High CPU load and frametime with Ryzen 9 5900x

[TripleHelix]

I've been having some trouble with high CPU frame times since upgrading my system which is making gameplay difficult. My system specs and VR settings can be found below:

AMD Ryzen 9 5900X cooled with an Artic Liquid Freezer II AIO (issue occurs with and without an overclock)
Gigabyte Gaming OC RTX 3080ti, no overclock
32GB 3200MHz Corsair Vengeance RAM (4x 8GB), XMP profile enabled
Samsung Evo 970 Plus 500GB M.2 SSD
Aorus Elite V2 motherboard
Valve Index: 90/120Hz refresh, SS 150%, motion smoothing on

I seem to get high CPU frame times when some instrumentation, AI jets, boats, buildings and rear-view mirror reflections are in view. On fpsVR it looks like an orange saw-tooth pattern with the performance graph. Weirdly it causes my virtual hands to stutter more than the VR headset display. This happens even when SS is set to 100% and refresh at only 90Hz.

My hardware is now pretty beefy and my last CPU which was an i7 9700K coupled with an RTX 2060 handled all of these things fine. I've searched these forums to see if anyone else was experiencing anything similar with this level of hardware but I couldn't find much. Do you have any advice as to how I can improve or stabilise performance?

[Slice]

Автор сообщения: TripleHelix

I've been having some trouble with high CPU frame times since upgrading my system which is making gameplay difficult. My system specs and VR settings can be found below:

AMD Ryzen 9 5900X cooled with an Artic Liquid Freezer II AIO (issue occurs with and without an overclock)
Gigabyte Gaming OC RTX 3080ti, no overclock
32GB 3200MHz Corsair Vengeance RAM (4x 8GB), XMP profile enabled
Samsung Evo 970 Plus 500GB M.2 SSD
Aorus Elite V2 motherboard
Valve Index: 90/120Hz refresh, SS 150%, motion smoothing on

I seem to get high CPU frame times when some instrumentation, AI jets, boats, buildings and rear-view mirror reflections are in view. On fpsVR it looks like an orange saw-tooth pattern with the performance graph. Weirdly it causes my virtual hands to stutter more than the VR headset display. This happens even when SS is set to 100% and refresh at only 90Hz.

My hardware is now pretty beefy and my last CPU which was an i7 9700K coupled with an RTX 2060 handled all of these things fine. I've searched these forums to see if anyone else was experiencing anything similar with this level of hardware but I couldn't find much. Do you have any advice as to how I can improve or stabilise performance?

Solution: Use Process Lasso to configure a CPU SET for "VTOLVR.EXE" for cores 0 to 11, your problem is fixed now.

[TripleHelix]

Автор сообщения: Slice

Автор сообщения: TripleHelix

I've been having some trouble with high CPU frame times since upgrading my system which is making gameplay difficult. My system specs and VR settings can be found below:

AMD Ryzen 9 5900X cooled with an Artic Liquid Freezer II AIO (issue occurs with and without an overclock)
Gigabyte Gaming OC RTX 3080ti, no overclock
32GB 3200MHz Corsair Vengeance RAM (4x 8GB), XMP profile enabled
Samsung Evo 970 Plus 500GB M.2 SSD
Aorus Elite V2 motherboard
Valve Index: 90/120Hz refresh, SS 150%, motion smoothing on

I seem to get high CPU frame times when some instrumentation, AI jets, boats, buildings and rear-view mirror reflections are in view. On fpsVR it looks like an orange saw-tooth pattern with the performance graph. Weirdly it causes my virtual hands to stutter more than the VR headset display. This happens even when SS is set to 100% and refresh at only 90Hz.

My hardware is now pretty beefy and my last CPU which was an i7 9700K coupled with an RTX 2060 handled all of these things fine. I've searched these forums to see if anyone else was experiencing anything similar with this level of hardware but I couldn't find much. Do you have any advice as to how I can improve or stabilise performance?

Solution: Use Process Lasso to configure a CPU SET for "VTOLVR.EXE" for cores 0 to 11, your problem is fixed now.

Really appreciate the input. I've tried something similar to this using Windows Task Manager before which didn't seem to work. I've just tried the above as you suggested. I would say there's sadly only a very mild improvement. I was very careful to do CPU SET, not PRIORITY or AFFINITY, although I did try affinity too as I know that is a much more strict way of assigning cores.

The saw tooth like pattern is reduced but only because it limits the number of cores it can jump to next when the one proceeding it experiences an increased frametime. I'm still getting frequent drops in fps even at 90Hz with lots of stuttering, especially of the hands. The frequency of the stutter is correlated to the frequency of jumping between cores. Each jump causes a lag. It's doesn't affect the GPU frametime at all. Sadly I can't run the game on a single core as it maxes out quickly and fps drops to unplayable levels.

I've messaged AMD to see if they have any more up to date work arounds for high end Ryzen CPUs and CPU bottle necked games that use the Unity engine. I'll report here if I hear back from them. Fingers crossed

[derSanitoeter]

Running Ryzen 5000 with 4 modules of memory is really not that great either to be honest.

99,9% of the boards a Daisy Chain topology which means that 2 modules perform better and can reach a higher stable clock.

Could be possible that your 4 modules XMP setup is not 100% stable and error correction maybe kicks in.

Did you ever check with Karhu or Aida 64 Extreme Stability Test or anything like that, that there are no error and / or your system runs without WHEA errors?


I using the AMD 5950X with 16 cores / 32 threads and have no issues at all. Everything runs fine and perfect. But i use 2 16GB modules with DDR4 3800 C16 right now.

[TripleHelix]

Автор сообщения: SaltySani

Running Ryzen 5000 with 4 modules of memory is really not that great either to be honest.

99,9% of the boards a Daisy Chain topology which means that 2 modules perform better and can reach a higher stable clock.

Could be possible that your 4 modules XMP setup is not 100% stable and error correction maybe kicks in.

Did you ever check with Karhu or Aida 64 Extreme Stability Test or anything like that, that there are no error and / or your system runs without WHEA errors?


I using the AMD 5950X with 16 cores / 32 threads and have no issues at all. Everything runs fine and perfect. But i use 2 16GB modules with DDR4 3800 C16 right now.

Very interesting. I'll check the above as you've suggested and get back to you. Thanks pal

[TripleHelix]

So I paid for a licence for Karhu as it was cheaper and good to have anyway. I ran the test but it kept on going after 100% coverage. Not sure when it would stop or if it just continues indefinitely. I stopped it at 126% and no errors were found. HWiNFO64 also sees no hardware errors (at the bottom of the page). I also don't appear to be getting any WHEA errors either.

I know UserBenchmark isn't the most hardcore of benchmarks but one thing I have found is that whilst my RAM with XMP enabled gets an overall score of 116%, it is doing well with multicore tasks but poorly with single core (67%) tasks and latency (44%). Could this be a warning that all is not well?

[derSanitoeter]

10000 %, then reboot and do another 10000 %.
Takes some time but that's the way to do it.

UserBenchmark is not a real benchmark and simply Intel biased ♥♥♥♥♥♥♥♥.
They had huge negative press last years after dropping AMD scores and pushing Intel CPUs into heaven even when they were a lot worse. It's crap.

[TripleHelix]

Автор сообщения: SaltySani

10000 %, then reboot and do another 10000 %.
Takes some time but that's the way to do it.

UserBenchmark is not a real benchmark and simply Intel biased ♥♥♥♥♥♥♥♥.
They had huge negative press last years after dropping AMD scores and pushing Intel CPUs into heaven even when they were a lot worse. It's crap.

Ah that's a shame. I'll be taking those UserBench scores with a very large pinch of salt. One thing I did do was download Memtest from PassMark. Booted the image of it from a USB. It took just under 5 hours but all my RAM sticks passed without any errors.

I've just gone on Performance Test by PassMark to use as a benchmark instead of UserBench. I'm getting OK single core scores with XMP enabled but again my latency is still crap at 43ns. Could this be an issue?

[derSanitoeter]

I only use Aida 64 Extreme for Cache/Memory Benchmarks as most people do - esp. with Ryzen.

[TripleHelix]

Whilst I would pay for the licence for it I am wondering whether we're barking up the wrong tree here? I've searched online and it appears my latency isn't atypical.

Going back to my previous setup, the game ran smooth as anything with an i7 9700K and half of the exact same speed/latency/brand of memory as I have now. It just doesn't make sense that you would need a system like yours which is more akin to a rendering workstation to run a game as basic as VTOL VR. I could upgrade my RAM to specs similar to yours, but surely brute forcing this issue with more powerful and expensive hardware shouldn't be necessary?

[derSanitoeter]

We have a second system with 3700X + 16GB DDR4 3600 C16 + 1070 and it can run VTOL VR flawless. o_O

I mean... something is wrong on your system, could be software-side too.

Most recent UEFI, most recent chipset drivers from AMD, most recent GPU drivers - clean windows. No stuff running in the background. Should easily get perfect performance.

Don't overclock the CPU AT ALL, it takes the highest clock it can get within the powerlimits and your cooling capabilities. ( I did read with / w/o OC... ) Never DO OC THE RYZEN!

Never use Ryzen Master, as it still seems to fight with UEFI setting all time in use.

OH what PSU do you exactly have? ( Brand, Name, Revision, Watt? )

[TripleHelix]

All excellent suggestions. Just done the below:

- Downloaded most recent BIOS from Gigabyte, BIOS updated
- Downloaded most recent AMD chipset drivers and updated
- Nvidia GPU drivers were updated about a week ago
- Disabled CPU overclock that had been applied via Ryzen Master
- Aside from Nvidia control panel and Steam nothing else is running in the background. No RGB controllers/monitoring software/Afterburner etc.

Back into VTOL VR, set at 90Hz, Desert Cobra mission 3, still the same saw tooth high CPU frametimes with NPCs in view etc. It seems worse though which I think is because it was lessened with the CPU being OC'd.

My power supply is a fully modular Corsair RM 850 2019 model. This is such a strange issue!

[derSanitoeter]

Mhm only in this mission or in ALL missions?

[TripleHelix]

Any mission that involves higher CPU usage than just flying around and hitting static targets. I get decent performance at 90Hz on the training tutorials, and even the target practice mission. But as soon as there are friendly NPCs around I get the saw tooth high CPU frame times. It also occurs every time I flip down the rear view mirror without fail, and I saw someone else who got something similar whenever they use S-Cam and that happens with me too.

I've still not heard back from AMD but hopefully I'll get something. I did just increase my Windows virtual memory working a problem I've been having with H3VR and Boneworks where I see an intermittent microstutter with perfect CPU and GPU frame times. It seems to have made a modest improvement with Boneworks but sadly not so with VTOL VR. It's a bastard this issue, I'll give it that!

[derSanitoeter]

And again - we have 2 AMD system here and both run the game with all stuff perfectly fine.

I still say it is an issue on your end and not a AMD issue.

[jtfb4u]

Adjust your virtual memory to reflect your actual available on your hard drives! If you have 16 set it to 16 if you have 32 set it to 32! the larger hard drives do not adjust automatically.