Jump to content
Greenfist

Yet Another Arma Benchmark

Recommended Posts

A comment about the comment you posted previously.

There is no much sense to speak about GPU limitations with Arma3 with a RX 470 4Gb.

You must understand that some of the so called "Video" parameters are NOT related  with the GPU !

It's the case for the General>Visibility>Overall parameter, this one is 100% CPU .

This parameter defining the radius to be rendered in game is setting the CPU load.

It's also switching some shadows rendering from the GPU to the CPU.

 

So setting the Arma3 Preset to "Low" is ...

- lowering the graphics quality to low,

- lowering the visibility to 1000m,

- lowering the CPU overall load but adding some more render load.

With a Ryzen 1700/ RX 470 4GB combo, it can tell the CPU is good but it's telling nothing about Arma3 CPU hierarchy.

 

So in order to get a good comparison base, it's important to run YAAB with "Standard" YAAB settings.

Share this post


Link to post
Share on other sites
On 19.10.2017 at 8:32 PM, oldbear said:

So setting the Arma3 Preset to "Low" is ...

- lowering the graphics quality to low,

- lowering the visibility to 1000m,

- lowering the CPU overall load but adding some more render load.

With a Ryzen 1700/ RX 470 4GB combo, it can tell the CPU is good but it's telling nothing about Arma3 CPU hierarchy.

 

So in order to get a good comparison base, it's important to run YAAB with "Standard" YAAB settings.

 

@oldbear I intended not to compare our Graphics backend so I set the Video Presets to "Low". I do not see why that should be making CPU-CPU comparisons difficult as long as you and me do the same.

 

However, I can report that a YAAB run under "Standard" preset plus pressing "S" for standard settings results in 51.5 FPS in YAAB on my 1700 @ 3.8GHz + 470 4gb.

 

Here are all results including RAM scaling on my sys.

 

Low means 'Low Preset'

Standard means 'Standard Preset + Standard Settings'

 

Hope this is comprehensive enough.

 

3eT44N2Bj564.png

Share this post


Link to post
Share on other sites

OK, you are on the right path!

Now, use "autodetection", so your rig is set to the best options Arma3 engine will support [we will speak about custom parameters afterward] then run YAAB and press "S" so for this test in order to get results based upon common community settings, run it 3 times and go on posting ...

 

  • Confused 1

Share this post


Link to post
Share on other sites
11 hours ago, oldbear said:

OK, you are on the right path!

Now, use "autodetection", so your rig is set to the best options Arma3 engine will support [we will speak about custom parameters afterward] then run YAAB and press "S" so for this test in order to get results based upon common community settings, run it 3 times and go on posting ...

 

 

Briefly, ... to better understand... you want me to use 'autodetect' which leads to custom, basically all settings ultra, incl. 8x FSAA and then post average FPS so we can conclude from that on CPU performance? What's the rationale? Could you briefly elaborate on that?

 

I used two reproducible settings that are most likely not GPU bottlenecked and tested them in three different conditions. What is missing here? 

 

Anyway, a brief run says 47.7 with autodetect + s but I consider that pretty meaningless as we are now substantially benchmarking the GPU.

 

I am off for a conference now for a week...

Share this post


Link to post
Share on other sites

Basics
Unlike most of FPS style games, Arma3 is giving an absolute priority to CPU simulation jobs.

This is related to the basic building principles of the Real Virtuality engine.

With Arma3, in game performances depend directly on CPU efficiency and operating frequency*.
The role of the graphics card** has never been completely secondary, it has gained weight since the Visual Upgrade but, but the role of the processor remains totally dominant.


Consequences
Some of the so called "Video" parameters are NOT GPU related.
The General>Visibility>Overall parameter, is 100% CPU .
This parameter is setting up the CPU load.

All the "video" parameters included in the YAAB standard settings are more or less CPU related.


Using YAAB as a reference
One of our main concern as gamers is to know if Arma3 is going to be playable with the rig we have or the rig we intend to get.
As a community concerned Arma veteran player, I believe it's interesting to enlarge our knowledge about Arma3 real performances in game.
From my point of view, YAAB is a rather good candidate because this benchmark-mission is built as a real playable mission with all the assets (buildings, AIs, FX ...) contributing to load the CPU.
Playing YAAB with the ["S"-key) standard is important for comparison. It's setting a common reference.

Of course, on the higher Arma3 settings some AA & PP assets are fully related to GPU, that why it's important to quote those custom parameters when you post a YAAB result.

Share this post


Link to post
Share on other sites

Thanks for clarifying @oldbear.

 

Well, if it helps somebody. Here are the results and settings after autodetect + 'S' key (two runs). Note that I have disabled FSAA for the second run, as this surely only impacts GPU performance. Everything else is as normal. Radeon RX470 @ stock and Ryzen 1700 @ 3800MHz; RAM@ 3200MHz with tuned timings as before.

 

3fwYYWfLh7DD.png

Share this post


Link to post
Share on other sites

Hi Guys. I just got my 8700k built with 4266mhz memory, the objective was to get the most out of it in arma.

 

For comparison I was getting 43fps on my 4770k @ 4.5ghz with a 1080 TI founders edition overclocked.

 

With the 8700k @ 5.1ghz and memmory running at 3733mhz cas 15( could not get my XMP profile to work ) this is what Im getting running the benchmark with staInsert image from URLndard settings + pressing S.

 

https://imgur.com/a/zn6Op

Share this post


Link to post
Share on other sites

Results interesting but not completely usable.
To have a good preview of what the overclocked i7 8700K can give in game, it should first make a simple adjustment using the auto-detection function built in  game then launch YAAB and  then activate the standard settings by pressing the S key.

So we can get a glimpse at what can really be expected in game, because with such a rig you are not going to play with "Standard" parameters and ugly graphics ...

We know that the post processing section can pull hard on CPU performances.

Share this post


Link to post
Share on other sites

I will try to show the impact of AA&PP parameters on performances

Here are the result of my last YAAB runs featuring the brand new Iiyama G-Master 1080p 144 Hz monitor I have receive as birthday present  :don16:

New monitor used on the 2 last runs :

 

XZ8JEa9h.jpg

 

On the first run my custom AA&PP were in use as suggested by the "Bear's method" [1]

 

oJMzK7Yl.jpg

 

On the last run [*] Ultra vanilla parameters as set by auto-detection were used.

 

Om8AfXql.jpg

 

... you can see the cost of FSAA 8x !

 

[1] the "Bear's method" has started as a kind of a joke in my Clan.

I have publish on our forums some findings about how to tweak the game parameters.

It has grown as a kind of private FAQ. I have then published as a method on some French forums.

I hope it has helped some players.

 

Some explanations :

- bloom, blur and depth of field are disable or reduced because in my opinion that those parameters are a kind of sabotage.

Do you think I have paid for a nice bunch of hardware, paid a lot for a nice pair of spectacles and more recently to undergo cataract operations just to get blurry screens? Really, are you serious?

Maintaining a low level of rotation blur is one of the concessions made to have a better perception of the played character in game.

- FSAA 4x & CMAA usage are related to some hints published a log time ago on HardOCP site and on French CanardPC Hardware n°34 newspaper last month.

 

  • Thanks 1

Share this post


Link to post
Share on other sites

@Greenfist Mighty fine test this is.

 

It helped me realise something i never realised before. How much ultra textures effect not just my FPS,but the smoothness.

 

I swear it was like playing this benchmark on and old rig compared to a new one once i went from ultra down to very high textures.

 

i74790  / 3.6Ghz 

1060 GeForce 6GB OC

8GB ram 

non SSD

 

mods: JSRS (Only notable) 

 

Not alot of Post processing :  4x AA / full antiscopic filtering / shadows high.

 

1920 X 1080P

 

ULTRA TEXTURES : 30.1 FPS AVG

VERY HIGH TEXTURES : 35.3 FPS AVG

 

Bravo. From my own tests i was never able to notice ultra holding frame rate much lower,particularly in a steady scene.Only ever noticed it with some highly detailed modded units(RHS) ,Now playing vanilla.

Thanks for this nice benchmark

  • Like 1

Share this post


Link to post
Share on other sites

@Greenfist Thanks for this last YAAB 1.00 version. 😎

 

Still doing tests on a Ryzen 3 2200G, now in 1080p :

 

KVVHspzh.jpg

 

I am wondering if the modified IA behavior since Jets and Tanks DLC release has an effect on results.

I am wondering if something is still wrong with the Wipeout spawn in the end.

 

Note :

1° RAM is a Corsair Vengeance LPX 16Go (2x8Go) DDR4 3000MHz C15 XMP 2.0 kit.

2° EZ is the ASUS OC for the noobs mode pushing CPU frequency from 1600 / 3700 MHz to 3850 MHz on all cores.

3° "Standard" in the test is NOT related to Standard mission test parameters,  it's the  "Standard" preset as given by game "Autodetection".

 

Edited by oldbear
"Esprit d'escalier", be slow-witted in english don't sound so good ...

Share this post


Link to post
Share on other sites

@oldbear

I wonder if difficulty settings affect the load that AI puts on the CPU and if so, then we should maybe agree to use recruit or regular difgiculty for our YAAB tests.

Otherwise, the one who tests at recruit, will get better FPS than the one who tests at regular or even veteran.

But I don't know if there is this difference or not between different AI difficulty settings.

Share this post


Link to post
Share on other sites

I would be surprised if skill didn't have an effect. Just the rate of fire alone on higher levels should create a lot more ballistic calculations to process. Although I doubt you'd be able to measure the difference with this test, the AI is just too random to begin with to see such (presumably) small change.

Should be fairly easy to check though, just by testing.

 

 

Share this post


Link to post
Share on other sites

I had run YAAB very often since it was released and still playing it now every day for tests.

Since Jets and Tanks DLC release, I had witnessed modified IA behavior.

I will give you 2 examples of obvious modifications :

- in the aerial sequence, previously, the choppers flying straight on were engaging the trucks as a priority, often disabling and sometimes destroying them.

Now the chopper are only engaging armor without much effect and are flying an evasive route.

- in the Varsuk rush sequence, previously the tank after firing go crash itself on the wreck, then was shot once or twice in the back by AT.

Now,  the tank run the curb cautiously killing them all before going to a stop.

So my question is only about the impact of the modified IA behavior since Jets and Tanks DLC release has an effect on results because of FX modifications.

 

 

Share this post


Link to post
Share on other sites
On 4/18/2019 at 11:57 AM, Greenfist said:

 

 

Greenfist... would it be possible for your amazing Yet Another Arma Benchmark to have a special tab or section that measures all the mission's running scripts individually and list them based on different parameters like memory usage, cpu impact and network bandwidth usage?

 

So perhaps we can figure out what script/mod is causing the most performance drop in terms of memory, cpu and also network bandwidth.

 

That would be just a dream come true for mission and even mod makers.

 

Thanks for all your hard work on this must have tool of yours!

Share this post


Link to post
Share on other sites

@LSValmont STOP, it's a "mistaken identity" case !

 

YAAB  is  Greenfist creation and one of his contributions to help to understand the game's performances.

I am only a user of the tool he built and maintain.
YAAB is important because it takes the place of a necessary tool that has not been published by BI.
That's why I support it, use it and try to promote its use.

 

Sorry, you must ask Greenfist 😥

 

 

  • Haha 1

Share this post


Link to post
Share on other sites

@Horus The dropbox link on the first post is now updated. It's been so long since I last touched this mission I didn't remember I even had DB download. 🙂

 

@oldbear You're right the test definitely runs different after some game updates, because most of AI actions aren't strictly forced by the mission. Instead it's mostly just basic waypoints and fire commands, so the outcome is sensitive to any updates to AI behaviour. Their effect on FPS is anyone's guess I suppose. There aren't any good tools to measure where the processing power goes on that deep level.

Although I don't know if something like the Arma script Profiler or the vanilla frame diagnosis tool could provide more insight on the matter. These two might also be what you're looking for too,  @LSValmont

 

 

  • Like 1
  • Thanks 1

Share this post


Link to post
Share on other sites
On 4/21/2019 at 7:31 AM, Greenfist said:

Although I don't know if something like the Arma script Profiler or the vanilla frame diagnosis tool could provide more insight on the matter.

They definitely do. They show quite precisely which part of the frame is spent rendering, ai calculating, script execution and so on.
My profiler can also record the entire YAAB run for later analysis. and give you averages/histograms of all the steps.

  • Like 1

Share this post


Link to post
Share on other sites

Here is a screen shot of a rather rare event in YAAB.

Helicopter came back and attack transport truck.

So far has happened 3 times over around 200 + YAAB runs since Tanks and Jets DLC release.

 

8WX6Rdxh.jpg

 

Note the very high GPU usage -99,4%- in this [R3 2200G+EZ  RX550 16Go 3000MHz  1080p] run using Test Standard settings, Arma3 High Quality and custom AA&PP parameters along the "Bear method".

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now

×