RAM Management - Serious Question

[I give up 152]

Give the mission for repro.

System specs in both situations would still be nice to know... We now know that you've 32GB of RAM and in that video you've 5870 but what other AMD card you've in the situation where things aren't good? And you likely have page file disabled or is it enabled? You could edit that stuff also in the first post.

Did you change any graphic settings or launch parameters?

This could be AMD only problem.

 

You also could post the other video where the performance was bad in that situation. You had a topic about it or something. In this topic you've shown RAM load after flying all over the map but you didn't show that now with your video. You could help a lot but you're too cryptic.

 

Some basic stuff needs to be answered first before the issue can be looked. Dwarden would likely happily forward these stuff but he very likely needs to get the same questions answered if not more.

Yes, I have 32 GB of RAM, 

Yes, I have pagefile disabled.

Startup parameters? No I dont have any related with hardware. Besides the only one that  is working is maxmem which is only related with arma3.exe process memory usage but the only changes with it are when you set it bellow 2047 MB, but is like useless because if you set it, let's say, to 1500 MB the game starts struggling, if you set it, let's say, to 1000 MB the game becomes unplayable, but again maxmem parameter is ONLY related with arma3.exe, which is not the subject here and also I dont have issues with it, arma3.exe, memory usage with it rarely exceed 2000 MB, in my end.

I have no doubts (based in my experience with both) that the game performs better with Nvidia, still is not the case here, the RAM usage and management are not related with AMD or Nvidia.

 

Now keep in mind that the subject here are not the FPS, is the RAM usage and management. Still, having above 50 FPS is Kavala with a 10 year old GPU with 1 GB of VRAM is something amazing.

But the most amazing is that do not give any issue with RAM and everything looks much more smooth. RAM load remains around 4 GB and the flush/refressh timings are being performed quite well, all this having less than 800 MB of VRAM in use.

Also, I am recording this video with Lagarith Losseless Codec, which eats aroud 7/10 FPS, when not recording I have stable 60 FPS (I have it limited to 60, there is no need for more), whatever the scenario.

 

https://www.youtube.com/watch?v=fLsr1w1fx1w

 

Are you kidding? I have found the perfect hardware for A3.

[domokun 515]

I'm just going to leave this here. --- https://www.youtube.com/watch?v=9cYNcnJvWH4

 

This game is amazing. However, it's biggest flaw has to be ingrained deep in Engine, and probably never to be touched, at least for this iteration of Arma.

You really shouldn't quote people who have no idea what they're talking about. That vblogger complains about A3's performance but he's running everything at the Ultra preset. He then tries to improve it by lowering every setting except the most important (View Distance). Next he compares A3 to Rainbow 6 Siege! Finally he draws a parallel with Squad: for sure it's a tactical shooter except that... Max map size is 2km². No AI. Fewer units. Fewer maps. Fewer weapons. No vehicles. No editor. No modding. So actually BF4 would be a better comparison, right?

[CaptainAzimuth 714]

You really shouldn't quote people who have no idea what they're talking about. That vblogger complains about A3's performance but he's running everything at the Ultra preset. He then tries to improve it by lowering every setting except the most important (View Distance). Next he compares A3 to Rainbow 6 Siege! Finally he draws a parallel with Squad: for sure it's a tactical shooter except that... Max map size is 2km². No AI. Fewer units. Fewer maps. Fewer weapons. No vehicles. No editor. No modding. So actually BF4 would be a better comparison, right?

Whoops, that was actually the wrong video. However I did watch it and he doesn't really quite understand, I agree, but he does mention view distance as being a major problem, despite not actually adjusting that in the video. I don't actually follow this particular Vlogger, but anyhow. The problems with performance as I've said before are Engine deep, to the point where it actually has me wondering if BI are really sure that the added eye candy coming with Apex are going to be worth it when people with medium end PC's like me still have issues with simple things such as firing my gun at another player and having my game freeze, and in-freeze with my character slumped over in a hail of lead.

[Defunkt 431]

As a programmer I get that it's hard to retrospectively convert the RV engine's core into something that's thread-friendly. What I don't understand is why BI don't make better use of the available processing power so we could have something more to show for our hardware. Specifically for the AI, complex path-finding and better strategic and tactical decision making could be constantly re-evaluated in parallel and the main thread just fetches the conclusions of those calculations as and when it has time ('what should this unit be doing now and which units need the greatest attention right now?'). Your FPS might not improve much but the perceived quality of the game would reflect the hardware it was running on. The headless client is a really poor substitute and far too complex to setup for the vast majority of players (like the guy who made that video).

 

Get somebody like William van der Steren (http://www.cgf-ai.com/docs/plannedassault_ai_planner.pdf) on the job for a year and work toward re-using the engine's current AI decision making timeslice to just exchange updates with a completely separate process.

[NoPOW 59]

Yes, I have 32 GB of RAM, 

Yes, I have pagefile disabled.

Startup parameters? No I dont have any related with hardware. Besides the only one that  is working is maxmem which is only related with arma3.exe process memory usage but the only changes with it are when you set it bellow 2047 MB, but is like useless because if you set it, let's say, to 1500 MB the game starts struggling, if you set it, let's say, to 1000 MB the game becomes unplayable, but again maxmem parameter is ONLY related with arma3.exe, which is not the subject here and also I dont have issues with it, arma3.exe, memory usage with it rarely exceed 2000 MB, in my end.

I have no doubts (based in my experience with both) that the game performs better with Nvidia, still is not the case here, the RAM usage and management are not related with AMD or Nvidia.

 

Now keep in mind that the subject here are not the FPS, is the RAM usage and management. Still, having above 50 FPS is Kavala with a 10 year old GPU with 1 GB of VRAM is something amazing.

But the most amazing is that do not give any issue with RAM and everything looks much more smooth. RAM load remains around 4 GB and the flush/refressh timings are being performed quite well, all this having less than 800 MB of VRAM in use.

Also, I am recording this video with Lagarith Losseless Codec, which eats aroud 7/10 FPS, when not recording I have stable 60 FPS (I have it limited to 60, there is no need for more), whatever the scenario.

 

Are you kidding? I have found the perfect hardware for A3.

 

Could you upload some vid's of Helo's benchmarks?

https://steamcommunity.com/profiles/76561197970869024/myworkshopfiles/?appid=107410

[I give up 152]

 

If I was going to guess (based on my experiences with several hardware brands/models) the reason for these issues related with memory management, I would say that we have 3 facts for the cause.

 

1 Fact,

Due to "32 bits breaking barrier" basically there are no limits for RAM load, usage and consequently management. Keep in mind that when this feature was implemented the average VRAM was around 1 GB (for the GPUs at that time).

2 Fact,

In this game (more than with any other) the graphics quality (mainly textures) have a significant impact in RAM load and usage.

Also graphics quality (mainly textures) have a significant impact in VRAM load and usage, and these two are linked. That's why with a GPU having a 1 GB of VRAM we don't have "Ultra" available for textures in game settings.

3 Fact,

More VRAM = More RAM load. Above certain limits in matters of VRAM, the RAM load is to high and the game engine cant handle with it within acceptable timings causing degradation in performance that also negatively affects GPU and CPU performance.

 

In short, "32 bit breaking barrier" works flawless below certain limits in matters of VRAM but it causes severe performance degradation above certain limits.

Like I have said, just a guess.

[jumpinghubert 48]

 

More VRAM = More RAM load. Above certain limits in matters of VRAM, the RAM load is to high and the game engine cant handle with it within acceptable timings causing degradation in performance that also negatively affects GPU and CPU performance.

 

 

To make things clear you have to compare different rigs in the same benchsituation so you need 2 runs of the same benchsituation, first run with hardware x, second run with hardware y. Its like you repeat:" the pencil x is longer than pencil y" and as "evidence" you post lots screenshots with single pencils.

 

I have in your theory the worst combination for arma3: 4gb vram (R290), texture settings at ultra and 8GB system Ram. No slowdowns at all.

 

The best for arma3 performance wise I have done was:

-switch from HDD to SSD

-changed and overclocked ram from 1600 to 2666Mhz.

-overclocking CPU to 4.8GHZ.

-using fred´s malloc.

 

All verifiable.

[I give up 152]

Could you upload some vid's of Helo's benchmarks?

https://steamcommunity.com/profiles/76561197970869024/myworkshopfiles/?appid=107410

Those benchmarks are mainly for FPS contest, still, not being FPS my purpose with this thread I have no problems in doing it, because also serves to prove my point.

Pagefile disabled, VRAM load max 800 MB, RAM load max 4800 MB Stratis average 54 FPS, Altis average 44 FPS.

Now if we look closer we see that in Stratis we have basically the GPU running at 99% and in Altis running at 50%. Does this make sense? Not at all, what would made sense was if we had the GPU running at 50% in Stratis and at 99% in Altis.

Why this happens? The answer is simple.

In Stratis we have a RAM load bellow 3.8 GB, which is perfectly normal and in this situation the game engine is able to manage it under acceptable timings, means that the GPU (and also CPU) are not being bottlenecked.

In Altis we have a RAM load of about 4.8 GB (which is pretty decent when compared with the 12 GB that I had when I was having a VRAM of 9 GB), still is enough to cause a huge GPU bottleneck (and also CPU) just because the game engine is unable to manage this amount of RAM under acceptable timings.

This will make me to say again, the real issue with this game is the memory management, the CPU and GPU bottleneck that common people usually complains about, is in reality a issue cause by memory management that cause bottlenecks in to every piece of hardware in our machine.

 

Stratis and Altis in a row.

https://www.youtube.com/watch?v=qm1kp4oSSSM

(btw, I am still using Lagarith codec, which eats 7/10 FPS)

[domokun 515]

To make things clear you have to compare different rigs in the same benchsituation so you need 2 runs of the same benchsituation, first run with hardware x, second run with hardware y. Its like you repeat:" the pencil x is longer than pencil y" and as "evidence" you post lots screenshots with single pencils.

 

I have in your theory the worst combination for arma3: 4gb vram (R290), texture settings at ultra and 8GB system Ram. No slowdowns at all.

 

The best for arma3 performance wise I have done was:

-switch from HDD to SSD

-changed and overclocked ram from 1600 to 2666Mhz.

-overclocking CPU to 4.8GHZ.

-using fred´s malloc.

 

All verifiable.

Interesting. What would be awesome is if you could provide us with some performance increases for each upgrade. It could help understand the "best bang for your buck". BTW do use use the in-built benchmarking to compare the effectiveness of an upgrade. Is it an accurate representation of real-world performance? i.e. isn't there a better tool?

[jumpinghubert 48]

@domokun

ram overclocking: https://forums.bistudio.com/topic/156993-arma-3-cpu-vs-ram-performance-comparison-1600-2133-up-to-15-fps-gain/

for cpu overclocking many benchmarks are spread over the internet and in this forum.

malloc: https://forums.bistudio.com/topic/154302-arma3-and-the-largeaddressaware-flag-memory-allocation-2gb/

 

I use Helo´s benchmarks (not much ai) and yab with lots of ai https://forums.bistudio.com/topic/177406-yet-another-arma-benchmark/

[domokun 515]

jumpinghubert: I'm familiar with those threads. I meant what about your personal experiences, e.g. upgrade A cost B and offered C% improvement...

[St. Jimmy 272]

jumpinghubert: I'm familiar with those threads. I meant what about your personal experiences, e.g. upgrade A cost B and offered C% improvement...

I can give my experience:

 

RAM from 9-9-9-24 1600MHz -> 9-11-10-28 2133MHz

- 8-10% performance increase

 

Running Arma 3 from HDD vs SSD

- I drove in Kavala with hunter PiP on. 5-10fps increase when playing from SSD ~10% increase in that scenario.

 

CPU overclock

- 3,8GHz -> 4,7GHz should probably be around 15% increase at least.

[jumpinghubert 48]

Thanks Jimmy. Same results here. I will add two things:

1. the performance gain affects the more important minimum fps, not the average fps.

2. freds malloc tweak gave me roughly 20% more minimum fps in multiplayer (PvP).

 

P.S.

Ram is very cheap at the moment :)

[domokun 515]

Thanks Jimmy. Same results here. I will add two things:

1. the performance gain affects the more important minimum fps, not the average fps.

2. freds malloc tweak gave me roughly 20% more minimum fps in multiplayer (PvP).

 

P.S.

Ram is very cheap at the moment :)

jumpinghubert: thanks. Do you use the LPS and/or fed41 malloc described here?

https://forums.bistudio.com/topic/189347-lockless-memory-allocator-and-fred41-version-of-tbbmalloc/#entry3001270

[I give up 152]

STOP.

Want to discuss hardware performance, ram, cpu, ssd or whatever, there are plenty of threads open for the purpose, leave this thread out of it.

Want to discuss Fred's malloc, there are threads dedicated to the subject, even one in this forum section, leave this thread out of it.

 

This thread is about of a specific subject, about which I am 100% sure that is having a negative impact in game performance, in the hope that the developers can look at it

Unless you have a similar hardware configuration so we can really compare whats happening, I'll ask you people to stop making guesses.

 

I am waiting for a 8 GB VRAM single GPU so I can be sure that this is not related with crossfire, then I'll ask moderators to close this thread.

Still I am also asking for a bit more time because I am having some troubles to get the GPU since Pascal is coming out in a couple of weeks and is not wise decision to buy any AMD hardware atm, but I am confident that I will get one just for testing purposes.

 

Thank you for understanding.

[jumpinghubert 48]

jumpinghubert: thanks. Do you use the LPS and/or fed41 malloc described here?

https://forums.bistudio.com/topic/189347-lockless-memory-allocator-and-fred41-version-of-tbbmalloc/#entry3001270

I use fred´s malloc only from here: https://github.com/fred41/tbbmalloc_arma/blob/master/binary/tbbmalloc.dll

 

Yes bratwurst, i was a little bit offtopic, sorry. But the largeaddressaware-thing might be ontopic.

[St. Jimmy 272]

Hey bratwurste! When you monitor memory, I guess you monitor RAM usage and not the Page File usage with the afterburner?

Just tried again without Page File with my 8GB of RAM and GTX 970 but I couldn't get any higher total RAM usage out of ordinary before and when the out of memory. It seems to load the memory somewhere else or the windows forces to use some small page file so this can't be monitored without 16-32GB of memory.

Or then it's Crossfire or AMD specific or something.

One thing you could also try is to play Skyrim with some graphic mods if you've that game. I've read that some people also get RAM load or out of memory errors if they don't have enough page file set. Not sure if it's just about that the VRAM runs out first or is some RAM/page file already used before that.

[domokun 515]

 

STOP.

Want to discuss hardware performance, ram, cpu, ssd or whatever, there are plenty of threads open for the purpose, leave this thread out of it.

Want to discuss Fred's malloc, there are threads dedicated to the subject, even one in this forum section, leave this thread out of it.

 

This thread is about of a specific subject, about which I am 100% sure that is having a negative impact in game performance, in the hope that the developers can look at it

Unless you have a similar hardware configuration so we can really compare whats happening, I'll ask you people to stop making guesses.

 

I am waiting for a 8 GB VRAM single GPU so I can be sure that this is not related with crossfire, then I'll ask moderators to close this thread.

Still I am also asking for a bit more time because I am having some troubles to get the GPU since Pascal is coming out in a couple of weeks and is not wise decision to buy any AMD hardware atm, but I am confident that I will get one just for testing purposes.

 

Thank you for understanding.

 

Chill out. We're discussing RAM optimisations. This topic is entitled RAM management. So I think we're on topic here.

[St. Jimmy 272]

I now just upgraded my RAM so I've 16GB of it. I disabled page file and seems like I've now enough memory to play this game without it. I get now more memory usage but my total memory usage was 6,8GB (I had Steam and Discord open) when I had flew around the Altis couple times and played maybe an hour. It got pretty soon to that ~6GB but it didn't really increase a lot from there. My starting memory was 2,5GB so Arma 3 used around 4,3GB of RAM. I guess there's something else used somewhere else when you can't play Arma without page file with 8GB of RAM or you get "out of memory" issues and crashes.

 

I didn't get any stutter like you've been saying. Actually my map texture loading felt better.I don't actually remember getting that smooth performance with high view distances but I need to double check that later.

 

Here are some pictures from the test. I literally flew around the Altis 3 times and played couple time Altis Benchmark:

 

 

 

 

 

If you're monitoring that MSI Afterburner Page File usage, it showed ~10GB but I don't believe that's actually indicating anything really.

You can see my rig in my signature.

[kbbw123 115]

hm.... I might be wrong here but here is something straight out of my hardware components class of school:

 

You video (1st page)  is showing 2 diffrent memory indexes so I have the feeling you might be wrong with your interpetation there or the software your using is combining some stuf giving you a wrong picture of your acctual ram usage.

 

Normal ram works the following way: it holds the calculations and applications your cpu is running. once its full it will divert it to your harddrive in so called virtual ram. now if you look in taskmanager you can see them both listed, your acctual RAM and you Virtual RAM. A game as arma ussually has a limit of how much ram it can use for calculations, everything it can't hold in the RAM it should store in the virtual drive. meaning that if you went over the entire island it wil save all that info in the ever growing ram stick tillits full and it wil pass older info into the virtual RAM, If you go back to a place or have to load anything from an earlier point it will load it from the Virtual drive into you ram again to process it taking longer because HHD or SSD are both slower than DDR3 or DDR4.

 

Take by example Adobe After Affects, after rendering multiple video's in there you will want to clear its cache because it will easily store 20+ GB of data in your virtual RAM, the same has to be done in arma, after a while you have to flush your cache to get a performance boost (there is acctually a shortcut key for that in arma's 'cheat' shortcuts)

 

Try your first mission again on your normal rig and when you hit 13GB try flushing your cash (It might acctually work ;) )

[Vasily.B 529]

Hey, i just increased my ram memory from 8 gb to 16, and found that thread.
I was negatively supprised how bratwurste is threated and there is no react from Moderators. But more suprissed was resoult of using new ram.
There was no gain in FPS UNTILL i increased pagefile to manually size 800BM-24GB. Now space on my C drive is free in large procent and game runs much better. Also i use new performance EXE from Dwarden and its available in public. I use system memory allcoator and game runs very steady now - its 25-55 FPS - when its mostly quiet and enemy isnt aware its 55 fps (!) when big heat is coming up it falls down sometimes to 20 FPS, but mostly 25 is keeped... EVEN IN THE CITIES. Its great resoult for me.

[domokun 515]

Hey, i just increased my ram memory from 8 gb to 16, and found that thread.

I was negatively supprised how bratwurste is threated and there is no react from Moderators. But more suprissed was resoult of using new ram.

There was no gain in FPS UNTILL i increased pagefile to manually size 800BM-24GB. Now space on my C drive is free in large procent and game runs much better. Also i use new performance EXE from Dwarden and its available in public. I use system memory allcoator and game runs very steady now - its 25-55 FPS - when its mostly quiet and enemy isnt aware its 55 fps (!) when big heat is coming up it falls down sometimes to 20 FPS, but mostly 25 is keeped... EVEN IN THE CITIES. Its great resoult for me.

Which OS are you running?

[Vasily.B 529]

Which OS are you running?

Windows 7 64 bits.

[Dwarden 1124]

@Vasily.B well on Windows 7 x64 I would say that default allocator (Intel TBB v4) should work way better than System allocator ...

only Windows 8.1 and Windows 10 system OS allocator might be good enough

[Vasily.B 529]

@Vasily.B well on Windows 7 x64 I would say that default allocator (Intel TBB v4) should work way better than System allocator ...

only Windows 8.1 and Windows 10 system OS allocator might be good enough

Yes, i know, but somehow, its working better on system allocator. Its just like you sayed, it cannot be the same for every PC.