4 hours ago, Cpt_Piett said:

I realised this spending a great deal of time over the last few weeks troubleshooting "memory could not be read" CTDs. I have a factory OC'd Asus 3090. When I remove the OC i.e. reduce core clock by 165MHz, the CTDs disappear.

For me I also had to disable DDR5 XMP as this combined with GPU OC caused these CTDs. But it's 100% reproducible. If I return to GPU OC, the CTDs will reappear, every time. 

I’m glad I’m not the only one. It’s very strange. Have you tried a lower OC? My impression (which could be wrong) from this issue back in early 2021 was that even a 10MHz OC would cause CTD issues. It was not the OC instability, but just the mere presence of an OC. Rolling back to an earlier Nvidia driver resolved the crashes and enabled an OC just fine. The issue was not OC instability but some driver, OC, sim interaction.

We have the same GPU. I’m able to increase the power budget on my card in Afterburner to max (+123%) but I cannot touch the clocks. Fortunately, increasing the power budget allows the card to reach its factory OC at least, which is better than nothing. Have you also tried this?

Are you on Win 11 or 10? I’m pretty sure when I put my sim setup back together a month ago, on Win 10, it was running without nearly as many memory CTDs. After upgrading to Win 11 last week the problem got a lot worse where I was guaranteed an issues every time I started the sim. 

Edited August 14, 20223 yr by Virtual-Chris

2 hours ago, SierraDelta said:

I don't agree with the "MSFS doesn't like overclocking" mantra.

When I built the system in my sig I carefully sourced the components and spent about 8 days tuning it, incl. endless hours of stability tests. Admittedly, I probably scored big in the silicon lottery with my CPU and could probably run it at 5.3 if I wanted to. Now, very very important - cooling. The box I use is absolutely brilliantly constructed, it has 2 180mm fans at the bottom blowing cool air towards the 3090 and I modded the AIO so it has 6 fans in total.

Running Fenix A320 in MSFS in 4k the 3090 lingers between 70 - 73 degrees and the CPU between 40 - 60, mostly Ultra settings, vsync'ed to 30fps. Just finished a flight from PHNL to KSAN, no problems, 5h32m block-to-block.

Are you running an OC on your 3090? What OC on what make/model? Win 11?

I have run a CPU OC with MSFS no problem. It’s a GPU OC (of any amount) that it doesn’t like (and maybe memory now as Piett points out). Back in 2021 when this was happening, rolling back to an earlier Nvidia driver solved the CTDs with a GPU OC which convinced me it wasn’t actually the stability of the OC but some strange inner workings of the OC, driver and Sim. 

Edited August 14, 20223 yr by Virtual-Chris

8 minutes ago, Virtual-Chris said:

Have you tried a lower OC?

Even a +15MHz OC and I will get the CTDs, lol. 

9 minutes ago, Virtual-Chris said:

We have the same GPU. I’m able to increase the power budget on my card in Afterburner to max (+123%) but I cannot touch the clocks. Fortunately, increasing the power budget allows the card to reach its factory OC at least, which is better than nothing. Have you also tried this?

No. I basically had to remove the factory OC completely and return the card to founders edition clocks. I feel sorry for those not so computer savvy that own a factory OC'd GPU, thinking all is well and good, and getting CTDs. 

10 minutes ago, Virtual-Chris said:

Are you on Win 11 or 10?

11. 

3 minutes ago, Cpt_Piett said:

Even a +15MHz OC and I will get the CTDs, lol. 

No. I basically had to remove the factory OC completely and return the card to founders edition clocks. I feel sorry for those not so computer savvy that own a factory OC'd GPU, thinking all is well and good, and getting CTDs. 

11. 

If you haven’t tried increasing the power limit in Afterburner while leaving the other settings at default, you may want to try that. It’s working crash free for me and unlocked a bit more performance. I think my OC gave me about 70FPS (where I test), no OC is around 60, and max power budget around 65. 

Did you notice any change in the problem going from Win 10 to 11?

6 minutes ago, Virtual-Chris said:

If you haven’t tried increasing the power limit in Afterburner while leaving the other settings at default, you may want to try that. It’s working crash free for me and unlocked a bit more performance. I think my OC gave me about 70FPS (where I test), no OC is around 60, and max power budget around 65. 

Did you notice any change in the problem going from Win 10 to 11?

Not sure if I understand completely. Aren't you talking about overclocking, while at the same time stating that OC leads to CTDs? Not sure what increasing power limit will do. Perhaps I misunderstand. 

Can't remember if I had the issue with Win10 - I don't think so. 

EDIT: Or do you mean increasing the power limit so it can reach higher boost clocks? While keeping core clock unchanged i.e. stock. 

Edited August 14, 20223 yr by Cpt_Piett

Just now, Cpt_Piett said:

Not sure if I understand completely. Aren't you talking about overclocking, while at the same time stating that OC leads to CTDs? Not sure what increasing power limit will do. Perhaps I misunderstand. 

Can't remember if I had the issue with Win10 - I don't think so. 

Interesting about Win 10. It definitely spiked for me after updating to Win 11 last week. 

When you Overclock your GPU in Afterburner, there are four attributes you adjust with sliders… Voltage, power, Core clock and memory clock. On our cards (3090 Strix) you can increase the power up to 123%. This simply lets the card draw more power, and achieve the advertised factory OC. If I only increase this value in Afterburner and leave the voltage and clocks at default (no overclock), I’m fine… no crashing. The card will run at about 1750MHz without the added power. It runs at 1860Mhz with the added power. My OC added an additional 75MHz to this but that causes crashes. 

You may want to try using Afterburner to increase the power to max, while leaving everything else at default. It works for me and unlocks some extra performance. 

13 minutes ago, Virtual-Chris said:

You may want to try using Afterburner to increase the power to max, while leaving everything else at default. It works for me and unlocks some extra performance.

Thing is, I don't want any OC on my card, not even factory OC. Take a look at my Afterburner settings and GPU-Z. I've reduced the core clock by 165MHz (from 1395 to 1230). Boost clock is reduced by the same amount, from 1860 to 1695. If I have ANY overclock on my card, even factory OC, I'll get the CTDs. 

Yeah, I increased the power after reading your posts, it used to be lower 🙂

Edited August 14, 20223 yr by Cpt_Piett

1 hour ago, Virtual-Chris said:

Are you running an OC on your 3090? What OC on what make/model? Win 11?

I have run a CPU OC with MSFS no problem. It’s a GPU OC (of any amount) that it doesn’t like (and maybe memory now as Piett points out). Back in 2021 when this was happening, rolling back to an earlier Nvidia driver solved the CTDs with a GPU OC which convinced me it wasn’t actually the stability of the OC but some strange inner workings of the OC, driver and Sim. 

All the info is in my signature - 😉

Haven’t touched the 3090 settings, so it’s the factory OC. And I’m normally on the latest driver, but right now I’m probably a release or two behind.

32 minutes ago, Cpt_Piett said:

Thing is, I don't want any OC on my card, not even factory OC. Take a look at my Afterburner settings and GPU-Z. I've reduced the core clock by 165MHz (from 1395 to 1230). Boost clock is reduced by the same amount, from 1860 to 1695. If I have ANY overclock on my card, even factory OC, I'll get the CTDs. 

Yeah, I increased the power after reading your posts, it used to be lower 🙂

Your system may be different,.  But you shouldn’t need to downclock. At least I don’t need to.  But then I don’t have an issue with my RAM (but I’m on Mainstream DDR4 3600).

Maybe you’ve done this, but I would suggest you try setting everything back to default and just increase the power slider to max and see if that works. if it does, you’re still getting the max performance from the card without manually over locking it. 

Edited August 14, 20223 yr by Virtual-Chris

2 minutes ago, SierraDelta said:

All the info is in my signature - 😉

Haven’t touched the 3090 settings, so it’s the factory OC. And I’m normally on the latest driver, but right now I’m probably a release or two behind.

Signatures don’t show on Mobile. If folks just typed the info rather than “it’s in my signature” we could have saved a post or two 🙂

So your GPU is not OC’d then. It’s when you use Afterburner or similar to start manually over clocking your GPU that you will run into this issue we’re discussing here… And on Windows 11 it seems. 

I underclocked my AMD 6900xt about 10% when in VR mode, because the AMD with the HP Reverb G2 is not quite as stable as the Non-VR in MSFS.
I don't think I had the issue with the Quest 2 and the AMD though. So far so good, though I've only flown a couple hours in VR since the changes.

I get more FPS than I need anyhow in almost every situation, so my other reason for underclocking is I don't like heating up the room.
I have problems trying to use any FPS limiter in VR mode though (which is what I prefer to do), but MSFS seems to ignore it so I just underclock more. 

My GPU averages at about 65c (the actual hot spot temp) and maxes around 72c or so, and my CPU maxes at about 80c and averages in the mid 70's.
It's all air cooling with 4 Noctua fans, Noctua NH-D15S heatsink, and my huge 240mm case fan. I installed the video card horizontally which did reduce the heat.

I set my memory to 3200 as well for an AMD 5800x, this is the max rating of my board, I could go for 3600, but it won't make much difference (maybe 1-2 fps).

If your case has a large side intake fan, installing the GPU horizontally towards the case fan will likely make it run cooler, but you have to buy an Asus Rog Strix PCIE riser cable for about $45 (it's worth it though and works fine). You also might have to cut the slot inserts on the back of the case to expose the DP ports since the GPU is oriented differently.
 

Edited August 14, 20223 yr by Alpine Scenery

3 hours ago, Cpt_Piett said:

The way I did it was to just reduce RAM clock a little. I.e. I had XMP enabled which is 6000MHz for my DDR5. Then I reduced it to 5800 and kept timings and voltages in place. 

 

3 hours ago, abrams_tank said:

I didn't change the timings, voltages, or anything else in my BIOS.  My computer is now very stable, for any video game I play, MSFS included.

Thanks guys... if the problem continues, I'll try to lower my RAM clock a little.  Will probably only try Memtest if downclocking doesn't work as imaging USB drives, removing memory modules, and running multi hour tests seems like a bit of overkill for a very isolated (if not very frustrating) issue.  I find it strange that I've only encountered it with MSFS in the last few months and it occurs very randomly. 

Update… After some extensive tweaking and testing this evening, I decided to try slowly re-introducing my GPU OC settings. To my surprise, I could increase my core clock back to what it was. I didn’t test the GPU memory OC. I now suspect that was the culprit, and it just wasn’t stable. I’m going to leave the memory at default clocks and stick with the core OC. It was good for about a 5% increase in FPS and after several flights, no CTDs. I guess there’s no Nvidia driver conspiracy after all. Maybe as my card has agreed, the memory isn’t as stable at high clocks as it use to. OCing the memory on the 3090 doesn’t seem to add anything anyway. 

On 8/14/2022 at 5:38 AM, MrFuzzy said:

First, that with the RTX 3090 it's more useful to undervolt in order to keep it more quiet 🙂

Then, MSFS has no way to determine if you overclocked your card or not. In fact, every card based on the same GPU works at different clocks depending on the manufacturer and the cooling system. The clock also varies with the workload, and this is the cause of your crashes in my opinion.

With other games which are not CPU limited and keep your GPU usage constantly in the 90-99% range, your GPU works basically at one single frequency, and evidently that frequency is rock stable at its respective voltage. 

This is my voltage / clock curve with MSI Afterburner:

You have to think in terms of a curve rather than a point. If your GPU is stable at the highest frequency that you targeted with your overclock, it may not be fully stable across the entire curve.

In the above example, my GPU may be rock stable at 1890 MHz with 0.875 V, but not so stable at 1700 MHz / 0.850 V. And if that's the case and the GPU frequency dropped with a drop of the workload (for example on the ground at a big airport with a complex aircraft and hence severely CPU limited), also the voltage would drop and my GPU could cause a CTD or a freeze even if it's rock stable with every other game which keep it at 1890 MHz constantly.

This doesn't happen of course, because my curve is 100% tested and rock stable since at least one year 🙂

One thing you could do is force the GPU to always work at its max frequency (and related voltage) regardless of the workload, by setting Max performance profile in NVCP, but the best thing to do is optimize the frequency / voltage curve in order to make it stable across all the points.

I also found it advantageous to undervolt my 3090, I currently have 1900MHz on the clock with 900mV and it is rock solid, I typically see around 65 degrees, fans around 1400RPM (I also have 3 X 140mm fans under my GPU that help) and the power draw sits around 300W and for a 3090 I was really pleased with that.

Edited August 26, 20223 yr by RJC68