Jump to content
Placebo

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

Recommended Posts

That also makes sense, yes. I am sure other people will be able to help you on the topic (I am not that knowledgeable).

 

I am now running RAM at 3600 CL14 and it seems stable 🙂

 

Benchmark shows a stable 72.6 fps at LOW Settings (on par with 3466MHz CL14 so far) but the minimum frame rate jumped to a 46fps! HOLY guacamole. Remember it did start at 21fps with the XMP profile..

https://imgur.com/UVtoeWt

 

UPDATE:

 

YAAB - LOW Settings, 1440p:

CPU on AUTO, RAM on XMP: 66.8fps

CPU on AUTO, RAM on 3200 FAST: 71.9fps - 7.6% increase

CPU on AUTO, RAM on 3466 FAST: 73.5fps - 10.0% increase

CPU on AUTO, RAM on 3666 FAST: 74.4fps - 11.4% increase

CPU on AUTO, RAM on 3733 FAST: 75.6ps - 13.2% increase

CPU on AUTO, RAM on 3733 FAST with CMA_x64 Memory Allocator: 80.2fps - 20.1% increase

 

YAAB - ULTRA Settings, 1440p:

CPU on AUTO, RAM on XMP: 43.1fps

CPU on AUTO, RAM on 3200 FAST: 43.8fps - 1.6% increase

CPU on AUTO, RAM on 3466 FAST: 45.6fps - 5.8% increase

CPU on AUTO, RAM on 3666 FAST: 46.7fps - 8.35% increase

CPU on AUTO, RAM on 3733 FAST: 47.2fps - 9.5% increase

CPU on AUTO, RAM on 3733 FAST with CMA_x64 Memory Allocator: 53.8fps - 24.9% increase

 

NOTES:

  • I can't seem to be able to POST with RAM @ 3800 CL16 (and Infinity Fabric at 1900). It seems the end of the line for the memory I have and the motherboard/cpu combination is 3733MHz CL14, which is - to my standards - insanely high
  • I was able to keep it stable by keeing MISC, TERMINATION and CAD_BUS parameters (see DRAM Calculator for what I mean) to AUTO, except for the COMMAND RATE which I kept at 1T)
  • RAM Latency at the AIDA64 benches reached 64.5ns, the lowest I could get it
  • I am running at 1.45v DRAM Voltage and 1.1v SOC Voltage which is pretty within safe ranges I'd say
  • When you read "CPU on AUTO" it means I only tweaked the Infinity Fabric speed to match 50% of the RAM speed (e.g. 3200 RAM = 1600 IF and so on)
  • I ran a benchmark at ULTRA but removing AA and PP (from the default FSAA:8x and SMAA Ultra) and the difference is a 3fps average fps, give or take. It's crazy how efficient these cards are considering the difference from FSAAx8 + SMAA ULTRA to both DISABLED..
  • For some weird reason Cinebench r15 scores keep going lower than my highest score with the memory running at 3466... but I don't care because YAAB consistently says that faster ram = more fps

 

 

As usual, the screens for all my tests are here:

https://imgur.com/a/tXPyZE3

Share this post


Link to post
Share on other sites
14 hours ago, domokun said:

@Reezo thanks for your analysis. What was the performance impact in YAAB of switching to memalloc to cma_x64?

 

@domokun I just tried the CMA_x64 memory allocator and I got a:

 

+10% from the same exact settings at LOW

+14% from the same exact settings at ULTRA

 

The BIOS settings where:

CPU on AUTO and RAM 3733 MHz CL14 exactly as suggested in Ryzen DRAM Calculator "FAST" Settings

 

YsNt6n4.png

 

jibXFx9.png

  • Thanks 1

Share this post


Link to post
Share on other sites

The parts are here! I'll start building soon and then report back.. I might be streaming this just to add to the suspense 🙂

 

EPIHCDbXsAE4MHb?format=jpg&name=medium

Share this post


Link to post
Share on other sites

Fine choice of motherboard. I have one here. 🙂

Share this post


Link to post
Share on other sites
3 hours ago, Tankbuster said:

Fine choice of motherboard. I have one here. 🙂

 

Thanks! For once I was able to get what I “wanted”. But kudos to the my colleagues because they get nothing but the best stuff. They told me it’s hands-down the best motherboard you can get for a gen9 i7, period. I trust my brudas.

 

I love the “armor plate” on the rear side of the motherboard, it makes everything way less risky. Now I only have one side to dent with my screwdriver skills 😉

 

 

 

 

Share this post


Link to post
Share on other sites

Let us know if memory speeds has the same effect on the intel than it has on the ryzen. 

Share this post


Link to post
Share on other sites

9700k is almost 400 euros where i live. Sigh, don't think it would be worth it over the 9600k. 

Share this post


Link to post
Share on other sites

9600k will deliver exactly same FPS, in Arma, as client, at same frequency, provided you don't use OBS to stream/record.

Server side happily uses more threads/cores than client.

Share this post


Link to post
Share on other sites
15 hours ago, Reezo said:

 

Thanks! For once I was able to get what I “wanted”. But kudos to the my colleagues because they get nothing but the best stuff. They told me it’s hands-down the best motherboard you can get for a gen9 i7, period. I trust my brudas.

 

😉

 

Same here. For the first time, I had a budget that meant I didn't have to make as many compromises as I have done in the past. It's lovely being able to have your list of parts sorted by price and not having to scroll down so far.

Share this post


Link to post
Share on other sites
7 hours ago, Groove_C said:

9600k will deliver exactly same FPS, in Arma, as client, at same frequency, provided you don't use OBS to stream/record.

Server side happily uses more threads/cores than client.

 

I see. Do you need water cooling to get to 5GHz? 

 

I'm currently using a Cryorig H7 on my setup. How far can i push it with something like that? 

Share this post


Link to post
Share on other sites

No need for water cooling.

A high end air cooler is more than enough. A 240/280 mm AiO should be more than enough as well.

Share this post


Link to post
Share on other sites

I wouldn't go past 5.0 GHz, even if the CPU was capable of 5.1-5.2 GHz, since no FPS increase past 4.7-4.8 GHz in Arma.

So no need to stress the CPU with a lot of voltage for just 100-200 MHz more if it won't improve your FPS in Arma or if it will by like 1 FPS, average, which is within margin of error.

Share this post


Link to post
Share on other sites

Agreed on not going over 5.0GHz, I am going to aim at that.

 

The Intel i7 9700k at 5.0GHz on stock XMP memory profile will be - to me - a fair comparison to the AMD Ryzen 3700x at stock CPU (AUTO) but with the overclocked RAM. The Intel will feature the OC on the CPU and the AMD on the RAM. The two fields the two rigs excel at. Then I could try and see if I could also overclock the RAM on the Intel and take it an extra step further..

 

If you have any idea or thing you want me to try on the Intel.. that I haven't mentioned.. do it now 😉 as I won't have both rigs forever 😉

 

Good afternoon, everyone. After attending at SSL certificates renewal for work (I don't work in IT but I gotta take care of my own web sites ;)) I'll be starting the tests!

 

13 hours ago, opus132 said:

Let us know if memory speeds has the same effect on the intel than it has on the ryzen. 

 

Will do, for sure! First thing I want to do is compare the baselines (same systems at stock with XMP memory profiles only) and then compare both systems where their overclocking shines: Memory for the Ryzen and CPU for the Intel. Last, I'll try and see how beneficial memory overclocking is on the Intel.

 

13 hours ago, froggyluv said:

@Reezo Nice Rack!

 

Thanks! Glad you like it!

 

12 hours ago, opus132 said:

9700k is almost 400 euros where i live. Sigh, don't think it would be worth it over the 9600k. 

 

Yes it's insane, I think here in Germany is 389 or something.. yup. The 3700x is 325 so.. not so far apart.

 

10 hours ago, Groove_C said:

9600k will deliver exactly same FPS, in Arma, as client, at same frequency, provided you don't use OBS to stream/record.

Server side happily uses more threads/cores than client.

 

Cool to know! Makes sense that my colleague had a 9700k as all they do is basically gaming while streaming.. but streaming is their main purpose. Question: isn't the encoding being offloaded to NVENC (Hardware) in OBS if one has it? I would think at that point the CPU usage for encoding is minimal and it won't make a difference between AMD and Intel? I might be wrong @Groove_C

 

6 hours ago, Tankbuster said:

Same here. For the first time, I had a budget that meant I didn't have to make as many compromises as I have done in the past. It's lovely being able to have your list of parts sorted by price and not having to scroll down so far.

 

Yes, yes. I know the feeling 🙂 I've done it once when my wife inherited a sale of a house and told me: we gonna go buy you THE computer! And I was like yeeeeeee *jumping off the window while smiling* - that computer was the one I played ArmA 2 and OA on, actually! Many fond memories, I converted it to a workstation later on, even for video editing. Talk about 2,000 worth of computer, it did last me waaaay longer than expected and even resold it later. 2600k golden chip that would do 5,0GHz, I have the screenshot the store owner gave me and ordered me to frame it 🙂

 

This one rig is mostly built from half stuff that I own and half stuff I am borrowing. It still feels great as I get boxes that I can re-open and build, swear, sweat, and enjoy 🙂

  • Like 1

Share this post


Link to post
Share on other sites
26 minutes ago, Reezo said:

isn't the encoding being offloaded to NVENC (Hardware) in OBS if one has it? I would think at that point the CPU usage for encoding is minimal and it won't make a difference between AMD and Intel?

If there is an option to select nVidia NVEC codec in OBS, then this is what I would choose. This way, even if you have only 4 cores i3, it doesn't matter, since it's the GPU that will take care of almost everything and CPU load will be minimal.

Share this post


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

If there is an option to select nVidia NVEC codec in OBS, then this is what I would choose. This way, even if you have only 4 cores i3, it doesn't matter, since it's the GPU that will take care of almost everything and CPU load will be minimal.

 

That's what I use indeed. The GPU is barely getting used in ArmA 3 compared to other titles so that's where I am moving all my OBS-related encoding.

 

PN3jtp0.png

Share this post


Link to post
Share on other sites

The best RAM specs for Intel are also not above 3800 MHz 14-15-14, 15-14-14 or 15-15-14, since 4200-4266 MHz CL17 or 4000-4133 MHz CL16 is slightly worse than 3800 CL14/15.

Only higher bandwidth, but worse timings.

Even DDR3 2400 MHz <40 GB bandwidth is not restricting anything in anyways, DDR4 3800 MHz ~60 GB banwidth wouldn't as well.

Should result in not more than 40 ns, if all secondary and tertiary timings tightened as well. I would say even <40 ns, whereas on AMD it's impossible to go <61-62 ns with same 3800 MHz CL14/15.

Arma will 100% approve such low latancy!

I'm also @ 39-41 ns (Aida64 results tend to fluctuate from run to run) with my DDR3 2400 CL10 and Arma likes it very much.

Share this post


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

The best RAM specs for Intel are also not above 3800 MHz 14-15-14, 15-14-14 or 15-15-14, since 4200-4266 MHz CL17 or 4000-4133 MHz CL16 is slightly worse than 3800 CL14/15.

Only higher bandwidth, but worse timings.

Even DDR3 2400 MHz <40 GB bandwidth is not restricting anything in anyways, DDR4 3800 MHz ~60 GB banwidth wouldn't as well.

Should result in not more than 40 ns, if all secondary and tertiary timings tightened as well. I would say even <40 ns, whereas on AMD it's impossible to go <61-62 ns with same 3800 MHz CL14/15.

Arma will 100% approve such low latancy!

I'm also @ 39-41 ns (Aida64 results tend to fluctuate from run to run) with my DDR3 2400 CL10 (de and Arma likes it very much.

 

 

This is a comparison between the RAM at XMP profile in the 9700k/Z390 AORUS Master and the same stick at its best overclocking, on the 3700x/x470-f:

 

ctsjHal.png

 

vqv9yr3.png

 

 

Share this post


Link to post
Share on other sites

Cache should also be OC'ed as much as possible. Depends on core voltage as well.

cjelc4A.jpg

Share this post


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

Cache should also be OC'ed as much as possible. Depends on core voltage as well.

 

 

Ah! I haven't ventured into overclocking the cache, yet.. do you have a guide for it, maybe? Thanks @Groove_C <--bottomless pit of knowledge 😉

Share this post


Link to post
Share on other sites

Increase cache frequency and if not stable -> increase core voltage, as core voltage is for core and for cache.

Share this post


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

Increase cache frequency and if not stable -> increase core voltage, as core voltage is for core and for cache.

 

Got it! Thanks! Stay tuned later tonight for my live stream in which I'll reveal the results. Looking forward to have a healthy geek talk with myself and possibly fellow ArmA 3 veterans 🙂

 

UPDATE:

Found the setting.. it is now called Uncore Ratio on Gigabyte boards, or Ring Ratio.. they are the same. I am now at 47 before the 43 it was before. So 4.7GHz which should scale correctly against the 5GHz core. Perfectly stable so far!

Share this post


Link to post
Share on other sites

Here is today's episode about "AMD vs Intel for ArmA 3: Which is Best?!" for those who missed it.

 

 

  • Thanks 1

Share this post


Link to post
Share on other sites

It's really wrong to say, that Intel doesn't benefit as much from RAM OC as Ryzen.

Share this post


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

It's really wrong to say, that Intel doesn't benefit as much from RAM OC as Ryzen.

 

@Groove_C Intel CPUs don't benefit as much as AMD CPUs from RAM overclocking, while AMD CPUs don't benefit as much as Intel CPUs from CPU overclocking. This at least, we know 🙂

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

×