From 14900K to AMD 9950X3D Is Lack Luster

[Raisuli]

8 minutes ago, niru27 said:

I know, what I meant to say was your thread gave Raisuli the (false) impression that 14900k to 9950x3D is a downgrade

That would be a false impression.  I got the feeling it's not enough of an upgrade to go through the brain damage of getting my sim machine apart, playing musical parts with two different computers, running my powershell scripts to re-set all the controls, etc etc.

That doesn't mean I'm not going through the brain damage, just not in the near future and I might wait for the next gen CPUs.  I consciously went i9-14 rather than -ultra just because of the gaming reviews.  I don't think that's going to fix the stutters.

I keep putting off Windows 11 because they've turned harvesting personal data into an art form and I don't have a ready tool to fix that in the registry, but the information here about tweaking it for games has been awesome.  Also noodling a PiMaxSuperDuperDeluxeUraniumEdition whatever the latest VR is later this year, so this entire area is priceless. 

Lots of great information.

[Hiob]

14 minutes ago, Raisuli said:

That would be a false impression.  I got the feeling it's not enough of an upgrade to go through the brain damage of getting my sim machine apart, playing musical parts with two different computers, running my powershell scripts to re-set all the controls, etc etc.

That doesn't mean I'm not going through the brain damage, just not in the near future and I might wait for the next gen CPUs.  I consciously went i9-14 rather than -ultra just because of the gaming reviews.  I don't think that's going to fix the stutters.

I keep putting off Windows 11 because they've turned harvesting personal data into an art form and I don't have a ready tool to fix that in the registry, but the information here about tweaking it for games has been awesome.  Also noodling a PiMaxSuperDuperDeluxeUraniumEdition whatever the latest VR is later this year, so this entire area is priceless. 

Lots of great information.

That depends a little bit on the exact circumstances and if the current CPU is in fact a bottleneck for the desired scenario and framerate.

Generally speaking and looking at the edge cases, a X3D(*) CPU is certainly more optimal for DCS than any current Intel. That doesn’t mean that a 14K is a slouch.

Upgrading between those two is certainly not a BIG step in most scenarios. I wouldn’t make this step unless there is another reason that the Intel has to go (like a defect e.g.).

(* actually the 9950 isn’t the most optimal for DCS though, the single die X3Ds (9800\7800\5800) are slightly superior due to the absence of any inter-die problems. But of course DCS isn’t the only use case)

 

[Raisuli]

2 minutes ago, Hiob said:

(* actually the 9950 isn’t the most optimal for DCS though, the single die X3Ds (9800\7800\5800) are slightly superior due to the absence of any inter-die problems. But of course DCS isn’t the only use case)

 

DCS isn't even the most annoying thing I run into; I run 2.5d (not VR) and get ~149 FPS most of the time.  If that's not enough it's time to check your entitlements!

It's those other guys where I get 55 FPS despite a 5090 drawing enough power to run the helium liquification plant at Fermi.

I haven't dug into the hardware enough to figure that out, or decided if I even care enough to do anything about it.  The overall upshot being it's not time for an upgrade yet, which is good given I built this nine months ago.  My last machine lasted five years before I decided to get a performance boost, and honestly all I really needed was a new video card.

OCD kicking in again.

[Hiob]

32 minutes ago, Raisuli said:

DCS isn't even the most annoying thing I run into; I run 2.5d (not VR) and get ~149 FPS most of the time.  If that's not enough it's time to check your entitlements!

It's those other guys where I get 55 FPS despite a 5090 drawing enough power to run the helium liquification plant at Fermi.

I haven't dug into the hardware enough to figure that out, or decided if I even care enough to do anything about it.  The overall upshot being it's not time for an upgrade yet, which is good given I built this nine months ago.  My last machine lasted five years before I decided to get a performance boost, and honestly all I really needed was a new video card.

OCD kicking in again.

5 years is my ugrade cycle as well. With the exception of upgrading the GPU from a 3080 to a 4090 prematurely (because the 3080 was NOT a worthy successor to the holy 1080ti which I sported 5+ years before that), I intend to keep to the rythm.

Therefore it is/was easy for me to skip the lacklustre 5000 series and DDR5 (for now). I was/am a little bit tempted by the 9800x3D though.

[Raven (Elysian Angel)]

51 minutes ago, Hiob said:

I was/am a little bit tempted by the 9800x3D though.

As I mentioned before in another thread, I went from a 5900X to 9800X3D and the performance upgrade is eyewatering in games that are CPU-heavy (not just simulation games but also ARPGs for example especially in endgame).

P.S. I kept the same graphics card but upgraded everything else - it was well worth it.

[Hiob]

29 minutes ago, Raven (Elysian Angel) said:

As I mentioned before in another thread, I went from a 5900X to 9800X3D and the performance upgrade is eyewatering in games that are CPU-heavy (not just simulation games but also ARPGs for example especially in endgame).

P.S. I kept the same graphics card but upgraded everything else - it was well worth it.

What keeps me from doing it is the necessary infrastructure update (mobo+ram) that comes with it. If it was the swap of the CPU, I would certainly do it.

It is just a gut feeling, but I don’t really want to switch to DDR5. And if I’m honest to myself, an upgrade isn’t really economically justifiable.

[The_Cokester]

Well I think I know what the issue was. I put Pimax plays quality to custom and put it at 20% and it was still crystal clear. Which I thought was odd. There was not perceived quality loss at all. Went through the config and noticed the resolution was at at like 1700x1700 per eye. Then put it at 100%  and it was at like 3840x3840 per eye. Deleted pimax play and reinstalled. 100% slider was at 2880x2880. ( I used steam VR to check the actual resolution coming through)

I guess a corrupted pimax play install was it. Or it thought my OG crystal was the super. Jesus I was going mad.

Also, there is an argument to put pimax play and DCS on the CCD1 without the cache. Average FPS was down like 10fps but it was stutter free.

Putting it on CCD0 with the VCache had a higher overall fps but there was more stutters.

Also, reapplied thermal paste and there is no change in temps.

So in conclusion Pimax play had my resolution almost two times higher than what my headset is rated for. A reinstall fixed the issue.

And P.S. There was a performance uplift going the AMD 9950X3D. Instead of 90 to 100fps. I was getting 110 to 120. Then when switching too DLAA and DLSS (the new transformer model got rid of the ghosting completely) pinned at 120fps.

Thanks everyone who chimed in.

[The_Cokester]

Also, never once did the frame counter say I was GPU bound. It just said CPU Bound Rendering Thread.

[Tomcat388th]

Glad you got it figured out.  I'm slowly contemplating an upgrade when my retirement kicks in next year.  Getting tired of Meta's update screwing the pooch for everyone. Have them locked out for now.  Running a 5800x3d quest pro  6950xt with virtual desktop.  Happy with the balance of gameplay and visuals but always wanting more

J

[Raven (Elysian Angel)]

6 hours ago, The_Cokester said:

Also, never once did the frame counter say I was GPU bound. It just said CPU Bound Rendering Thread.

It says the same for me: in 2D I’m ‘GPU bound’, but in VR I’m ‘CPU bound rendering thread’ with some dips to ‘CPU bound main thread’ on unoptimised sections of maps for example. My hope is that Vulkan will fix this, so that we are GPU bound even in VR.

[Rufuz64]

On 8/26/2025 at 1:48 AM, The_Cokester said:

Also, never once did the frame counter say I was GPU bound. It just said CPU Bound Rendering Thread.

Being CPU bound is normal with DCS.
But your 90°+ temperatures are a problem. I had the same issue with my upgrade. I completely redesigned my cooling system and installed three additional fans, which brought the temperature back down to below 80°.

[MoleUK]

On 8/26/2025 at 12:48 AM, The_Cokester said:

Also, never once did the frame counter say I was GPU bound. It just said CPU Bound Rendering Thread.

Yeah the in-game FPS counter will lie like that.

Essentially if it can't tell what is bottlenecking you, it will default to saying the CPU render thread is the problem. 

Capping the FPS externally via nvidia control panel can change this behavior and get it listing you as GPU limited.

Edited yesterday at 08:32 AM by MoleUK

[Hiob]

11 hours ago, MoleUK said:

Yeah the in-game FPS counter will lie like that.

Essentially if it can't tell what is bottlenecking you, it will default to saying the CPU render thread is the problem. 

Capping the FPS externally via nvidia control panel can change this behavior and get it listing you as GPU limited.

Actually the ingame fos counter is pretty reliable in reporting the limiting factor.

When you use Nvidia driver to limit fps, the limit is enforced by the GPU-driver, therefore it is reported ingame as GPU-limited….duh….

When you use the ingame fps-cap, it is enforced by the main-render-thread and therefore reported as CPU bound (as it is).

If there is no artificial fps cap, it will report the actual bottleneck correctly. Reasons for being  CPU limited with a Giga-Chad CPU could be (e.g.) GFX settings to low, playing in VR, CPU choked by outside factors (other applications, bios settings etc….)

[MoleUK]

11 hours ago, Hiob said:

Actually the ingame fos counter is pretty reliable in reporting the limiting factor.

When you use Nvidia driver to limit fps, the limit is enforced by the GPU-driver, therefore it is reported ingame as GPU-limited….duh….

When you use the ingame fps-cap, it is enforced by the main-render-thread and therefore reported as CPU bound (as it is).

If there is no artificial fps cap, it will report the actual bottleneck correctly. Reasons for being  CPU limited with a Giga-Chad CPU could be (e.g.) GFX settings to low, playing in VR, CPU choked by outside factors (other applications, bios settings etc….)

It's unfortunately not reliable in my experience.

If running in VR at 72hz and maintaining 72 fps, and just relying on the games internal FPS cap (or your headsets) it will report as CPU render thread limited as you say. But when the GPU struggles to maintain that 72hz/fps, and you dip below 72, it will still report as CPU render thread limited. Despite the GPU being definitively the limiting factor.

Additionally, if you have enabled reprojection and your GPU can;t keep up and you end up at half refresh rate being reprojected to hit 72fps, it will also list you as render thread limited. Despite the GPU again being the limiting factor there.

In the same scenario, with NVCP being used to cap the game to 72FPS: when maintaining 72 FPS it will report you as GPU limited. But more importantly, when you dip below 72 if the GPU can't keep up, it will correctly report the game as being GPU limited.

With the external cap in place, if you run into main sim thread bottlenecks it will still accurately point towards the main sim thread being a problem. But the render thread will no longer be inaccurately reported as the problem when the GPU can't keep up.

Additionally to all of this, DCS' internal FPS cap appears to have frame pacing problems in my experience. This isn't entirely uncommon in games, which is why it's often recommended to use an external tool to cap FPS for better frame-pacing generally.

The FPS graph being incorrect could be a quirk of my particular VR setup. Given that Quest/Pico headsets can sometimes run the PC at 99% GPU usage instead of 100%, and use additional encode. Maybe the game is misinterpreting that as having some GPU overhead leftover idk, I don't currently have a wired displayport headset to compare to atm.

Edited 10 hours ago by MoleUK

[Hiob]

3 hours ago, MoleUK said:

It's unfortunately not reliable in my experience.

If running in VR at 72hz and maintaining 72 fps, and just relying on the games internal FPS cap (or your headsets) it will report as CPU render thread limited as you say. But when the GPU struggles to maintain that 72hz/fps, and you dip below 72, it will still report as CPU render thread limited. Despite the GPU being definitively the limiting factor.

How do you figure that the GPU can't keep up? It is much more likely that the CPU hits a wall with calculations. Especially in VR, the CPU is much more a limiting factor.

[Raven (Elysian Angel)]

13 minutes ago, Hiob said:

Especially in VR, the CPU is much more a limiting factor.

This is also my experience: I’ve never seen GPU usage higher than 92-94% in VR.

If I were actually GPU limited, usage would be 98-99+.

[Hiob]

25 minutes ago, Raven (Elysian Angel) said:

This is also my experience: I’ve never seen GPU usage higher than 92-94% in VR.

If I were actually GPU limited, usage would be 98-99+.

I'm not sure how telling the GPU utilisation is nowadays. In yesteryears, when it was just rasterization power, it was much more straight forward.

Today GPU can hit several different walls. Rasterization, Frame gen, Raytracing etc..... I'm not sure how the utilization is calculated exactly. 

I have seen the GPU stuck at 97% reported in benchmarks while setting new (personal) records.

Long story short. When the reporting is somewhere in the 90s, I usually assume the GPU is properly loaded. Otherwise I usually see it hovering at 70something when running a frame cap at 120Hz.

[MoleUK]

4 hours ago, Hiob said:

How do you figure that the GPU can't keep up? It is much more likely that the CPU hits a wall with calculations. Especially in VR, the CPU is much more a limiting factor.

You can monitor GPU usage quite easily and see when it can't keep up.

DCS VR being primarily CPU bound hasn't been true for over 2 years, since the introduction of MT.

Before MT it was heavily CPU bound while still heavy on the GPU. Now it's vastly more likely to be GPU (or vram/ram) bound, except for unusual missions.

Many VR users are also using quad view, which increases the CPU load and decreases GPU load. The only reason quadview gives such performance gains is because we have the available CPU overhead to exchange for GPU gains.

Even a 5800X3D which is fairly dated will not struggle to keep up with DCS VR atm for the most part, even with quad view enabled.

This is why I've been saying the FPS graph in DCS is misleading, if you've been led to believe you were CPU bottlenecked when you were not.

Edited 3 hours ago by MoleUK

[MoleUK]

3 hours ago, Raven (Elysian Angel) said:

This is also my experience: I’ve never seen GPU usage higher than 92-94% in VR.

If I were actually GPU limited, usage would be 98-99+.

Not if you are hitting max refresh, and not if you are using reprojection.

Using VD on my settings and hardware I am either hitting 72fps at 80-90% usage (I try and aim for 72fps, no reprojection with some free overhead), or if things get particularly busy I drop below 72fps and GPU pegs at 98-99% usage. That's expected.

VD is far better at just maxing out the GPU load than Meta link is mind, as link likes to cut down to half refresh rate if it can't maintain the full refresh rate even with ASW disabled.

Edited 3 hours ago by MoleUK

[niru27]

4 hours ago, Raven (Elysian Angel) said:

This is also my experience: I’ve never seen GPU usage higher than 92-94% in VR.

If I were actually GPU limited, usage would be 98-99+.

Anything >=95% you can assume it is GPU bound. Even though the average utilization is less than 100%, some frames are taking longer than the 1000/<headset Hz> ms to render so you will be seeing stutters on those frames, so you're at the limit.

 

18 hours ago, Hiob said:

Actually the ingame fos counter is pretty reliable in reporting the limiting factor.

You're missing the point that MoleUK is trying to make: in VR, it will NEVER show GPU limited even if you are GPU limited. Unless you enable Vsync. Weird, but try it out.

Crank up the pixel density so you know you will definitely be GPU bound,

and see what the fps counter says with/without Vsync.

On 8/26/2025 at 4:56 AM, The_Cokester said:

Putting it on CCD0 with the VCache had a higher overall fps but there was more stutters.

That's because Windows puts all programs on Core 0 by default, which on the 9950x3D is the one with the cache, so in Process Lasso set default affinity for all apps to the non x3D cores, and only DCS to x3D

Even on other CPUs preventing DCS from using Core 0 often results in a smoother experience