3 hours ago, GSalden said:

I choose for a. Asus Rog Strix DDR4 version instead of DDR5 about a year ago.

Because of the 13900 cpu I had to turn down the 32Gn GSkill 4266 MHz cas 17/18/18/39 memory to 3200 MHz but I lowered cas to 13/14/14/32….

The 13900 series will work with DDR5 4800 MHz but the timings are so much higher than DDR4 that the complete set is slightly slower icw MSFS …..

That's an important point -- I have traditionally gotten better performance running lower latency sticks as opposed to higher clock speed.   But I've never built a rig specifically for MSFS.  This assumes MSFS responds the same way as P3D or FSX or FS2004 etc. did as to memory.

In other words I'll take the CL16 ram over CL20 or whatever.  Even if the CL20 is, say, rated up to 6000 and the CL16 is 4266.  Is that thinking wrong or flawed?

6 hours ago, Mace said:

That's an important point -- I have traditionally gotten better performance running lower latency sticks as opposed to higher clock speed.   But I've never built a rig specifically for MSFS.  This assumes MSFS responds the same way as P3D or FSX or FS2004 etc. did as to memory.

In other words I'll take the CL16 ram over CL20 or whatever.  Even if the CL20 is, say, rated up to 6000 and the CL16 is 4266.  Is that thinking wrong or flawed?

Here an overview of DDR4 : you can see the speed in nanoseconds for a certain frequency and certain cas latency.

 

 

Edited November 10, 20223 yr by GSalden

11 hours ago, GSalden said:

Because of the 13900 cpu I had to turn down the 32Gn GSkill 4266 MHz cas 17/18/18/39 memory to 3200 MHz but I lowered cas to 13/14/14/32….

Slightly surprised that the IMC of your Raptor Lake seems to be handling OC RAM worse than your predecessor CPU, but good to know, then I do not consider OC my RAM anything beyond I already have (3200 CL14). 

9 hours ago, Mace said:

In other words I'll take the CL16 ram over CL20 or whatever.  Even if the CL20 is, say, rated up to 6000 and the CL16 is 4266.  Is that thinking wrong or flawed?

No, not flawed at all. This specifically relates to DDR4, as I have no experience with DDR5 (although the general philosophy will be similar), after watching all of the online testing with DDR4, it was clear there is not much advantage after 3800.  You are better off working on the other timings then, which can give rewards if you get it right.  With DDR4, anywhere around 3200-3800 at CL16 is not a bad place to be.  (CL16 at 4266 as you stated above may be a hard stretch , I had to be at 18 with good RAM and even then I doubted the stability).

The problem is the testing.  It takes a lot of time.  I go into the BIOS, alter a few settings, then you have to reboot to have the settings take effect and go into the bios again before running Memtest86 for an hour to prove stability, although normally, if the RAM is going to fail the test it normally does it within 20 minutes I have found.  I got into a cycle of getting in from work, booting into the bios, making a tweak, rebooting into the bios again and running memtest for an hour while I relaxed and had my dinner - one tweak per day! 

Just a footnote:  The last thing you want is RAM instability, so get it stable and then even tweak it back a notch if you can to make sure.  During a bad blue screen caused by over stretching my RAM before I started using memtest, it appeared it was writing to the SSD and it corrupted some files.  It took me a few hours to recover and get back to where I started.  It was a bit frightening at the time, as it affected my windows and MSFS installations.

The moral is, never adjust your RAM in the direction of more performance without performing a memtest. 
If you just wait to see if it crashes on boot into windows or in MSFS, dragons lie in that direction - trust me!

10 hours ago, raknal said:

Just odrered a new PC with Kingston Fury Beast 32gb (2x16GB Kit) 2 of for 64GB 5200MHz Non-ECC Ram Black heat spreader, so is the above comments going to apply to my purchase?

Possibly Raknal.  You will have 4 sticks, and the XMP profile for overclocking will only apply to 32 GB.  You may be lucky though as the RAM quality can vary. 

You just need to test it to prove stability.  The best thing to do is go into the bios, then into tools, and run Memtast 86 for an hour while you do something else. 
If it passes, then fine.  If not, just tweak the speeds back a bit and try again. 

A proportion of RAM at 64GB may hit the stated specification, but from my experience, and after reading about a lot of issues on various hardware forums, a lot of RAM won't quite do it.

You would think that if you get two 32GB packs at a certain rating, you can put both of them in for 64GB, and run them at the rated XMP profile.  But for whatever reason, sometimes it doesn't work out like that.  The stated speed rating will be guaranteed for 32GB only, even though each individual pack is rated at those speeds, when they are combined, for whatever reason, it doesn't always work out.

2 hours ago, AnkH said:

Slightly surprised that the IMC of your Raptor Lake seems to be handling OC RAM worse than your predecessor CPU, but good to know, then I do not consider OC my RAM anything beyond I already have (3200 CL14). 

As I have understood it from the store both 12900 and 13900 have the same memory controller. What you gain in MHz with your cpu speed you loose in memory speed MHz by the controller…. Intel really has done a great job 🤒

Luckily I can run  the cas settings lower than when I was using the 12900. 

And as Windows + MSFS never have used the complete memory the difference in game is neglectable…

The framerate in total is about 10% higher than when using the 12900. But the biggest advantage is using the GPU’s extra AI frames which make the total fps twice as it is. That makes it way smoother.

 

DDR5 6000MHZ here. Will be testing next week so will let you know. Thats using AMD which I believe they call XMP = AMP. 

45 minutes ago, GSalden said:

As I have understood it from the store both 12900 and 13900 have the same memory controller. What you gain in MHz with your cpu speed you loose in memory speed MHz by the controller…. Intel really has done a great job 🤒

Luckily I can run  the cas settings lower than when I was using the 12900. 

And as Windows + MSFS never have used the complete memory the difference in game is neglectable…

The framerate in total is about 10% higher than when using the 12900. But the biggest advantage is using the GPU’s extra AI frames which make the total fps twice as it is. That makes it way smoother.

Yes, the difference for me between the 12900 and 13900 is about 10-15%.  No problems with RAM timings here, but I run lower than XMP anyway.

I can't measure it exactly, as the FPS jump about whichever measuring software you use, but in the Task Manager performance display, I noticed two of the 12900 cores were topping out and flat-lining occasionally, causing me stutters.  With the 13900, none of the cores / threads go above about 90% (at my 4k ultra settings).  So basically, it isn't all about % increase in frames, but stopping my 12900 from being overloaded.  MSFS is very smooth now by adding just that little bit of extra headroom.

A worthwhile upgrade, and I can sell the 12900k for a fairly good price as well.

Have you disabled any of the 16 Eco cores?  These are useless for MSFS, but I have heard the cache gets added to the other performance cores.  Not sure if it is a myth. 

1 hour ago, bobcat999 said:

Yes, the difference for me between the 12900 and 13900 is about 10-15%.  No problems with RAM timings here, but I run lower than XMP anyway.

I can't measure it exactly, as the FPS jump about whichever measuring software you use, but in the Task Manager performance display, I noticed two of the 12900 cores were topping out and flat-lining occasionally, causing me stutters.  With the 13900, none of the cores / threads go above about 90% (at my 4k ultra settings).  So basically, it isn't all about % increase in frames, but stopping my 12900 from being overloaded.  MSFS is very smooth now by adding just that little bit of extra headroom.

A worthwhile upgrade, and I can sell the 12900k for a fairly good price as well.

Have you disabled any of the 16 Eco cores?  These are useless for MSFS, but I have heard the cache gets added to the other performance cores.  Not sure if it is a myth. 

Beside MSFS I also run Prosim, SIOC, PSXT+ RT, OpusFSI and Fs2Crew RAAS. And the apps with the lowest cpu usage are assigned to the E cores automaticall.
 

What I do wonder is if it is better to run with HT on (default) or off …?

19 minutes ago, GSalden said:

What I do wonder is if it is better to run with HT on (default) or off …?

To my knowledge, if you do not run into temperature issues with HT on, there is no reason to turn it off. And even then, in my honest opinion, rather take the OC slightly back than sacrifying HT. But that is just me using my rig also for games where HT might provide some extra performance, not only MSFS...

On 11/9/2022 at 12:27 PM, altenae said:

Anyone noticed more CTD's when going from DDR4 to DDR5 ?
Thanks 

 

After upgrading my rig to a DDR5 platform with G.Skill @ 6000MHz in XMP I struggled with "memory could not be read" CTDs for a while. Not sure if DDR5 was the main cause though, but I think it contributed to the CTDs. 

Jayz did a video in the early days of DDR5 and found that running 4 sticks could cause issues. So it's not unlikely that some DDR5 configurations can cause issues. 

EDIT: That was back in the SU10 beta. CTD frequency was a lot less when I disabled XMP and reduced the speed to 5800. IIRC I'm now back on 6000 in SU10/11 with no issues. Like Rob says - if there's issues (CTDs or even BSODs) - run memtest, reduce speed until failure-free. Or even better: run memtest first, just to be sure. 

2 hours ago, bobcat999 said:

Have you disabled any of the 16 Eco cores?  These are useless for MSFS, but I have heard the cache gets added to the other performance cores.  Not sure if it is a myth. 

What I do with my 12900KF is that I assign cores using Process Lasso. Basically I remove the E-cores from MSFS (after MSFS has started - it won't work beforehand). Then I put other stuff - TDS GTNXi, FSLTL injector, video recording software etc - on the E-cores. With DX12 the P-cores get a good workout when running MSFS - often running at 100% on all P-cores in demanding scenarious. 

And - I can record 4K videos at 60fps without any performance hit on MSFS. Which I've never been able to do before. The apps that are assigned to E-cores only simply won't compete with MSFS and affect mainthread frame times. 

1 hour ago, GSalden said:

What I do wonder is if it is better to run with HT on (default) or off …?

I would say HT on - let MSFS have as many P-core threads available as possible. With DX12 they will all be used. That's my experience anyway. My 12900 has 8 P-cores 16 threads with HT. E-cores don't have HT so 8 cores / 8 threads. 

Edited November 10, 20223 yr by Cpt_Piett

52 minutes ago, GSalden said:

Beside MSFS I also run Prosim, SIOC, PSXT+ RT, OpusFSI and Fs2Crew RAAS. And the apps with the lowest cpu usage are assigned to the E cores automaticall.
What I do wonder is if it is better to run with HT on (default) or off …?

Looking at the activity of the hyperthreading in Task Manager, I am leaving mine on.  Maybe it might help temperatures a bit, but when MSFS and addons are running (like FSRealistic), every core and thread on the performance cores is doing some work. 

My thinking is turning hyperthreading off will have to dump that work onto the remaining performance cores, so things could go backwards.  I may try this at the weekend though.

1 hour ago, Cpt_Piett said:

What I do with my 12900KF is that I assign cores using Process Lasso. Basically I remove the E-cores from MSFS (after MSFS has started - it won't work beforehand). Then I put other stuff - TDS GTNXi, FSLTL injector, video recording software etc - on the E-cores. With DX12 the P-cores get a good workout when running MSFS - often running at 100% on all P-cores in demanding scenarious. 

And - I can record 4K videos at 60fps without any performance hit on MSFS. Which I've never been able to do before. The apps that are assigned to E-cores only simply won't compete with MSFS and affect mainthread frame times. 

Nice - I might try that!  Is that the free version of PL, or the paid one where it lets you save profiles?

My thoughts on disabling the E-cores is because, looking at the AMD X3D chips, the extra cache really helps performance, and not just by a little bit.

There are now 16 E-cores on the 13900, too many for general and gaming usage I think (great for video processing no doubt).

So if by disabling them, the cache gets shared out (which I still haven't seen proof) then this extra cache on the P-cores should help.

I would probably keep 4 out of the 16 E-cores activated in the bios, to do background tasks as you say, but as they run at 4.3GHz on the 13900, four should be more than enough for background tasks.  I must experiment with this and Process Lasso soon, as I intend to capture video also.  Last time I experimented with video capture (about a year ago), The Nvidia experience one worked the best - any preference for yourself?

PS. Apologies to the OP for wandering slightly off topic, but it still relates to performance and memory / cache etc. for MSFS.

4 minutes ago, bobcat999 said:

PS. Apologies to the OP for wandering slightly off topic, but it still relates to performance and memory / cache etc. for MSFS.

Same. I hope it's possible to have this little slightly off-topic (but still relevant!) discussion.

I have the paid version of PL, but I don't think it's necessary. 

In terms of video recording - I discovered Action! a few weeks ago and absolutely love it. Highly recommended. I did use GFE before, but nowadays I "DDU" my display driver i.e. don't bother to install GFE. With Action it's possible to assign it to certain cores. It has a multithread-enabled encoder, and has lots of options in terms of audio and video settings. It records up to 4K 60fps and can upload directly to YouTube. Much more flexible than GFE. 

Ctd in this sim are typically an issue with Community addons.  If it's not that then usually video driver.  Try s clear community folder?  (Or rename yours and make a blank comm folder)

Im using Z690 with 5600mhz ddr5 for my 12700K and haven't had any hardware issues yet.  Windows 11 as well.