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

[Tankbuster 1733]

Groove are you saying that the more cores you have, the more you could gain by turning hyperthreading off?

[Groove_C 267]

@Tankbuster potentially, yes.

[oldbear 390]

From my point of view, it's better to use a Ryzen 5 3600 just out of the box without using "special" settings

 

Here are YAAB results from my last build ...

 

 

The only tweak has been to set the XMP DDR4 profile in the BIOS.

Letting all the cores/treads allows for a smoother gaming experience.

Here the limits are not in the hardware but in the game itself.

You will gain near nothing trying to expel some more FPS.

[oldbear 390]

At the risk of appearing excessively insistent, the limits are not in the hardware but in the game itself. You will gain near nothing trying to expel some more FPS at the risk of less smooth game experience.

 

 

[oldbear 390]

From my point of view, the only bench that is worth it is the practical experience that everyone can have while playing Arma3.

I don't have a different gaming experience with the slightly overclocked i7-7700K, the R5 3600X and the R5 3600.

It's just enjoyable and that's what matters, it's an answer to the founding questions of the topic.

 

 

 

[Groove_C 267]

If the min. and avg. FPS are higher in YAAB, it's an indicator of it possibly being also higher in MP, provided the server hardware is good and the mission is well and not heavily-scripted.

[Tankbuster 1733]

OK. If I'm mission developing, I'll often have a local server running too. It'll be interesting to see how it handles that.

[Groove_C 267]

The thing is that Arma server can deal with a lot of cores very well, but not Arma client.

[Tankbuster 1733]

3 hours ago, Groove_C said:

The thing is that Arma server can deal with a lot of cores very well, but not Arma client.

I'd like to think 8 cores would be enough for the OS, game server and game client. I'll try it out soon

[Groove_C 267]

We've checked the performance in YAAB with SMT off and less than 8 cores and the performance is literally the same.

Significantly Increasing CPU's L1 cache read/copy bandwidth per core by reducing the number of threads/cores, thus increasing the cache size per core, didn't help at all.

Must certainly be, that Arma simply isn't pushing the limits of CPU's L1 cache read/copy bandwidth and cache size per thread and there is more than enough headroom left.

 

Therefore I have decided to remove my recent posts on this matter.

 

Keep all the threads/cores enabled for silky smooth gaming experience.

[Tankbuster 1733]

If that's the case, how can we explain the performance of the 5775?

[oldbear 390]

In 2015, Hardware.fr testers had pointed out in the Core i5-5675C test that certain applications including Arma3 took advantage of the presence of the 128 MB eDRAM chiplet associated with the main cpu chip.

I seems to be related to the way the eDRAM is "doing well with a 2.4x / 2.7x higher bandwidth than DDR3-1600 and a latency of 1/3 lower!"

 

[Groove_C 267]

@oldbear with my i7-4790K, my DDR3 2400 MHz 10-11-12-18-1T, in Aida64, are producing 36,5 GB read, 37,5 GB write and 36,5 GB copy.

With same RAM, my i7-5775C produces 38,3 GB read, 37,6 GB write and 41,4 GB copy.

Then to this DDR3 bandwidth, you have to add 55,3 GB read, 37,3 GB write and 45,5 GB copy from eDRAM @ default 1800 MHz, that is now OC'ed to 2200 Mhz, but I haven't remeasured since, because lazy )

This results in total bandwidth of 93,6 GB read, 74,9 GB write and 86,9 GB copy, since eDRAM sits right inbetween the cores and RAM.

 

My i7-5775C has better bandwidth than even dual channel DDR4 4600 MHz + lower RAM latencies )))

A very high bandwidth is very important for games, where there is a lot of textures of very high resolution and a lot of highly detailed 3D objects.

Same goes for 128 MB L4 cache/eDRAM (16 MB per thread with HT enabled).

 

In A3, because of not so high resolution textures and not so detailed 3D objects, even DDR3 bandwidth (without eDRAM) is not limiting.

The FPS boost comes only from 16 MB cache per thread vs. only 1 MB cache per thread for something like an i9-9900K.

 

I have a friend, who also plays Arma, since A2, that had an i7-2700K 4.9 GHz + very high speced 4x8 GB DDR3 2400 MHz 10-12-12-35 + GTX 1070 and 1080p monitor and was having only min. 27-30 FPS in YAAB @ 1080p standard.

He's not a lawyer and has not so much money, but wanted to increase FPS in Arma, since it's the main game he plays.

So yesterday we've made him a budget upgrade, for Christmas, by only replacing the mainboard and CPU, so he could keep his DDR3 2400 MHz RAM and his GTX 1070, which are both more than enough for max FPS @ 1080p ultra.

 

We've bought him for only 90€ an Asus Z97 Deluxe USB 3.1 Wi-Fi + Bluetooth mainboard, which has 16 top quality VRMs (overpower for the tiny i7-5775C), USB BIOS Flashback button, clear CMOS button, MemOK button and reset + start buttons.

Also at rear 2x USB 3.1, 4X USB 3.0, 4x USB 2.0 and 2x Intel 1 GB lan + 2x 2 USB 3.0 headers for front of the case connection.

Very good quality of onboard audio, with separation for left and right tracks, built in amplifier for high end headphones and japanese capacitors.

+ 200€ i7-5775C, which is now overclocked to 4.2/3.8/1.8 GHz core/cache/eDRAM.

 

In YAAB 1080p standard he now has min. 41 FPS, avg. 69 FPS and max.105 FPS.

This is much much better, instead of min. 27-30 FPS that he was having with his i7-2700K 4.9 GHz.

And it's also only 290€.
It's much much less than buying a new mobo for more than 300€, in order to have a lot of good quality VRMs, display for error codes, MemOk button, if OC'ed RAM and can't boot, clear CMOS button if OC failed and can't boot and USB BIOS Flashback button if changing for new CPU, but no other more old CPU available to update the mainboard BIOS, when it comes with not latest BIOS, required for brand new latest CPU.

Also if you see the prices of high performance 32 GB DDR4 RAM kits...

And also don't forget the prices for new top CPUs.

 

Arma not only just flies, because of the FPS, but also feels very very smooth as well.

 

[Groove_C 267]

I was thinking, that only 8 threads of i7-5775C would result in less FPS in heavily-threaded games, like Battlefield 5, where even a R9 3900X, with its 24 threads is very well used.

And that it would throttle the GPU, like RTX 2080 Ti and won't let it reach >90% load in games.

 

But I was completely wrong!

 

Look here. One geek, who decided to invest in i7-5775C (he knows what he's doing), instead of changing to something like i9-9900KS + new expensive mobo + expensive custom watercooling + 32 GB expensive DDR4 RAM.

And he definitely has the $$$ for something like i9-9900KS, since he has a RTX 2080 Ti and high refresh rate 1440p monitor!

 

The guy is playing BF5 in 1440p with everything at max + RTX enabled and DLSS, on his RTX 2080 Ti.

The FPS is very very high (considering RTX enabled).

His RTX 2080Ti load is over 90%, closer to 100%, frametime is very low and the line is relatively flat.

And this is despite him using a heavily fragmented HDD and only 8 threads of his i7-5775C, which is almost all the time fully loaded.

 

So I can conclude, that it's the 128 L4 cache/eDRAM magic.

Because I have seen other i7 CPUs with 8 threads and only 8 MB L3 cache in BF5 and the FPS was not that high, frametime was higher, not so flat and the GPU load was lower, which indicates, that the CPU is holding the FPS/GPU back.

 

And look at the CPU power consumption... simply ridiculous )))

 

So if a game like BF 5, that is so heavily-threaded, is not a problem for the tiny i7-5775C with only 4.2 GHz and only 8 threads, then next Arma 100% won't be a problem as well, at all.

Since I don't think, that next Arma will be as heavily-threaded as BF is.

Will only need to upgrade the GPU, since I already have 32 GB RAM and ~90 GB/s read/copy bandwidth of RAM+eDRAM is simply more than a lot )

Also 128 MB cache won't be too much for nextgen games with very high resolution textures and highly detailed 3D objects.

[Groove_C 267]

@oldbear btw, hardware.fr have tested the i5-5675C at 4 GHz vs. i5-4670K not in the correct way, since they have only increased core frequency of both to 4 GHz, but i5-4670K has its cache at 3.4 GHz, whereas i5-5675C has it's cache at 3.2 GHz.

So the performance difference would probably have been a few % higher, for the i5-5675C.

[Groove_C 267]

Btw, with 128 MB L4 cache/eDRAM disabled, in YAAB 1080p standard 4.2/3.8 GHz core/cache, FPS goes down to:

min. 25, avg. 47, max. 74

vs.

min. 42, avg. 69, max. 107

 

And this is with top performance RAM.

So the CPU behaves like a normal CPU, with only 4.2 GHZ, only DDR3 and only 6 MB L3 cache.

Imagine people playing with less than 4.0 GHz CPUs, like 3.3-3.6 GHz + these CPUs are like real dinosaurs, by now... + really cheap RAM.

[Groove_C 267]

Windows 10 + BIOS up to date

Launcher mods, browser, Discord, Skype, Teamspeak etc + all known background programs/processes off

 

YAAB 1080p standard

 

R7 3700X ~4.2 GHz | 2x8 GB 3600 MHz 14-15-15-35 1.35 V | GTX 1080 | SATA SSD

Run 1: min 43 avg 65,4

Run 2: min 42 avg 63,8

Run 3: min 43 avg 63,5

vs.

Run 1: min 36 avg 53

Run 2: min 33 avg 52

Run 3: min 37 avg 56

i9-9900K 4.8/4.3 GHz core/cache | 4x8 GB 3200 MHz 16-18-18-36 1.35V | GTX 1080 Ti | SATA SSD

 

R7 3700X 4.2 GHz (~330€) + Gigabyte X570 Aorus Elite (~205€) + 2x16 GB 3600 16-16-16-36 (~300€) + Scythe Mugen 5 Rev. B (~45€)

total 880€

vs.

total 995€

i9-9900K 4.8/4.3 GHz core/cache (~530€) + Asus ROG Strix Z390-E Gaming (~230€) + 4x8 GB 3200 MHz 16-18-18-36 (~150€) + Noctua NH-D15 (~85€)

 

i9-9900K build is 115€ more expensive, despite much worse performance and despite 2x cheaper RAM.

 

Ryzens right now hold the performance crown even in low-threaded games, like Arma, Tarkov, Squad and Fallout.
Imagine their performance in well multi-threaded games!
+ SMT is more efficient than Hyper-Threading from Intel
+ don't forget about the in-game smoothness on Ryzens, since they have whopping 32 MB L3 cache vs. just 16 MB L3 cache for i9-9900K
+ even R5 3600X for just 230€ has same frequency per thread as R7 3700X and still has whopping 32 MB L3 cache and better SMT than Intel's Hyper-Threading
+ Ryzens are more than well served with just 35-45€ regular air coolers and don't even heat that much

+ lower consumption

+ for same price as an i9-9900K one can buy 12/24 cores/threads from AMD with whopping 64 MB L3 cache vs. just 8/16 cores/threads and just 16 MB L3 cache )))

[Groove_C 267]

On 12/7/2019 at 8:38 PM, oldbear said:

Sorry, but I don't understand how "PCI-E 4.0 can give you 3-5 FPS more than PCI-E 3.0" in Arma.

 

a B450 Mobo featuring an updated BIOS will do the job.

Here is a proof/demonstration of higher FPS and also lower and flatter frametime (smoothness) with PCI-E 4.0 vs. PCI-E 3.0.

And it's just a low end RX 5500 XT, not even a RX 5700 XT.

Also less vRAM, negatively impacts FPS and frametime, because of data, that starts to flow from vRAM to much much slower RAM or to even much much much slower SSD, even if it's NVME M.2.

Nothing to do with whether it's Arma or not.
It's physical/software behavior in any game.

 

I WOULDN'T BUY ANY MOBO, THAT HAS NO PCI-E 4.0 SUPPORT ANYMORE.

Source: https://www.pcgameshardware.de/Radeon-RX-5500-XT-Grafikkarte-275529/Specials/PCI-Express-3-vs-PCI-E-4-GPU-1339415/

 

Intel with its 10th gen CPUs will still have only PCI-E 3.0 support 🤣

Bye bye Intel (at least for 2020).

 

AMD Ryzen 3xxx CPUs not only support PCI-E 4.0, have more cores/threads for much less money, with lower consumption and higher performance.

But in 2020, new Ryzens will also operate at even higher frequencies and will have 64 MB L3 unified cache, instead of 2x16 MB L3 cache, where half of threads can access only half of cache.

[oldbear 390]

Sorry, but here, you miss the whole picture.

The RX 5500XT issue is related to the way this card has been designed, why AMD decided to make the 5500XT run on only PCIe x8 instead of full PCIe x16 ?

So in this very special situation, PCI-e 4.0 has some advantage over PCI-e 3.0 due to an AMD mistake allowing 4GB GPU release .

[Groove_C 267]

@oldbear I missed the part about 8x.

But anyways, at least, one can see how the FPS and frametime (smoothness) suffer on <8 GB GPUs.

 

Have switched my mainboard to PCI-E 3.0 x8 and benched YAAB 1080p standard like 7 times, and there is no difference vs. x16.

 

They must have chosen worst case scenario games, to demonstrate the issues with restricted bandwidth and vRAM.

It's maybe due to the fact, that Arma has not that good quality/resolution of textures/3D objects, that there is no difference.

[oldbear 390]

So now and here, go buy a RX 570 ... it's a lot cheaper and runs on all PCI-e 3.0 Mobo or get a  GTX 1650 Super without those limitations 🙄

[Groove_C 267]

@oldbear

I suggest you to run YAAB 1080p standard with CMA x64 AVX2 memory allocator.

It further improves FPS (by using AVX2 instructions set of the CPU, instead of much older and not so efficient SSE instructions set).
It further improves the smoothness of the game (frametime) (by using 2 MB chunks of memory, instead of 4 KB chunks of memory).

- Download it here: http://www.armaholic.com/page.php?id=31217

- Navigate in archive to malloc -> 64-bit -> AVX2 folder

- Copy/move cma_x64.dll file to Steam -> steamapps -> Common -> Arma 3 -> DLL folder

- Navigate in archive to malloc -> 64-bit -> Config -> WithLargePages folder

- Copy/move cma.ini file to Steam -> steamapps -> Common -> Arma 3 root folder

- Type secpol in search and select Local Security Policy - Navigate in Security Settings to Local Policies -> User Rights Assignment - > Lock pages in memory

- Right click on Lock pages in memory and select Properties

- Click on Add User or Group - in the field Enter the object names to select, type your PC username and click on Check Names

- it should add your PC user name

- then click OK and OK and reboot the PC

- Navigate to Arma 3 root folder, right click on arma3_x64 and in Properties -> Compatibility, select Run this program as an administrator

- launch Arma 3 launcher

- navigate to PARAMETERS tab -> ALL PARAMETERS

- locate and check Memory allocator (64-bit) in Advanced section

- select cma_x64 memory allocator and start the game

 

Let us know if it changed something for you.

 

P.S. - should be done only on systems with not less than 16 GB RAM.

[Reezo 45]

Just dropping in to thank you for the wealth of information, research and dedication that is this thread. I'm building a rig to use for streaming the whole campaigns and scenarios from ArmA 3 to gather donations for charity for a dear friend of mine and this thread was all I could ask for and more. As of right now I am going the Ryzen way, possibly 3800x if a colleague of mine is reeeeeeally willing to let go 😉

[oldbear 390]

Hi Reezo,

Even if the boost to 4.5 MHz of the R7 3800X is tempting, the cost (I am not talking only about the price as new equipment) from the point of view of electrical consumption and heating is high.

With a TDP up to 105W, this processor requires a more adequate cooling solution than the standard Wraith Spire cooling which does not lack quality, but which in this case is really undersized.

I think you have to integrate the cost and size of a high-end cooling solution in the plan.
More precisely, because my choices for cooling this processor go more towards air cooling, you must consider a solution at the Noctua NH-U12S level.

 

 

[Reezo 45]

6 hours ago, oldbear said:

Hi Reezo,

Even if the boost to 4.5 MHz of the R7 3800X is tempting, the cost (I am not talking only about the price as new equipment) from the point of view of electrical consumption and heating is high.

With a TDP up to 105W, this processor requires a more adequate cooling solution than the standard Wraith Spire cooling which does not lack quality, but which in this case is really undersized.

I think you have to integrate the cost and size of a high-end cooling solution in the plan.
More precisely, because my choices for cooling this processor go more towards air cooling, you must consider a solution at the Noctua NH-U12S level.

 

 

 

Pretty much what I was thinking. This 2700x at 4.3 GHz with OC'd RAM is pushing quite high in the scores (only doing CineBench/RealBench/aida for the moment).

 

For cooling, I have a Noctua NH-D15 Premium (with 2x NF-A15 PWM 140 mm) - I suppose this would be ok for the 3800x as well?