThreads=5"?

3: Without "exThreads=" in the command line, what default value is it using?

1: Supposedly, yes. By default the game now uses exThreads=7 with a quad-core processor.

2: You can always try it, but in my experience exThreads=7 was better.

3: Quote from Dwarden in another thread:

0 at <3 cores

7 at 3> cores

i think on 3 core AMDs it also uses 0 as default

Also, the game now sets cpuCount=4 to quad-core processors by default as well.

---------- Post added at 06:45 PM ---------- Previous post was at 06:19 PM ----------

I did some flying around over Chernogorsk in this beta, and though there seems to be more stutter when you fly at a very low altitude than past betas, I did not get any "Receiving..." screen or "Out of Memory" errors at my usual settings, nor have I gotten any crashes so far in my basic benchmarking with this beta. Granted, I only flew around for 10 minutes or so, but it usually happens by then.

One thing I did notice is that when I restarted the mission (just a one plane flying mission over Chernogorsk in the editor) the "Receiving..." screen was up for much longer than before. So I guess maybe more of the loading is done pre-mission rather than having the screen pop up in the middle of the mission now. But hey, that works for me. Waiting a bit longer to load is much improved over flying a plane and having the screen black out, causing me to crash into the ground.

EDIT: I forgot there was something else I wanted to mention. It seems like, in general, what is causing the most stutter for me is when trees/objects change to the highest LOD while in a fast-moving plane. It seems to be speed and altitude dependent, being that the lower you are and the faster you're going seems to produce more stutter (because of all the high LODs trying to load very quickly). I would almost say that, like with the grass, BIS should look into making the highest LOD models/textures not appear while in a fast-moving aircraft, even when fairly low to the ground. You're going too fast to really tell the difference anyway, and I think it would improve performance a lot.

Edited June 8, 2010 by MavericK96

[kabong 1]

Further to Richey79's post here I can confirm that the latest version of Okt_Noblur (for beta 71141) still works with this beta (checked with HashCheck after extracting bin.pbo).

Now BIS, about adding an option in video settings to do what oktane's mod does... ;)

-------

Update: Just after posting this I found that oktane has created a new version of his mod, marked for this beta. Thanks oktane!

p.s. Ya, this post is chock full of shameless plugs for this wonderful mod, so sue me. :D

Edited June 8, 2010 by KaBoNG

[NoRailgunner 0]

I got overall better fps with this build without using parameters.

Cross fingers that BIS will optimize & improve AI too and bring back the direct control over them. For example its imho better to have an option to override AI behaviour + weapon usage - as long as they arent able to act like humans (using human cognitive skills). No need to make it annoying or complicated if we can have a choice. :)

[Tankbuster 1735]

Yes, the beta comes with an ArmA2server.exe ;)

Nope. But clients having it as well shouldn't hurt. ;)

Thanks DE. I think you know why I'm asking. :)

[Beagle 684]

I noticed that the call outs introduced in 69645

[65703] New: Units report when reloading in combat situations.

[65703] New: Units report when throwing grenades in combat situations.

are removed. Thats sad since this was one of the best ambience producing new features in this beta line.

Furthet testing showed another fps gained with removal of -exThreads parameter for me.

fps in Benchmark 1 rised from 22 with 1.05 to 28 with 71275

Edited June 8, 2010 by Beagle

[Big Dawg KS 6]

Cross fingers that BIS will optimize & improve AI too and bring back the direct control over them. For example its imho better to have an option to override AI behaviour + weapon usage - as long as they arent able to act like humans (using human cognitive skills). No need to make it annoying or complicated if we can have a choice. :)

Are you asking BIS to introduce cheats? :j:

[maddogx 13]

Are you asking BIS to introduce cheats? :j:

How is having more direct control over the AI cheating?

[Big Dawg KS 6]

How is having more direct control over the AI cheating?

Gives you an advantage in combat that BIS doesn't want you to have. If I recall, they were very adamant about making the AI more human... and that means relinquishing some control over them.

Unless he was talking about through editing. In which case the more flexability the better.

[NoRailgunner 0]

I am talking about more control over AI eg

> you can simply + directly order the AI to use the Laserdesignator or AT/AA launcher or pistol or other weapons

> you order to switch behaviour/combat stance does have an effect

> player should be + stay responsible for his orders/team and AI should follow them

As I said - as long as the AI isnt able to use cognition and common sense it would be better to have a "AI override" option.

Big Dawg KS - this isnt cheating or gives you any more advantage in combat - you can give even more and different order in mp to your human teammate. AIs aren't that good like humans. ;)

[Big Dawg KS 6]

order to switch behaviour/combat stance does have an effect

Why do you think that it doesn't? It certainly does, I guarantee you.

Big Dawg KS - this isnt cheating or gives you any more advantage in combat - you can give even more and different order in mp to your human teammate. AIs aren't that good like humans.

But if you gave an order to your teamate player that was difficult/impossible/stupid he wouldn't do it, at least not as you ordered him to. Why should AI act like mindless robots? Remember, BIS has always had a strong position on AI having the same "rights" and abilities as human players... so that would include the ability to deny following orders (when appropriate of course).

[NoRailgunner 0]

AI behaviour/combat stance

The order doesnt work like it should - for example:

if you want the AI group move quickly from place A to place B or

if you want them to get in a vehicle in the fastest way

Now the player as team leader can order "safe" - "aware" etc but the AI will stay in danger loop until #2 is saying "Its safe".

about "mindless robots"

AI developments are great and good but they still have a long way until they will act like humans. There is no need to reduce or dumb down controls over AI - only for the sake of having somekind of an faked "human touch".

Imho the overriding AI option should work with existing AI mode - as additional control to get over some annoying AI issues.

[MadMike-Brig2010- 10]

Yep! No more CTDs with weapon rest. :)

Not for me, ArmA still crashes with ACE2 when i want to deploy the bipod. :confused:

[ast65 10]

Please BIS, get rid of those really annoying antialliasing- artifacts, had them in every beta since 63826; in nightvision- mode they´re worst:(

[Big Dawg KS 6]

Please BIS, get rid of those really annoying antialliasing- artifacts, had them in every beta since 63826; in nightvision- mode they´re worst:(

Did you see Suma's post?

Probably never for ArmA 2. Fixing it would require DX 10.1, there is too little control over anti-aliased surface sampling in DX 9 (and this is also the reason why you can see the same artefact in many games).

Maybe for the next iteration of RV.

[ast65 10]

Thanks for info, Big Dawg...

[stick_hogue 10]

I didn't get a chance to post this before discovering beta 71275 is already up. I was running beta 70951 with a mission/map that runs for days (there's so much to do on it). I noticed that nighttime conditions have changed a lot since the last time I played nighttime conditions. This map generates random weather, but the nighttime implementation of the engine seems to have overriden the fog parameter. Visibility is something like 180m at night, even on a crystal clear moonlit night. After daybreak there's a persistent fog - the fog weather parameter seems to be overriden by it. And the visibility is so low that it appears as if this fog is what is causing the ridiculously low nighttime draw distance. So: at night there's 180m draw distance, after daybreak there's a fog that makes an equivalent draw distance, possibly from that same fog which is overriding weather fog parameter. You can't see squat, but apparently the AIs can see a little farther than you.

Maybe this only happens as time progresses into nighttime and then daytime, as opposed to when you initialize mission time to night. I know a lot of maps don't run as long as mine so maybe nobody has experienced this happening as time progresses. For me this ruins the game. Wanted to register it here just in case. I'll go look for comments about this in recent beta threads.

Edited June 9, 2010 by Stick_Hogue

[langgis08 10]

about "mindless robots"

AI developments are great and good but they still have a long way until they will act like humans. There is no need to reduce or dumb down controls over AI - only for the sake of having somekind of an faked "human touch".

Imho the overriding AI option should work with existing AI mode - as additional control to get over some annoying AI issues.

This could become an almost philosophical discussion ;) I like that.

I dunno how many of us have ever been in a real war as soldier (hopefully not so many, of course) ... luckily I've never been to war, all of my real experience as soldier comes by serving 12months in the German Bundeswehr, early nineties when I was around 20years old (damn, am I old but so what ;) ) I remember when we went for a simulated infantry combat "blue country" vs "green country" - therefore we had laserpointers mounted on our G3's and everyone was wearing sensors on chest, back and on the helmet ...

I remember, too - and that's my bridge to topic - that somehow everyone of us was kind of "alone on his own", the group wasn't a group anymore after 10 seconds of shooting our blank cartridges, some simulated grenades had exploded and there was a little (in fact just a very little of course) feeling of "how it could be" --> I guess a soldier will follow his orders (as a "robot") until a certain point when combat becomes too intense and everything turns diffuse and bizarre, and then it's time for his own survival instinct to take command ...

So:

Why should AI act like mindless robots?

You're right Big Dawg, they shouldn't. I like the way AI acts atm because imo it's closer to reality than 100% obeying orders in combat situations (which can be suicide, btw) -- that's why it's ok that for example the team looks/turns 3 times around before it gets "into the car", before it moves "to that tree" and so on ... (I don't wanna have a button "get your ass out of here at any costs", because not everyone would do in real, I guess)

I got overall better fps with this build without using parameters.

Yep, me too. Especially the frame drops become more and more rare :)

Edited June 8, 2010 by langgis08

[Tankbuster 1735]

I didn't get a chance to post this before discovering beta 71275 is already up. I was running beta 70951 with a mission/map that runs for days (there's so much to do on it). I noticed that nighttime conditions have changed a lot since the last time I played nighttime conditions. This map generates random weather, but the nighttime implementation of the engine seems to have overriden the fog parameter. Visibility is something like 180m at night, even on a crystal clear moonlit night. After daybreak there's a persistent fog - the fog weather parameter seems to be overriden by it. And the visibility is so low that it appears as if this fog is what is causing the ridiculously low nighttime draw distance. So: at night there's 180m draw distance, after daybreak there's a fog that makes an equivalent draw distance, possibly from that same fog which is overriding weather fog parameter. You can't see squat, but apparently the AIs can see a little farther than you.

This has been happening for ages. It precedes betas. If you let a persistent server run for a couple of days, you get this happening. A ring of fog closes in restricting VD.

See my post here.

http://forums.bistudio.com/showthread.php?t=95914&highlight=fog

[MavericK96 0]

After playing this build a bit longer (online MP in ACE, mind you) it seems to have considerably lower framerate, or at least more stutter than previous builds, but I haven't gotten a single crash or "Receiving..." screen the entire time I was flying around in an A-10, so that's a plus. One weird thing was when I was going in for a landing, I blew up for no apparent reason (could have been AA though, who knows) and then suddenly the game sort of locked for a little while, and when I respawned all the textures were gone, replaced by blocky, single-color ones and the game was going at about 5 FPS. I couldn't hit ESC to get out of the game and I had to close it directly from the task manager. Never had that issue before, but overall stability has improved greatly, though perhaps at the cost of framerate/stutter.

[no use for a name 0]

Works great for me

[stick_hogue 10]

This has been happening for ages. It precedes betas. If you let a persistent server run for a couple of days, you get this happening. A ring of fog closes in restricting VD.

See my post here.

http://forums.bistudio.com/showthread.php?t=95914&highlight=fog

Aw, damn. Thanks for the info.

I posted on that thread to confirm it does happen on hosted server (that's my situation) with a repeating weather script running. (fog seems to be ignored, but I have experienced changes in rain after the VD and fog set in)

stick

[jpinard 10]

It's taken me a while, but I've finally completed a comprehensive testing & analysis of the beta upgrade path. 50+ hours and includes every Benchmark available including two of my own. Links and demonstrations further below.

All tests were completed in duplicate or triplicate.

Since there's such a major limit on picture size here in the forums, the only screenshot of the data I can provide is blurry & reduced, but is just to give you an idea of what's included:

Here is the full spreadsheet in standard Excel format:

http://www.jeffpinard.com/ArmA2_Benchmarks_1.xls

The Excel spreadsheet data is divided into 4 parts.

* Lines 1-16 are the individual tests - distance @3500m

* Lines 20-25 are the individual tests - distance @6000m

* Lines 29-35 are the averages & this is the data that counts - distance @3500m

* Lines 37-39 are the averages for the 6000m distance. once again, this is the data you want.

* I outlined the most important data points in red.

* All tests performed with all graphic settings maxed out (FSAA & AF both set to Very high).

* Screen resolution set to 1920x1200.

* View distance set to 3500m, latter tests include 6000m as well.

* Responsiveness tested running around a massive 300 person battle defense & assault.

Versions Tested:

* v1.05

* Beta v63826 - Old video card

* Beta v63826 - current HW

* Beta v70313

* Beta v70313 -exThreads=3

* Beta v70951

* Beta v71275 -exThreads=7

* v1.05 (distance=6000)

* Beta v70951 (distance=6000)

* Beta v71275 -exThreads=7 (distance=6000)

Benchmarks & abbreviations used:

* Scenario Benchmark 1 (Bench 1)

* Scenario Benchmark 2 (Bench 2)

* ArmAllMark 1.0 (Test One, Two, Three, Four, Five, OFPMark)

* Chernogorsk City FPS 6 (C6 ave, min, Highest, Lowest)

* jpinard Helo Test (H frames, H Time, H Min, H Max, H Avg)

* jpinard Jet Test (J Frames, J Time, J Min, J Max, J Avg)

To see how my Helo bench test works click here:

To see how my Jet test works click here:

To download both tests click here:

http://www.jeffpinard.com/jpinard%20ArmA2%20Benchmark%20Tests.zip

I created the extra tests because it didn't feel like any of the synthetic tests pushed every aspect of the game. Please read the Youtube description for more details.

In summary, each beta I tested has generally improved performance most notably when looking at 6000 meter distance. Supersonic jet performance is amazing in this build. The current beta has the greatest increase in speed but there are 2 major problems:

1. I've experienced Game hangs I've never seen before while the missions are loading in ArmaMark and Chernogorsk City (C6_FPS_6). The lock-up rate was quite high at 33%.

2. Sound is cutting out worse in this beta than in any other build I've tested. Beta v70951 was by far the best at managing massive amount of sound properly.

System Specs:

* Intel Core i7 920 2.66GHz overclocked to 4.3 GHz stable

* Asus P6X58D Premium Motherboard BIOS v.0703

* G.Skill DDR3 2000 in Triple Channel Mode - 6 Gig

* NVidia GeForce GTX 480 Driver v197.75

* PCIe Titanium Sound Blaster X-Fi Titanium Fatality

* Planar PX2611W 26" widescreen LCD

* Win7/Boot drive on 80 Gig Intel X25M SSD

* ArmA2 installed by itself on second 80 Gig Intel X25M SSD

Edited June 9, 2010 by jpinard

[MavericK96 0]

jpinard, your link doesn't work but WOW. Amazing work. Thank you for this. :)

EDIT: Spreadsheet link, that is. This one works: http://www.jeffpinard.com/ArmA2%20Benchmarks-1.xls

Edited June 9, 2010 by MavericK96

[jpinard 10]

Link fixed: http://www.jeffpinard.com/ArmA2_Benchmarks_1.xls

Few more notes:

* Reason there are x's for several categories is because Armamark sets its own view distances, so re-testing with 6000m range is redundant, and I hadn't designed the jet test yet for older betas.

* There are observational notes in the far right columns highlighted in orange.

* No mods were loaded or tested.

* Link for those without Excel (image): http://www.jeffpinard.com/Arma2%20Benchmarks.jpg

Thanks Maverick. Hopefully BI and others can use data. I'll continue to update the spreadsheet as beta patches come out if I remain healthy. :)

Edited June 9, 2010 by jpinard

[MavericK96 0]

What I'm most impressed by is your personal benchmarks, because like you said, the _FPS_6 bench does not show a lot of variance between betas, where your benchmarks definitely do.