Upgraded to i9 9900K from i7 8700K and can see no difference in VR

[hansangb]

 

--snippage--

 

I would say for anyone thinking of upgrading don't get hopes too high, because for as long as we cannot get above 90 constantly, the margins are going to be very fine indeed.

 

 

Although 90 is buttery smooth, I got so used to CV1's 45FPS (with ASW) that it's fine as far as I'm concerned. The fact that I'm getting 90FPS at times with Reverb is just bonus points.

[pimp]

On the PC Hardware thread, it gets discussed weekly. DCS today is bound by single core performance...

I checked my affinity settings for DCS and all the cores were checked. So since DCS is automatically doing this now, how can it be a single core performer? Or am I missing something?

[Lucky]

I was about to buy an i9 9900k for almost $500.00. But after reading some reviews it’s only marginally better then my i7 8700k so I bought the RTX 2080 instead.

[dburne]

I was about to buy an i9 9900k for almost $500.00. But after reading some reviews it’s only marginally better then my i7 8700k so I bought the RTX 2080 instead.

 

Yeah probably a smart move, 8700k - 9900k not that big an advance, and you also already have DDR4 ram.

I upgraded from i7 4820k with DDR3 ram, to the 9900k with faster DDR4 ram, along with a new MB platform, and the difference for me was substantial.

[imacken]

I checked my affinity settings for DCS and all the cores were checked. So since DCS is automatically doing this now, how can it be a single core performer? Or am I missing something?

Yep, it would be good to get a definitive explanation for this.

Personally, I am very confused by all the posts on these forums that say 'DCS is single core'. followed by endless, 'no it's not', 'yes it is' type responses.

[Svsmokey]

Yep, it would be good to get a definitive explanation for this.

Personally, I am very confused by all the posts on these forums that say 'DCS is single core'. followed by endless, 'no it's not', 'yes it is' type responses.

 

Yep , that argument has been running at least as long as i have been on these forums . Glazed eyes after a bit reading the esoterica of each side defending their respective positions .

[imacken]

Well, as my signature reflects, I've returned the 9900K. Now back to my trusty 8700K!

Doing even a little O/C on the 9900 sent the temps way too high for my liking.

Even the Intel utility couldn't do much with it.

[etherbattx]

Yep, it would be good to get a definitive explanation for this.

 

Personally, I am very confused by all the posts on these forums that say 'DCS is single core'. followed by endless, 'no it's not', 'yes it is' type responses.

 

different users test different things with different options in different scenarios.

 

and some users use different names for different technical details... so yes it can be confusing.

 

short answer is, all windows programs package work into ‘threads’. some use 3, some use 200+. these threads are then scheduled onto cpu “cores” to execute and do the work.

 

if most of your program “work” is done in a single thread, then that work can only be done when that thread is assigned to a core (one thread -> one core).

 

DCS does a LOT of work on 1 thread and a bunch of work on 2 other threads.

 

in my MP test scenario, those 3 threads gobbled up over 73% of the entire CPU time used by DCS.

 

ideally, if those threads were always ready to execute, they would run on 3 cores simultaneously.

[imacken]

different users test different things with different options in different scenarios.

 

and some users use different names for different technical details... so yes it can be confusing.

 

short answer is, all windows programs package work into ‘threads’. some use 3, some use 200+. these threads are then scheduled onto cpu “cores” to execute and do the work.

 

if most of your program “work” is done in a single thread, then that work can only be done when that thread is assigned to a core (one thread -> one core).

 

DCS does a LOT of work on 1 thread and a bunch of work on 2 other threads.

 

in my MP test scenario, those 3 threads gobbled up over 73% of the entire CPU time used by DCS.

 

ideally, if those threads were always ready to execute, they would run on 3 cores simultaneously.

Hmm... see attachments. With and without DCS running.

[Headwarp]

Hmm... see attachments. With and without DCS running.

 

In steamVR settings, on the "Video resolution" page, whats your resolution per eye?

 

I'm personally getting much higher framerates since the VR optimizations and when I lowered my PD to 1.0 I'm getting 90fps over Caucasus for the most part with no shadows and msaa@2x

 

Run a repair.. reinstall directx all that. Ensure chipset drivers are up to date and the like, and make sure "prefer maximum performance" is set in NV Control panel.

 

Also make sure that beefy GPU has some aggressive fan curves they run better when they're cooler.

[fearlessfrog]

Hmm... see attachments. With and without DCS running.

 

Single thread apps can be spread around cores, it's what the Windows 10 thread scheduler does on a shared cached chipset that supports it, which is what you are seeing. It didn't used to be like this. It allows the CPU to run more evenly and cooler rather than a core0 pegged at 100%. The newer Intel (and AMD) designs have a shared cache between cores that makes this 'core spread/hop' have practically no run-time impact.

 

If you find that hard to believe (as a guess) and think that perhaps DCS is now a fully multi-threaded changed in secret or something, then please try disabling your cores in your BIOS and track the linear relationship between load and cores left. Post your results.

 

As for the general comparison between an i7 8700k and a i9 9900K then you're talking about a 6% difference on a single threaded app. Depending on how you systems is I can believe that you didn't see a dramatic difference in DCS. I personally find an i9 overclocks much more nicely than a i7 8700, but perhaps you got unlucky in the chip lottery or weren't sure how to best use it for that or your cooling etc. Cool and stable at 5 GHz is very common for an i9 I think.

 

The misunderstanding of what a 'bottleneck' might be happening as well, in that there will *always* be some sort of limiting factor, it just shifts around as people do different settings/things.

 

To say 'DCS is CPU core speed limited' is only true if the CPU is the limiting factor in how you run DCS. If you use VR and have lots of MSAA and super high resolutions with millions of vertices being drawn then this isn't true - your GPU will be the 'bottleneck'.

 

To say 'DCS is GPU limited' is also not true if you use a very high visibility/object setting with calculations done on the CPU side. A mission with many units or perhaps even a MP session can easily make the CPU the limiting factor. The only other factor in this side is that currently some things *could* be put to another thread but the use of DX11 prohibits it, as in the thread you initialize the render with has to be the one issuing instructions. This makes it hard to change program architecture over to something not designed initially around multiple threads. The Vulkan API isn't 'faster' than DX11 but is at a lower level to be able to do this sort of change, hence the look at that.

 

So, yeah, you're going to get lots of different people saying different things. In a way, some of them can all be right.

[imacken]

Single thread apps can be spread around cores, it's what the Windows 10 thread scheduler does on a shared cached chipset that supports it, which is what you are seeing. It didn't used to be like this. It allows the CPU to run more evenly and cooler rather than a core0 pegged at 100%. The newer Intel (and AMD) designs have a shared cache between cores that makes this 'core spread/hop' have practically no run-time impact.

 

If you find that hard to believe (as a guess) and think that perhaps DCS is now a fully multi-threaded changed in secret or something, then please try disabling your cores in your BIOS and track the linear relationship between load and cores left. Post your results.

 

As for the general comparison between an i7 8700k and a i9 9900K then you're talking about a 6% difference on a single threaded app. Depending on how you systems is I can believe that you didn't see a dramatic difference in DCS. I personally find an i9 overclocks much more nicely than a i7 8700, but perhaps you got unlucky in the chip lottery or weren't sure how to best use it for that or your cooling etc. Cool and stable at 5 GHz is very common for an i9 I think.

 

The misunderstanding of what a 'bottleneck' might be happening as well, in that there will *always* be some sort of limiting factor, it just shifts around as people do different settings/things.

 

To say 'DCS is CPU core speed limited' is only true if the CPU is the limiting factor in how you run DCS. If you use VR and have lots of MSAA and super high resolutions with millions of vertices being drawn then this isn't true - your GPU will be the 'bottleneck'.

 

To say 'DCS is GPU limited' is also not true if you use a very high visibility/object setting with calculations done on the CPU side. A mission with many units or perhaps even a MP session can easily make the CPU the limiting factor. The only other factor in this side is that currently some things *could* be put to another thread but the use of DX11 prohibits it, as in the thread you initialize the render with has to be the one issuing instructions. This makes it hard to change program architecture over to something not designed initially around multiple threads. The Vulkan API isn't 'faster' than DX11 but is at a lower level to be able to do this sort of change, hence the look at that.

 

So, yeah, you're going to get lots of different people saying different things. In a way, some of them can all be right.

I don’t think anything about whether DCS is ‘fully multi threaded’ or not. I just patiently listen to what others say! All I was pointing out is that it is certainly not true to suggest that DCS only seems to hit a single core. Look at the graphs.

Re 9900X, my experience over the years is that although overclocking is fun, it is ultimately not worth the grief. There is very little difference between 4.8 and 5.0 in practice, and certainly, the very marginal improvement compared to the 8700 was not worth £50, far less £500!

[fearlessfrog]

I don’t think anything about whether DCS is ‘fully multi threaded’ or not. I just patiently listen to what others say! All I was pointing out is that it is certainly not true to suggest that DCS only seems to hit a single core. Look at the graphs.

Re 9900X, my experience over the years is that although overclocking is fun, it is ultimately not worth the grief. There is very little difference between 4.8 and 5.0 in practice, and certainly, the very marginal improvement compared to the 8700 was not worth £50, far less £500!

 

But you get the 'why' on how DCS can still be a single threaded (more or less) app and you still have graphs like that on Windows 10 and a Coffee Lake architecture or later, right?

 

The i9 is expensive as top of the range premium, and if you don't like to overclock or run things that use the cores (video editing etc) then it doesn't represent very good value for you over an i7 8700K. You are literally paying extra for that 'K' CPU feature of the unlocked multiplier *for overclocking*, so paying for it and not using it is never going to be good value.

 

A lot of these things could be sorted out before purchase, but it is good you can return things anyway and get your money back. By sharing experiences on forums like this you'll probably help others in a similar boat.

[imacken]

But you get the 'why' on how DCS can still be a single threaded (more or less) app and you still have graphs like that on Windows 10 and a Coffee Lake architecture or later, right?

 

The i9 is expensive as top of the range premium, and if you don't like to overclock or run things that use the cores (video editing etc) then it doesn't represent very good value for you over an i7 8700K. You are literally paying extra for that 'K' CPU feature of the unlocked multiplier *for overclocking*, so paying for it and not using it is never going to be good value.

 

A lot of these things could be sorted out before purchase, but it is good you can return things anyway and get your money back. By sharing experiences on forums like this you'll probably help others in a similar boat.

 

Sorry, confused. Does the 8700K not also have unlocked multiplier?

[fearlessfrog]

Sorry, confused. Does the 8700K not also have unlocked multiplier?

 

It does. Makes overclocking much easier, so for those that do that, worth the extra money you pay for that feature.

[imacken]

It does. Makes overclocking much easier, so for those that do that, worth the extra money you pay for that feature.

 

Sorry, I thought you were suggesting this was a feature of the 9900K, and one of the reasons for the price premium.

[nrosko]

Both show in fpsVR.

 

 

Thanks but we've gone through this in other threads, and it's almost impossible to compare with different settings, SS, HMDs, etc.

Still, if we take Wags' test mission, i.e. flight over Tbilisi, then I get between 16 and 18ms frame time at first, then dropping to 20-22ms if I dive towards the city when over it. It's pointless me quoting fps as I am on 45fps almost regardless of what else is going on! (There is another thread on this very subject, amd yes, I have motion smoothing switched off for this test.)

If I drop out MSAA, and all shadows, then I go to 12ms at start and 14-15ms over Tbilisi.

However, my point is that I am not seeing any improvement after spending £500 on this 9900K compared to my 8700K!

 

 

 

Are you locking at 45 even when looking to the sky and/or turning MSAA off?

[imacken]

Are you locking at 45 even when looking to the sky and/or turning MSAA off?

 

Yep!

[nrosko]

Yep!

 

Make sure your Nvidia setting are set to default & your memory profile in your bios is set to the correct profile as this resolved the issue for me I was also stuck at 45.

 

Sorry if you already checked this.

 

Something seems off.

I'm easily getting 70-90fps in that benchmark with a lesser system although i have to turn MSAA off.

[imacken]

Settings are all good.

It's just a fps display thing and I don't know why it's happening. As you know, there is another thread on that particular topic that hasn't come to any conclusion.

Still, I am very happy with performance. I run between 12 and 20 ms in general, and that is about 50 to 83 fps regardless of what the fps counter says.

Also, bear in mind my VP is pushing 70% more pixels around than your Rift and that makes a big difference.

[nrosko]

Settings are all good.

It's just a fps display thing and I don't know why it's happening. As you know, there is another thread on that particular topic that hasn't come to any conclusion.

Still, I am very happy with performance. I run between 12 and 20 ms in general, and that is about 50 to 83 fps regardless of what the fps counter says.

Also, bear in mind my VP is pushing 70% more pixels around than your Rift and that makes a big difference.

 

 

I'm running it at 1.5 PD are you able to push yours up? interested as incoming Index.

[RealDCSpilot]

My latest findings on optimizing DCS is to set DCS.exe's affinity in Process Lasso only on the 2 best cores of my CPU (they are marked in the bios with an *) This way the thread(s) will be kept on those cores. Another setting is to force motion smoothing in SteamVR settings. Because it will never reach 90 fps anyway, this prevents SteamVR's reprojection algorithm from searching for opportunities to get to 90 fps (saves some CPU workload).

[Vullcan]

Why not force all the re-projection off completely? Should gain a few FPS there and as long as you can maintain a decent framerate (assuming 5.0ghz CPU, 2080ti and medium settings with extras like shadows off) you'll be better off without the patchwork guessed frames.

[RealDCSpilot]

I'm on the best image quality side (no shadows is a no go) turning off reprojection leads to very inconsistent fps with additional stutters. Motion smoothing does a pretty good job nowadays, it's been getting better and better over the last months. However, DCS is still not the best optimized game for VR in general.

[boedha68]

I'm on the best image quality side (no shadows is a no go) turning off reprojection leads to very inconsistent fps with additional stutters. Motion smoothing does a pretty good job nowadays, it's been getting better and better over the last months. However, DCS is still not the best optimized game for VR in general.

 

+1 This is a definitly go. I have explored different settings, but these i find the best too! :D