It's all about the CPU*

[Brun]

*not literally, just more than you may expect.

 

In June I upgraded my vanilla 1080 to a Ti. I always suspected VR was particularly CPU dependent so wasn't expecting the 30% performance increase the card was capable of, but the reality was very little increase at all.

 

Last weekend I replaced my 6700k (not a bad gaming CPU @4.8GHz) with a 9900k and the difference is incredible. Only tried DCS for the first time tonight and haven't had much chance to mess with settings, but I've got pixel density at 2 and draw distance on ultra. Free flight in and around Vegas is completely smooth, even when looking to the side which was almost nauseating before.

 

For anyone considering replacing a decent graphics card with something like a 2080 or 2080Ti, it seems well worth bearing in mind that you won't get the benefit unless your CPU is up to the job. It's also probably more about clock speed than core count so I reckon the likes of an 8700 would also do a fine job.

 

Wish I'd done some more specific tests before upgrading, but I suspect 9900/1080 would run DCS VR much better than 6700/1080Ti.

[VirusAM]

It's all about the CPU*

 

out of curiosity...dis you overclock the CPU?And what settings do you have?

 

mine is a t 5GHz just enabling MCE in bios (some sort of factory auto overclock) but i am not so satisfied in VR....yes its better then before but i am still struggling to have 45 fps in every condition....i did a passmark benchmark to understand if there is something wrong in my system...but the numbers are where they should be....99 percentile both multicore and single core performance...as i should expect from the fastest desktop cpu on the market....

To mantain 45fps in every condition i have to put visible range to medium shadows to flat only civil traffic to low...pd is to 1.0 in Dcs but 60% in steamvr (with general steamvr screen resolution on auto where it puts to 200%)

using the wmr motion reprojection indicator i see that most of the time i am cpu limited (light blue indicator) and sometimes also gpu limited (dark blue indicator) i almost never see the green indicator (90fps)...i tried an experiment putting the hmd (lenovo explorer) at 60hz in windows mixed reality app in windows...and so reducing the workload...then i have almost always the green indicator (60fps) and when on ground/carrier or in intense scenes 30 but never under it...and this is with much higher settings (shadows to low water high etc)...

its tempting to mantain these settings but obviously 60fps is different from 90 in VR.

 

my specs are in the signature...i was not expeting 90fps at high settings but 45 in every condition yes....

For all the money i spent to update the system i am quite disappointed and hoping in DCS multithread/VR optimization and Vulkan runtime will come early (but it is not the only famous flight sim with the same issues).

Edited November 6, 2018 by VirusAM

[dburne]

Still waiting on my i9 9900k I have on Pre-Order with Amazon - should be soon now.

 

Sounds promising, looking forward to my new build coming up.

[uri_ba]

It's CPU and memory speed.

 

Have a look.

https://pit.uriba.org/uriba/vr/8700k-overclock-performance-in-dcs-2-5/

[VirusAM]

i have 3200MHz clocked ram....5GHz clocked cpu and os and dcs loaded on nvme ssd x4...so i don’t know really how to have a faster rig....still much unsatisfied

[uri_ba]

1. "Sync all cores"

2. VR shaders mod

3. Reduce CPU intensive stuff (shadows most promenantly)

 

I see no problem 90% of the time, however, on ocassions, there are microstutters I can really figure out. My guess is first time use of a shader.

[Sn8ke_iis]

Chipset, bandwidth between components, DD4 memory, PCIe lanes, new motherboard. All variables besides the CPU change.

 

DCS does like single thread performance, newer processors also have different architecture, not just higher clocks and smaller transistors. 9900k has fast single thread performance, but so does a 9600k at less than half the cost. If you are running multiple GPU's or multiple Nvme drives be aware of available PCIe lanes.

[Mr_sukebe]

OP, overall, I think you’re right, i.e. that it’s the CPU loading that’s currently killing us.

 

 

For example, I was testing this week. When comparing my frame rate and resource usage between say running my own mission and a far more complex server based squadron mission, when online, not only did my frame rate drop, but also my GPU usage. That implies to me that playing online is NOT adding to the stress on my GPU, but clearly on the CPU, which is proving to the bottleneck.

 

So my current belief is:

- CPU speed is absolutely the bottleneck (unless you’re running a seriously ancient GPU), also explaining why people with MUCH newer CPUs are not getting say double my frame rate

- GPU resource facilitates the level of eye candy, e.g. PD values in VR, MSAA, textures and it appears that you can play with these to your hearts content when you know what GPU free resources you normally have available

- System RAM volume affects how much can be “pre-loaded” with it’s slider. I’ve been playing with that recently and am now over 100k with no issues. Very easy to check RAM usage

- VRAM volume and Ram/motherboard/SSD speeds affect the level of stutters. My system is ancient, using DDR3, probably resulting in more stutters than I’d like. My guess is that DDR4 with NVME SSD would be MUCH smoother and reduce troughs in performance

 

Some thoughts from the above:

- I was playing with CPU core affinity last night using Processlasso. DCS is primarily single core, but with some secondary core usage. So what I tried doing was giving is core 0 & 1 to DCS and running the rest of my applications on my other two cores (including teamspeak, Oculus, MSI afterburner, processlasso itself, tacview, voiceattack etc). Core 3&4 were still not unduly taxed running everything else. At this stage, not really sure it helped, but seemed worth a try and I’ll be testing more over the next few days. Certainly didn’t seem to result in any negative consequences

- Vulkan will apparently not only spread it’s resource usage across CPU cores, but also reduce CPU resource usage (certainly from the online data that I’ve read). That can only be a good thing and should help

- I’m also hopeful that the planned standalone server will de-couple the graphics engine from the game “logic”, possibly allowing the two to be run on separate CPU cores.

[boedha68]

*not literally, just more than you may expect.

 

In June I upgraded my vanilla 1080 to a Ti. I always suspected VR was particularly CPU dependent so wasn't expecting the 30% performance increase the card was capable of, but the reality was very little increase at all.

 

Last weekend I replaced my 6700k (not a bad gaming CPU @4.8GHz) with a 9900k and the difference is incredible. Only tried DCS for the first time tonight and haven't had much chance to mess with settings, but I've got pixel density at 2 and draw distance on ultra. Free flight in and around Vegas is completely smooth, even when looking to the side which was almost nauseating before.

 

For anyone considering replacing a decent graphics card with something like a 2080 or 2080Ti, it seems well worth bearing in mind that you won't get the benefit unless your CPU is up to the job. It's also probably more about clock speed than core count so I reckon the likes of an 8700 would also do a fine job.

 

Wish I'd done some more specific tests before upgrading, but I suspect 9900/1080 would run DCS VR much better than 6700/1080Ti.

 

 

 

Yep, I imagine. Tests from the past say that DCS is highly CPU demanded.

 

 

core 1 is the highest demand for the program

core 2 is for audio

 

 

9900k is the highest value on this moment. Is kicking in for 600 dollars ????

 

 

What about AMD? 7nm is coming, rumour is that the new processors come at 5.3 Ghz. And i believe not for 600 dollars. :D

 

 

Besides that Vulcan Api can gain us huge performance. Only the gap in time and impatience of every individual can hold this back. :D

 

 

Well the race is on. Next year i build a new pc... but on what components? Time will tell.. :D

[toutenglisse]

I run fps limitation (mostly VR, so 45/90, or 60hz screen).

 

The only times I loose fps is when GPU touch 98-99 or 100% usage. So no CPU bottleneck seen.

 

On other hand if I uncap fps to see what limits system's fps, I only see GPU at 100% some times, and most other times nothing limits fps (nothing at 100% usage) wich tends to mean nothing logical.

 

So I personnaly won't be that sure anymore about what gives perf in DCS.

[Mule]

It's all about the keyboard.

[javelina1]

It's all about the keyboard.

 

 

actually, most problems usually lie somewhere between the keyboard and the chair.... ;):music_whistling:

[sea2sky]

actually, most problems usually lie somewhere between the keyboard and the chair.... ;):music_whistling:

 

 

yeah, that thin gasket is a usual problem :)

[mdee]

I run fps limitation (mostly VR, so 45/90, or 60hz screen).

 

The only times I loose fps is when GPU touch 98-99 or 100% usage. So no CPU bottleneck seen.

 

On other hand if I uncap fps to see what limits system's fps, I only see GPU at 100% some times, and most other times nothing limits fps (nothing at 100% usage) wich tends to mean nothing logical.

 

So I personnaly won't be that sure anymore about what gives perf in DCS.

 

Everything matters. GPU, CPU speed (single core clock atm as DCS is not multithreaded extensively) ram speed, ram size, SSD speed (for streaming performance). Running ancient (<7700k) CPU with current GPU is no better than running latest CPU with 980ti.

 

We are running very complex simulation engine, with high graphical fidelity. Don't expect miracles. It's not "poorly optimized" it's just demanding. It can be better for sure, better multithreading will help, but this isn't a simple thing to do and may take a very long time. Current DCS performance is really decent across wide range of hardware, graphics settings are extensive and can be fine tuned to even very old hardware and go beyond currently available hardware . ED is very honest about system specs too. But people will always try to crank it up to the max on their 3 years old laptop and come back saying "not optimized". It happens with every PC game which has graphics settings.

 

Vulcan, heh.. good luck with that :megalol::megalol: even if it's going to happen, the performance gains are not what people think they are.

Edited November 10, 2018 by mdee

[SharpeXB]

Running ancient (<7700k) CPU with current GPU is no better than running latest CPU with 980ti.

Someone will have to correct me but are PCIe lanes even limiting the 2080Ti bandwidth?

Currently there is so much overhead in those lanes and PCI slots even SLI Titans won’t utilize it all. In Theory your CPU lanes could bottleneck graphics cards

Between 4th and 9th gen i7s there’s only been about a 15% gain in performance. But older CPUs will certainly stop up DCS.

 

Vulcan, heh.. good luck with that :megalol::megalol: even if it's going to happen, the performance gains are not what people think they are.

I would fully expect the Vulkan engine to take the same amount of time to develop as the current DCS engine, (EDGE?) meaning about 5-6 years. By that time hardware and VR will have changed so much it’s impossible to say what performance benefits will be. VR will probably be extinct in the market by then :music_whistling:

Edited November 10, 2018 by SharpeXB

[Sn8ke_iis]

PCIe lanes are relevant to number of GPU's and NVMe drives. Check the fine print on your CPU and MB specs before you order components for a build. i.e. there may be available lanes for 2 GPU's and one NVMe drive or 2 drives and 1 GPU. But not enough lanes for 2 GPUs and 2 NVMe drives. Or one of your expansion slots will be disabled, etc. Depends if you need the slot.

 

IIRC Intel has cut some PCIe lanes on the lower end 9000 series chips as a cost saving measure. Whether you need the lanes depends on component choice/configuration. FYI for people doing a new build this shopping season.

[uri_ba]

PCIe lanes are relevant to number of GPU's and NVMe drives. Check the fine print on your CPU and MB specs before you order components for a build. i.e. there may be available lanes for 2 GPU's and one NVMe drive or 2 drives and 1 GPU. But not enough lanes for 2 GPUs and 2 NVMe drives. Or one of your expansion slots will be disabled, etc. Depends if you need the slot.

 

IIRC Intel has cut some PCIe lanes on the lower end 9000 series chips as a cost saving measure. Whether you need the lanes depends on component choice/configuration. FYI for people doing a new build this shopping season.

 

I don't remember exact numbers, so take it with a grain of salt.

 

Intels main line of CPUs have 16 PCIe lanes directly to the CPU. those are usually hooked up directly to the GFX cards slots. sometime an NVME slot is sitting there too and the MOBO will control it. as stated before, 1 GPU on X8 and two NVME on X4 or two GPUs on X8. but you will probably won't be able to run two cards and an NVME on THOSE lines.

 

However, the CPU usually has dedicated PCIE lanes that are managed by the chipset. I think it's another X8 (these are the numbers I'm not sure about). those lanes are shared between all the periferials. USB, SATA and also sometime M.2 slots. in these cases the MOBO might disable SATA ports if you hook up an M.2 drive. but that is completly up to the chipset and you need to consult your MOBO manual.

 

AMD Ryzens are different, they have the same X16 PCIE3 for GPU, another (i thing) X8 for the chipset and addtional x4 PCIE2 dedicated for M.2 NVME.

 

Higher end CPUs can go up to 64 PCIe lanes (Threadripper).

[SharpeXB]

But running a graphics card x8 for example in SLI where you have two cards at x8 instead of one card at 16x does not mean the graphics card is running half speed. The x8 PCIe is more than enough bandwidth for the card.

[uri_ba]

But running a graphics card x8 for example in SLI where you have two cards at x8 instead of one card at 16x does not mean the graphics card is running half speed. The x8 PCIe is more than enough bandwidth for the card.

 

waaay more then enough.. and SLI comms are going via the Bridge anyways.

[Bob_Bushman]

And DCS as with most VR titles does not use more than one GPU anyways so rather a moot thing to worry.

 

I noticed myself a rather staggering boost in performance upgrading from an i7 4790k to my i7 8700k.

Not in DCS but in pretty much all other VR titles.

 

Yes, DCS isn't multithreaded, but VR is, so the tracking and software side of your VR plattform gets a benefit over single core performance with a newer multithreaded CPU.

And DCS is an oddity in this regard, pretty much every other game takes adavantage of more than eight threads.

And as soon as DCS goes to vulkan so will it.

I don't know when that will be, three months or three years from now, there is no question single core performance is a secondary these days.

 

I honestly find it quite funny the amount of threads I am seeing other places about "I am running a seven year old i7, why is VR bad? And why won't it support windows 7?" threads.

Edited November 15, 2018 by Bob_Bushman

[metzger]

From my testing not a single CPU core goes above 50-60% but I can easily hit GPU 100%, when I loose frames or stutters, it is usually GPU 100%. In comparison, running other more civil aimed sim with VR, I can clearly see Core 0 100% and gpu stays at 30%.

 

I think for DCS, AI is a major CPU consumer, especially ground units, so if you load a mission with lots of AI action, then you might have a CPU bottleneck but by only graphic changes I can't Utilize the cpu before my GPU hit 100

 

I7 8700K 4.7ghz, 32gigs DDR4-2400, NVMe TLC M.2 Solid State, 1080TI.