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

[Smart Games 76]

I wonder if the game prefers ram with tight timings or ram at high speed.

 

What would be faster? :

3600 cl14 vs

4400 cl19

 

And I'd like to know which Ram you used and your settings.

I mean.. 3200 cl14 to 4000 cl15 seems to be to good to be true. 

@1,5V?

 

 

 

 

[Tankbuster 1744]

On an intel platform, I found latency mattered more than frequency. My figures are here in this thread somewhere.  I'm away from pc now, so can't be any more specific.

[Groove_C 267]

3600 MHz CL14 gives more FPS in Arma than 4400 MHz CL19.
But if you manage to get your RAM to 3700 Mhz or 3733 MHz CL14, it's even better. Or 4000 MHz CL15.

And you can't look at RAM frequency or latency separately. RAM responsiveness (nanoseconds in Aida64) depends on the combo of frequency and latency, not only from 1 of both.

 

Higher RAM frequency is only GB/s bandwidth, not RAM responsiveness.

If a game is not limited by RAM bandwidth (Arma isn't even by DDR3 bandwidth), better responsiveness gives better FPS.

 

The better RAM responsivness, the quicker it can accept data from CPU, treat it and send it back to the CPU, so the CPU can transmit it quicker to the GPU, so the GPU can not only produce higher FPS, but also next frame will follow after the previous much sooner/more often = smoother game.

 

If you see RAM with very high frequency, but because of this its timings are much higher, then it's not worth.
Good RAM to buy, to be sure it has good memory chips for further squeezing out of additional performance, is G.Skill Trident Z (Neo/RGB) 3600 MHz 16-16-16-36.

[oldbear 390]

I would like to show results from with my R5 3600X test config and different DDR4 kits on YAAB

• Corsair Vengeance LPX 16 GB (8GBx2) 3200 MHz C16 [16-18-18-36]
• G.Skill FlareX 16 GB (8GBx2) 3200 MHz C14 [14-14-14-34]
• G.Skill TridentZ Neo 16 GB (8GBx2) 3600 MHz C16 [16-19-19-39]

[Groove_C 267]

3200 MHz CL14 (XMP) and 3600 MHz CL16 (XMP) is almost the same, 3600 MHz CL16 (XMP) being slightly better and 3200 MHz CL16 (XMP) is not far from 3200 MHz CL14 (XMP).

And it's only primary timings that don't matter as much as dozens of secondary and tertiary timings, so...

[Groove_C 267]

So it's 3600 MHz CL16 (XMP) > 3200 MHz CL14 (XMP) > 3200 MHz CL16 (XMP).

But 4000 MHz CL15 > 3800 MHz CL14 > 3733 MHz CL14 > 3800 MHz CL15 > 3700 MHz CL14 > 3600 MHz CL14 > 3700 MHz CL15 > 3600 MHz CL15 > 3600 CL16 (XMP).

 

From best to worst.

[Groove_C 267]

4000 MHz CL15 is for example just 1.79% less responsive than 3800 MHz CL14 (in nanoseconds), but has higher bandwidth, so 4000 MHz CL15 is still slightly better than 3800 MHz CL14 overall.

That's why you can see the rest of frequency vs. timing(s) in this specific order.

 

Same goes for 3600 MHz CL16, that's just slightly less responsive than 3200 MHz CL14, but has gigher bandwidth, so 3600 CL16 is still slightly better than 3200 MHz CL14 overall.

[Smart Games 76]

What Ram do you recommend for best Performance? 

[Groove_C 267]

G.Skill Trident Z RGB 4000 MHz 17-17-17-37 if Intel only

2x8 GB - https://geizhals.de/g-skill-trident-z-rgb-dimm-kit-16gb-f4-4000c17d-16gtzr-a1755388.html

4X8 GB - https://geizhals.de/g-skill-trident-z-rgb-dimm-kit-32gb-f4-4000c17q-32gtzr-a1755380.html

 

G.Skill Trident Z Neo 3600 MHz 16-16-16-36 if AMD or Intel
2x8 GB - https://geizhals.de/g-skill-trident-z-neo-dimm-kit-16gb-f4-3600c16d-16gtzn-a2099434.html

4X8 GB - https://geizhals.de/g-skill-trident-z-neo-dimm-kit-32gb-f4-3600c16q-32gtzn-a2099487.html

2X16 GB - https://geizhals.de/g-skill-trident-z-neo-dimm-kit-32gb-f4-3600c16d-32gtzn-a2099396.html

 

G.Skill Trident Z black/white 3600 MHz 16-16-16-36 if Intel only

2x8 GB - https://geizhals.de/g-skill-trident-z-schwarz-weiss-dimm-kit-16gb-f4-3600c16d-16gtzkw-a1505578.html

 

G.Skill Trident Z RGB 3600 MHz 16-16-16-36 if Intel only

2x8 GB - https://geizhals.de/g-skill-trident-z-rgb-dimm-kit-16gb-f4-3600c16d-16gtzr-a1561158.html

4X8 GB - https://geizhals.de/g-skill-trident-z-rgb-dimm-kit-32gb-f4-3600c16q-32gtzr-a1561113.html

2X16 GB - https://geizhals.de/g-skill-trident-z-rgb-dimm-kit-32gb-f4-3600c16d-32gtzr-a2152645.html

[Smart Games 76]

My new ram is arriving tomorrow. I'll be testing some speeds with YAAB and sharing the results here.
Even if the settings are optimized for me, clear patterns should be visible.

 

[Groove_C 267]

@Smart Games which RAM model have you ordered and what is your CPU and motherboard model?

 

Have foud Trident Z silver/red 3600 15-15-15-35 2x8 GB. Thought it wasn't available anymore.

https://geizhals.de/g-skill-trident-z-silber-rot-dimm-kit-16gb-f4-3600c15d-16gtz-a1439046.html
 

It's slightly more responsive than 4000 MHz 17-17-17-37 and also cheaper. Even cheaper than 3600 MHz 16-16-16-36 Neo/RGB and even more responsive.

But it's for Intel - not AMD + the color is not that nice, if you care.

[Smart Games 76]

@Groove_C Got some cheap G.Skill Ripjaws 3600 cl16 Ram. I read some reviews and should be able to oc it quite a bit. 

[Groove_C 267]

@Smart Games CPU and motherboard model?

 

There are 2 different RipJaws V 3600 MHz CL16 -> 16-19-19-39 and 16-16-16-36, which is not the same.

[Smart Games 76]

@Groove_C 16-16-16-36, Z390 Phantom Gaming 4s

[Groove_C 267]

@Smart Games CPU model? 9350KF?

[Smart Games 76]

i3 9350kf @5/4,6 Ghz

[Groove_C 267]

@Smart Games RipJaws V 3600 16-16-16-36 has same memory chips as all Trident Z 3600 16-16-16-36 variants. Just look simpler and thus cost less.

It's possible to OC it a lot with very low timings.

But you could have bought these, with same specs and same price.
https://geizhals.de/g-skill-trident-z-schwarz-weiss-dimm-kit-16gb-f4-3600c16d-16gtzkw-a1505578.html
Just look better and are slightly colder.
They were in the list I made few messages higher.

[Smart Games 76]

5 minutes ago, Groove_C said:

@Smart Games RipJaws V 3600 16-16-16-36 has same memory chips as all Trident Z 3600 16-16-16-36 variants. Just look simpler and thus cost less.

It's possible to OC it a lot with very low timings.

But you could have bought these, with same specs and same price.
https://geizhals.de/g-skill-trident-z-schwarz-weiss-dimm-kit-16gb-f4-3600c16d-16gtzkw-a1505578.html
Just look better and are slightly colder.
They were in the list I made few messages higher.

Thanks for your suggestion, but my Pc case is closed 🙂

[Groove_C 267]

@Smart Games they can do 4000 15-15-15 or 3800 14-14-14 or 3733 14-14-14, provided your CPU memory controller and motherboard are good + needed but still human SA, IO and VDIMM voltages have to be applied.

See here:
https://www.hardwareluxx.de/community/threads/intel-ram-oc-guides-und-tipps.1230518/
https://www.computerbase.de/forum/threads/intel-core-i-serie-ram-overclocking-auswirkungen-auf-spiele.1849970/

[Smart Games 76]

My Setup:

• I3-9350KF @5Ghz/4.6Ghz
• Zotac Gtx 1080 Ti Mini @Stock
• Z390 Phantom Gaming 4s
• HyperX Predator / G.Skill Ripjaw V

 

My In-Game Settings:

• Sampling: 100%
• Texture: Ultra
• Objects: Ultra
• Terrain: Ultra
• Shadow: Ultra
• Particles: Low
• Cloud: Ultra
• PIP: Disabled
• Overall: 2500m
• Object: 2500m
• Shadow: 100m
• HDR: Standard
• Dynamic Lights: Ultra
• Water Reflections: Disabled
• Display Mode: Fullscreen
• Resolution: 3840x2160 / 16:9
• Bloom: 200
• R.B., R.B., D.of F., S.F.: 0
• AO: Disabled
• Caustics: Disabled
• FSAA: 2x
• ATOC: All trees and grass
• PPAA: Disabled
• Anis. Filtering: Ultra

 

(Just for the Screenshot a fullscreen window)

 

Result (3x Yaab): 

 

• 4000 15-15-15-28 300 (+ opt. Subtimings) = (52.2), (51.9), (52.0) = 52.03 (100%)
• 4000 15-15-15-28 300: (50.4), (50.9), (51.2) = 50.8 (-2.36%)
• 4000 15-15-15-28 500: (51.1), (50.4), (49.8) = 50.4 (-3.13%)
• 4000 15-15-15-35: (51.8), (49.1), (49.2) = 50.0 (-3.9%)
• 4000 16-16-16-36: (50.1), (49.8), (49.4) = 49.8 (-4.28%)
• 4000 17-17-17-37: (49.9), (50.4), (48.2) = 49.5 (-4.86%)
• 4000 18-18-18-38: (50.4), (48.3), (49.1) = 49.3 (-5.24%)
• 4000 19-19-19-39: (50.7), (48.5), (48.4) = 49.2 (-5.44%)
• 3600 14-14-14-34: (48.7), (47.9), (49.3) = 48.6 (-6.59%)
• 3600 15-15-15-35: (48.4), (49.4), (47.8) = 48.5 (-6.78%)
• 3600 16-16-16-36: (48.1), (47.6), (47.5) = 47.7 (-8.32%)
• 3200 13-13-13-33: (48.2), (47.2), (46.3) = 47.2 (-9.28%)
• 3200 14-14-14-34: (48.1), (46.8), (46.3) = 47.0 (-9.66%)
• 3000 14-14-14-34: (47.6), (45.4), (44.3) = 45.8 (-11.97%)
• 3200 15-15-15-35: (46.3), (46.0), (44.9) = 45.7 (-12.16%)
• 3200 16-16-16-36: (46.6), (45.4), (45.0) = 45.6 (-12.36%)
• 2666 13-13-13-35: (46.1), (43.3), (43.5) = 44.3 (-14.85%)
• 2400 12-14-14-35: (42.2), (43.1), (42.3) = 42.5 (-18.31%)
• 2133 12-14-14-35: (41.1), (39.2), (40.3) = 40.2 (-22.73%)
• 1600 12-14-14-35: (33.1), (33.7), (31.7) = 32.8 (-36.96%)

 

Standard = X x-x-x-x 650 

 

Even in 4k RAM speed matters. Ram OC seems to be worth it!

 

Optimized RAM Settings:

[Groove_C 267]

Well, this must have cost you some time to do, but I have to tell you, that primary timings are not those that bring the most FPS, but are only partially responsable for higher FPS.

 

There is even more min. FPS to squeeze out (vs. XMP), despite your 4K resolution and pretty high graphics settings, once you manage to tighten every single timing to the absolute min. stable/working values.

+ it makes FPS more stable = less and not so severe dips.

 

So while your work already shows slight improvements in having better configured RAM, despite your 4K resolution and pretty high graphics settings, it's only partially true, since FPS difference can be even higher than that.

Hope nobody will take your work to show that better RAM is useless, because of "so tiny" improvements.

 

So yes, playing at "only" 1080p and not highest graphics settings, lets the RAM make even bigger min. FPS difference, because at higher resolution and graphics settings, you're more and first limited by the GPU, long before CPU and/or RAM become the limiting factor.

But even in this case, without even fully tunning the RAM, there is like 5 FPS difference already.

[Smart Games 76]

34 minutes ago, Groove_C said:

 

@Groove_C that's true, but most players won't change any timings. Im not very experienced with RAM OC.. Show me a good tutorial for the overclock of secondary items and i will update my post

[Groove_C 267]

@Smart Games I already posted links for 2 best german forums, concerning RAM OC (Intel specific). There is already a lot of info and you can ask for advise.

 

But 6 FPS more at 4000 MHz 15-15-15-28-300 vs. your 2666 MHz 13-13-13-35-650, considering your 4K resolution and pretty high graphics settings, it's already a good improvement, especially considering how low FPS is in Arma.

It's not like going from 100 to 106 FPS )))

[Groove_C 267]

R7 3800 XT is second from the bottom, not even talknig about 3600(X/XT), 3700(X) or 3800X
Well, it's only synthetics and not a game or rendering program, but still.

[Groove_C 267]

Tested today in multiplayer
my 4790K 4.8/4.4 GHz core/cache + 32 GB 2400 CL10 MHz DDR3
vs. 9700K 5.1/4.8 GHz core/cache +32 GB 4000 CL15 MHz DDR4

 

Results were same (+- 2-3 FPS difference, when view distance was 500-1500m).
So the higher the view distance, the lesser FPS difference between newer and older CPUs. Same goes for the GPUs.
The only scenario, where a more powerful CPU/GPU can make a visually unnoticeable, but still possible to measure (for the record) difference is between 500-1500m view distance.
So really not worth to OC your Intel CPUs higher than 5.0/4.7 GHz core/cache.
Also 4000 MHz RAM vs. 3200 MHz (XMP) made less than 5 FPS difference.

We both have same Windows version and SSD model and only Arma, Steam, Discord, Samsung Magician and MSI Afterburner were running on our computers.
GPUs weren't helping/hindering none of us, since we've done the tests in 1080p standard and 1080p low + everything possible disabled.

Server FPS was same all the time (almost 50 FPS) + hundreds of AI infantry + not more than 15 armored vehicles and not more than 10 jets/helis.
We were always looking at same spot, had same resolution and graphics settings and view distance (tried 500-7000m).
PvE server was full and we tested FPS as infantry at base, in Pirgos and Kavala, with 500-1500m.
And also hoovering 4 kms away from Pirgos and Kavala in the heli at 500m altitude, looking at these 2 villages with 7000m view distance.
 

So in MP, in Arma, you're not limited by server hardware, when server already always runs at max possible FPS (default 50 FPS).
But FPS you experience is limited by network traffic and Arma's netcode.
Even if the server would have been something like a new R9 5900X, instead of OC'ed i7-3930K, since FPS is already now almost 50, server FPS would have remained same, hitting 50 FPS limit.
And I doubt that with more recent/powerful hardware, there would have been less/more network traffic.

So what data gets requested/received by the server from what number of clients and what's synchronized to what number of clients is what dictates the client FPS, provided client already has a very good hardware basis.
So past certain client hardware performance level, there is 0 performance increase in MP, in Arma.