Jump to content
Placebo

Will-my-pc-run-Arma3? What cpu/gpu to get? What settings? What system specifications?

Recommended Posts

7 FPS more in YAAB @ 1440p ultra with my config using AVX2 dll.

Share this post


Link to post
Share on other sites
16 hours ago, ruPal said:

I just don't know how to run Arma3 with admin privileges from arma launcher.

Go to arma3.exe and force run as admin in compatibility settings.

 

Also when testing this and enabling lock pages privilege and running as admin. Keep in mind that Arma now has Large Pages support built-in already with a startparameter.

Which it didn't have in 2016 when that allocator was compiled.

 

If large pages is making the big difference, you can already get it with the normal allocator.

 

The allocators that BI ships are not built with AVX2, I don't even think they have SSE2 enabled, for compatibility reasons with older CPUs.

I also built own allocators for testing once.

 

Ah I found them: https://drive.google.com/drive/folders/1HomJlRkK9-eWRexqcqMMwtUsTjgr9BAq

Afaik only AVX and not AVX2 because my own CPU doesn't even support AVX2 😄

Here are some recent allocator benchmarks: https://www.youtube.com/watch?v=kNmeMxEel4M

 

When testing my allocators half a year ago we saw 46fps->49fps increase.

 

1 hour ago, Groove_C said:

7 FPS more in YAAB

That alone doesn't say much. 7fps more at over 60fps, are barely 2fps increase on 20fps base. Always need to mention the base, or the range.

 

Share this post


Link to post
Share on other sites
29 minutes ago, Dedmen said:

Also when testing this and enabling lock pages privilege and running as admin. Keep in mind that Arma now has Large Pages support built-in already with a startparameter.

Which it didn't have in 2016 when that allocator was compiled. 

 

If large pages is making the big difference, you can already get it with the normal allocator.

Yeap, I knew that. Tested with BI huge pages and CMA huge pages AVX2. The first one gave me 60FPS and last one 69FPS. Also tested other CMA (default and AVX) versions. But the most performance boost was with AVX2 version (default - the same as BI, AVX - 64FPS).
Thank you for admin privileges clarification, will try.

Btw, @Dedmen, did you try SuperMalloc https://github.com/kuszmaul/SuperMalloc?

Share this post


Link to post
Share on other sites
35 minutes ago, Dedmen said:

Keep in mind that Arma now has Large Pages support built-in already with a startparameter.

If large pages is making the big difference, you can already get it with the normal allocator.

I have large pages checked with Intel default one.

If I only change from Intel's default to cma_x64 and keep or recheck large pages in the launcher -> same FPS.

Selecting to launch arma3_x64.exe as admin, enabling large pages support via "secpol.msc" and having "cma.ini" in Arma 3 folder, @ 1440p ultra FPS increases from 44 to 50-51.

A friend, using the AVX2 dll with an i5-4570 3.4 GHz all cores + 1x8 GB RAM DDR3 1600 MHz CL11 + GTX 1060 6 GB @ 1080p ultra now has 28-29 FPS instead of 24.

Share this post


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

Selecting to launch arma3_x64.exe as admin, enabling large pages support via "secpol.msc"

Armas own hugePages support might need all of that too.

Share this post


Link to post
Share on other sites

@Dedmen

It's not because of large pages, because large pages already enabled. It's because of AVX2 actually.

@ruPal already told, that testing the default cma_x64 dll with large pages enabled via "secpol.msc" and as admin gave him same FPS as with BI Intel default.

With AVX 4 FPS more and with AVX2 9 FPS more.

What is difficult to understand here?

Share this post


Link to post
Share on other sites
1 minute ago, Groove_C said:

What is difficult to understand here?

Just trying to make sure that the effect is really only AVX2, and not AVX2 combined with hugePages.

Share this post


Link to post
Share on other sites
14 minutes ago, Dedmen said:

Armas own hugePages support might need all of that too.

It is a pity that BI just through that flag in without any instructions about that. I used fred41's malloc and knew that large and huge pages are need more work to do to work because of that. Also it is pity that you are unable to check if hugepages enabled, like in fred41's malloc and _blub malloc, where you may check that in logs.

Share this post


Link to post
Share on other sites

I have tested a lot of mallocs through the years, starting with A2.

And huge pages gave me "better" FPS only within the margin of error. But despite "same" FPS, it really helped on Altis, in the beginning, when it was lagging/stuttering and textures were loading not fast enough without huge pages and fred41's malloc solved this issue for me back then.

But AVX and especially AVX2 really boost FPS.

 

If not all CPUs support AVX and AVX2, why not include those mallocs in the list of the launcher with Intel's by default, that supports not higher than SSE, like now?

Several mallocs are already in the list as not default. Lets add AVX and AVX2 to the list of available ones.

Share this post


Link to post
Share on other sites
1 minute ago, Groove_C said:

And huge pages gave me "better" FPS (within margin of error).

In my situation, fred41's malloc gave me 2-3 FPS on FX8350 system. When I had 30-40FPS it was a good boost. What about AVX and AVX2, more tests needed. I run only 1 YAAB test per default and AVX mallocs without PC restarts in between.

 

6 minutes ago, Groove_C said:

But AVX and especially AVX2 really boost FPS.

More tests needed. Too early to give such conlcusions. Just run different mallocs yourself and test. To get proper results you will have to reboot PC when malloc switch because of large pages specific.

Share this post


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

on FX8350 system

FX8350 doesn't support AVX2 sadly. I'm also on that CPU. That's why my allocators are only SSE2/AVX1

Share this post


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

I've tested with 7 runs per AVX and per AVX2.

Personally I don't trust your results. If you do so, you won't refer to my result in your post and write 7 minutes later that you have tested it by yourself 7 times.
I have made second test run AVX vs AVX2 (didn't test default CMA vs Default CMA), but won't post results here. Don't want people to trust my results and want them to run test by themselves. Actually, I got contraversary results.

Share this post


Link to post
Share on other sites

@Dedmen

My findings:

No FPS difference between:

Intel malloc (no admin + no huge pages launcher) vs. Intel malloc (huge pages launcher)

Intel malloc (no admin + no huge pages launcher) vs. Intel malloc (admin + huge pages launcher)

Intel malloc (no admin + no huge pages launcher) vs. Intel malloc (admin + huge pages windows + huge pages launcher)

 

1-2 FPS difference between Intel and cma_x64 AVX2 (admin + no huge pages windows).

6-7 FPS difference between Intel and cma_x64 AVX2 (admin + huge pages windows).

  • Thanks 1

Share this post


Link to post
Share on other sites

So basically it's not just about huge pages by itself, but the combination of huge pages with AVX2 maloc dll.

Share this post


Link to post
Share on other sites


You must always remember that even if YAAB is the best tool at hand, results in a run can be affected by many factors.

The main variation factor being the IA behavior.

So you can get up to a 5 FPS variation range in 10 tests made in a row.

 

And, as always, you must be cautious not to select results according to your hopes 🙄

 

Share this post


Link to post
Share on other sites

I know what I'm doing. I run YAAB as many times as needed. If 3 results in a row are close enough, I make the average of them. If not, I let run YAAB more and then just drop results that are to fare ahead and/or behind from the rest of already available results. I can run YAAB 5-7-9 times to have the average and remove "suspicious" runs.

Share this post


Link to post
Share on other sites

After triple run with system reboot in between, I got:

CMA default: 66,9  (65,1; 67,1; 68,6)
CMA AVX: 66,3 (67,4; 64,8; 66,7)

CMA AVX2: 67,2 (70; 66,2; 65,6)

CMA vs CMA AVX vs CMA AVX2 delta is less than 1%. As we see the most FPS boost gives custom malloc with large pages support itself, not AVX or AVX2 instructions.

Personally, I choose AVX2 because it gave me absolute maximum - 70 FPS. Not sure but maybe CMA default and AVX may provide the same results in more runs. But still AVX2 gave me 69 and 70 FPS, CMA default and AVX gave me 68,6 max.

 

P. S. I think (didn't test), that there is no large or huge pages support in default arma 3 tbbmalloc, -hugepages parameter do nothing. So if you want play with large pages you have to run custom malloc. If somebody can check and test that, please do.

 

Share this post


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

P. S. I think (didn't test), that there is no large or huge pages support in default arma 3 tbbmalloc, -hugepages parameter do nothing. So if you want play with large pages you have to run custom malloc. If somebody can check and test that, please do.

+++

 

Definitely needs to be checked by somebody!

Share this post


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

I think (didn't test), that there is no large or huge pages support in default arma 3 tbbmalloc

There should be. But might be broken, I tested it back when I made my allocators and don't remember the exact results.

It certainly exports a method that says it supports them, whether it's correctly implemented dunno.

Share this post


Link to post
Share on other sites

-hugepages do nothing for me. No influence on performance with tbbmalloc or cma.

Share this post


Link to post
Share on other sites

r7 3700x

32 gb ddr4 3000

gtx 970

mx500

 

will that run as good as SINTEL 9900 in KOTH(64 players)?

 

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

×