Arma 3 CPU vs RAM performance comparison 1600-2133= up to 15% FPS gain

[rundll.exe 12]

It might depend a bit on your definition of "performance". If your definition is FPS, you HDD doesn't matter at all as long as the scene is stationary, and requires no memory swapping.

On the other hand, if your definition is smoothness, then the swapping is noticeable and a SSD is useful. But in my experience, any SSD will do, and if it keeps up with the scene loading you're good. I didn't experience any stuttering at all after I switched from a HDD to SSD, so what's the gain of even more read speed?

[jumpinghubert 49]

I wrote the same as you. No performance gain with ssd, only less/no stuttering (=more smoothness). Nothing new.

New&crazy is only bratwurste´s statement you might have overlooked ...:

a) faster ram bandwidth is only placebo (its trivial lower clocked ram with low latency might have same bandwidth as higher clocked ram with very high latency)

b) faster ssd bandwidth (from ssd to ssd-raid or ramdisk) = gain in performance (fps).

a) ignores the evidence in this thread

b) without proofs, tests...only assumption.

The tests i am waiting for are two benches. First one with ssd, second one with ssd-raid or ramdisk regarding frames per second or smoothness (frametimes and so on). But you don´t will see it because there is a little bit trolling going on lol.

Edited July 27, 2015 by JumpingHubert

[I give up 152]

It might depend a bit on your definition of "performance". If your definition is FPS, you HDD doesn't matter at all as long as the scene is stationary, and requires no memory swapping.

On the other hand, if your definition is smoothness, then the swapping is noticeable and a SSD is useful. But in my experience, any SSD will do, and if it keeps up with the scene loading you're good. I didn't experience any stuttering at all after I switched from a HDD to SSD, so what's the gain of even more read speed?

It has become clear that You do not understand the operation concept of Arma 3, or even what is RAID0 and you are not interested in knowing it.

@JumpingHubert

Get the faster RAM, whatever may be, dont think twice.

Peace.

[dnk 13]

I think it's pretty well established that Arma's bottleneck is most definitely the CPU's single threaded performance, with simplified front ends (ie shared across cores) causing further issues most likely. I think this has been established since late beta, basically as soon as we realized its "multithreaded/core" support was still getting caught up on this as before (though it's still an improvement, but not enough to overcome the increased performance demands). I'm not sure there's much to be done there short of a DX12-esque miracle cure for draw calls or a total rewrite of the engine (or new engine).

I have an SSD. I performed a lot of tests between it and a 7200RPM drive (posted somewhere on the forum). There was next to no actual FPS gain, though it did cut out a LOT of microstutters (and a few macro ones) as well as, imo, "fixing" a lot of crashes (game stalls for 2sec, but unlike before doesn't CTD).

The only reason my RAM upgrade improved performance was that (contrary to my prior beliefs) Arma needed more than 4GB on the system to run smoothly. It wasn't a huge difference anyway (in performance or speeds, just size).

[rundll.exe 12]

It has become clear that You do not understand the operation concept of Arma 3, or even what is RAID0 and you are not interested in knowing it.

Peace.

 

It has become clear that you jump to conclusions quickly. I'm perfectly aware of what RAID does, and I do have interest in it.  Anyway, I'm done beating a dead horse.  

 

Peace out.

[I give up 152]

It has become clear that you jump to conclusions quickly. I'm perfectly aware of what RAID does, and I do have interest in it.  Anyway, I'm done beating a dead horse.  

 

Peace out.

If you know RAID 0 and Arma 3, you should know that the game streams data directly into Hard Disk (not in to RAM as some believe), streams in a continuous and peristent way things like terrain or new objects.

 

It means that when we have a fast read/write the game becomes incredibly smooth and fast, also the LOD transition/load becomes incredibly fast and smooth allowing to have a stable fps permanently and under every situation, even when we fly at high speed and low altitude and we are approaching high density scenarios like huge towns in Altis, tons of vehicles or even AI.

 

When we have the necessary read/write speed in our Hard Disk, our CPU and GPU starts to dictate the parameters in matters on performance because is when our CPU and GPU does not have to wait for Hard Disk to perform the operations.

Is 500MB/s of read/write speed enough to achieve this? NO. We need more. How we get it? By the methods described above.

 

Now some may say, but this does not increase fps, if anything it will make the game more smooth with stable fps. WRONG. In fact it also helps with fps and with performance in general. Why?

Because a fast Hard Disk read/write speed allow to extract more "juice" from CPU and GPU, these 2 pieces of hardware will work close of its real capabilities in a consistent and persistent way.

Under these circumstances the often described as "CPU bottleneck" (that causes the often described as "GPU bottleneck") is barely non existent.

Our cpu (and cores) will have more and constant usage since it does not need to wait an eternity for Hard Disk to perform the operations

Our GPU will have more and constant usage since it does not need to wait an eternity for CPU to perform the operations because CPU is waiting for Hard Disk.

 

So, in fact faster Hard Disk read/write not only makes the game faster and smooth but also will give more fps, obviously if we have CPU/GPU to achieve that.

At bottom, greatly reduces the called cpu bottleneck (and gpu).

 

These are the facts and what really happens, all the rest are fairy tales.

[jumpinghubert 49]

 

These are the facts and what really happens, all the rest are fairy tales

Time for a benchmark to change your fairy tales into proofed facts like the others you call fairy tail tellers? LOL

[NeMeSiS 11]

If you know RAID 0 and Arma 3, you should know that the game streams data directly into Hard Disk (not in to RAM as some believe), streams in a continuous and peristent way things like terrain or new objects.

 

It means that when we have a fast read/write the game becomes incredibly smooth and fast, also the LOD transition/load becomes incredibly fast and smooth allowing to have a stable fps permanently and under every situation, even when we fly at high speed and low altitude and we are approaching high density scenarios like huge towns in Altis, tons of vehicles or even AI.

 

I am pretty sure that the game keeps most stuff in RAM most of the time, and loads terrain in/out whenever needed, but that won't be a huge amount of data. I have installed the game on an SSD as well as a normal HD and i cant really report a difference in FPS. LODs may load faster, maybe less microstutter (though i am not sure about that). But the game certainly isn't waiting for data from the HD until sending the next frame to the screen, our FPS would be atrocious.

[I give up 152]

Time for a benchmark to change your fairy tales into proofed facts like the others you call fairy tail tellers? LOL

You look a nice guy and because of that I will give you a few tips and you can do the benchmark (for free).

 

But first try to run Arma 3 and load Altis with pagefile disabled.

If you can do it without errors then you dont even need a SSD, because the data will be loaded in to HDD only one time when you open/load the game/island for the first time, after that is all about RAM. Doing this successfully will place you in the same level of the guy that have discovered the powder.

 

If you cant, then is because Arma 3 in fact needs and use system pagefile for file swap. In this case the myth, that the game do not stream (textures/lod/terrain/objects) from Hard Disk swapping files between system pagefile and physical memory, is dead.

 

If you are not the guy that have discovered the powder, then you can do the following and benchmark (for free).

To perform this I will assume that you have at least 12GB of Ram installed (the minimum required to run the game decently on very high/ultra with having a 4GB Vram gpu).

Grab this free tool (the freeware).

http://www.radeonramdisk.com/software_downloads.php

Will allow you to set a 4GB ramdisk (for free). Set a ramdisk with 4GB.

Then set the size of your system page file let's say to 3.9 GB and move it in to the ramdisk that you just created.

Now that your system pagefle is located in your ramdisk, the swap file (between physical memory and pagefile) will be significantly faster since the ramdisk is about 100x faster when comparing with a SSD

Last step, load Altis and perform the benchmark.

Come back later with results.

 

In my case I have a noticiable performance increase and without spending a cent.

Let's see your case.

[St. Jimmy 272]

I've 2x4GB sticks and my MB is dual-channel. Is there any disadvantage or advantage to buy two more of those sticks so I'd have 4x4GB = 16GB? Does the speed or bandwith get some decrease or something?

[rundll.exe 12]

I didn't want to respond to this, but I find your claims very hard to believe. I downloaded HDD Sentinel and tested the SSD usage while running a benchmark, flying over altis, and a normal ground mission with some action (helicopters etc)

The results: 

 

A maximum transfer rate of 8MB/s!  That's next to nothing. Actually, Arma writes a constant of 1MB/s (even in the menu).

That means the disk activity is at most 2%, and you say I need 2 SSD's in RAID0?  Why would that be?

 

My system: 

i7-4770K, 8GB RAM , GTX 770 3GB, Intel SSD 330 Series 180GB

[I give up 152]

@ rundll.exe

According to your readings and benchmark we can safely say that getting a SSD for Arma 3 is a pure waste of money.
Every HDD (even the lousy ones) can have read/write speeds of 40 MB/s which according to your readings is more than enough for Arma 3....

[jumpinghubert 49]

@bratwurste

Wrong order. Please test your own hypothesis at first :p.

 

Some details to me: when I changed arma2-3 from HDD to SSD (swap file on ssd too) I got less short framedrops/stuttering, thats true - but zero advantage in fps/benchmarking. In the past some guys tested the ramdisc-method (up to 16GB on ramdisc) with arma2 and got zero performance gain compared with ssd. Sadly I´ve only 8gb of ram.....

 

Whats about a new thread with your hypothesis AND some testing/benchmarking? I am very interested.

[I give up 152]

_{Arma 2 and Arma 3 have different architectures  in matters of file caching, swap file and memory management
The impact of SSD vs HDD in both games is not (and cant be) the same.
Any particular reason for you to put both games in the same "bag"?
Copy and past of images showing whatever  have a value close to zero. Do yourself the benchmark, or not.}

[jumpinghubert 49]

I found this https://forums.bistudio.com/topic/139258-how-to-install-arma-3-on-a-ramdisk/

 

Seems you are wrong. Regarding microstutters SSD or ramdisc will help compared to a HDD but not performance wise (more frames per second). Same behavior like arma2....the same bag regarding to this.

[dasa 20]

found something that may interest some people here while looking through skylake reviews

 

they test ddr3 & ddr4 performance in arma

http://www.hardware.fr/articles/940-5/cpu-ddr4-vs-ddr3-pratique.html

[St. Jimmy 272]

found something that may interest some people here while looking through skylake reviews

 

they test ddr3 & ddr4 performance in arma

http://www.hardware.fr/articles/940-5/cpu-ddr4-vs-ddr3-pratique.html

i5 5675C and i7 5775C are also very interesting results. If those CPUs can overclock nicely from the 3,6GHz then holy cow! At least C is the same as K so it's unlocked. That's a dam nice article!

 

What have those 5675C and 5775C eaten when they're that powerful compared to many others? Are they all running the same speed RAM in those results? They're running @ 3,6-3,7GHz compared to the i7 4790K @ 4,4GHz and even leaving it behind by ~3fps, the difference is pretty big.

 

/ I haven't heard about those CPUs so that was bit shocking to see that

 

//Aaah Broadwell is DDR3L. Hmmmm that's bit of a backlash there but also it's bigger surprise then because it likely isn't running anything higher than CL9 1600MHz DDR3. Well it can get behind in the future then because of that.

[I give up 152]

I am also waiting for next intel socket to assemble my new pc exactly because DDR4.
With the socket 1150 we are stuck at max memory frequencies of 2133 (with luck), also its funny to see some people with 2400 (and above) when the cpu memory controller does not have support for it, if anything it will brings instability and performance degradation (and money wasted).
With the upcoming socket 1151 socket is expected to have support for DDR4 operating at frequencies above 3000 mhz (close to 4000) and with these we can have, in fact, noticeable improvements in matters of performance with Arma 3.
Anxiously waiting.

[jumpinghubert 49]

6600k @3,9Ghz: 32,6fps

5675C @3,6Ghz: 36,5fps

 

???

 

My good old socket 1155 gives me rockstable 2666Mhz ramspeed (my signature) so I hope ddr4 is significantly faster.

 

@bratwurste

foreseeable you ignored my link with crystalclear falsification of your hypothesis. Now lets make new assumptions (noticeable improvements only with ddr4)....but wait a moment..... look into this thread there will you find noticeable improvements (up to 15%)  :D

Now you are on my ignorelist, bratwurste.

[St. Jimmy 272]

As long as the latency remains high there's no real gain. So don't wait for the MHz increases if they're achieved with making the latency worse... CL9 2133MHz DDR3 is at the par with CL15 3000MHz DDR4. If they can lower the latency by 2 from DDR4 and keeping the MHz the same or better, then it would get very interesting.

[I give up 152]

As long as the latency remains high there's no real gain. So don't wait for the MHz increases if they're achieved with making the latency worse... CL9 2133MHz DDR3 is at the par with CL15 3000MHz DDR4. If they can lower the latency by 2 from DDR4 and keeping the MHz the same or better, then it would get very interesting.

That's a good point and you are partially correct.

When we compare DDR 3 (frequency vs latency) we easily conclude that the performance of 1866 CAS 8 is basically the same of 2133 CAS 9. Here you are correct.

But we are now speaking of DDR 4 which is a completely new architecture and looking at it we see improvements under several aspects which may change the math about frequency/latency as we know it.

Here some of basic changes, in matters of architecture improvements

http://www.micron.com/products/dram/ddr3-to-ddr4

Now we just need to wait to see how the new Intel CPU's will interact with DDR4 and its frequencies, from my side I believe there will be improvements in matters of performance, DDR4 specifications leads me to believe.

[dasa 20]

 

What have those 5675C and 5775C eaten when they're that powerful compared to many others? Are they all running the same speed RAM in those results? They're running @ 3,6-3,7GHz compared to the i7 4790K @ 4,4GHz and even leaving it behind by ~3fps, the difference is pretty big.

 

 

They have a 128m L4 cache​ which arma loves

 

Here is another review with overclocking

http://translate.googleusercontent.com/translate_c?depth=7&hl=en&rurl=translate.google.com&sl=pl&tl=en&u=http://pclab.pl/art65154-26.html&usg=ALkJrhj0alS52oor9_1qlHHzwwoFrn2_1g

 

[St. Jimmy 272]

OK Skylake is the winner then because it isn't limited by RAM. Skylake can likely squeeze couple fps more with nice sticks and Broadwell remains with their DDR3L sticks. The RAM is likely the reason why Broadwell doesn't scale as well as Skylake.

That 8fps extra and likely more with the newer generations is already teasing a bit :P Maybe I'll upgrade after 2 years.

[dasa 20]

If this is accurate then the l4 cache while much faster than slower ram may not be all that much quicker than high speed ram so ddr4 may surpass it one day

I guess that also explains why the performance drops off a bit in the second review as they were running higher ram speeds

http://forums.aida64.com/topic/2864-i7-5775c-l4-cache-performance/​

[I give up 152]

Would be nice to know the hardware configuration used in this benchmark.

 

In matters of ram I believe that with Skylake having nice RAM modules with frequencies above 3000 Mhz we will see even a bigger gap in matters of performance when compared with Haswell, in Arma 3.

With Skylake the memory bandwidth is around 30% faster (in relation to Haswell), the ram frequency gains a new meaning and relevance.

 

Also thinking in Arma 3 and for those who are assembling a new pc with Skylake (like me) maybe should consider a PCI Express SSD, currently we can get one with 240GB for the same price of a decent SSD (sata), the advantages are more than many, here some basic concepts about both solutions.
http://blog.fosketts.net/2013/06/12/pcie-ssds-fast/

 

I am pretty sure that Arma 3 will be smoked having a decent hardware configuration based on Skylake/DDR4.