7950x3d Benchmarks for MSFS 2020

decided to go down the 7950x3d route with a 4090
1)just wondering what the best memory is to get for it - I am reading 32gb 6000mhz and CL30 - is this correct?
2)also in terms of tuning the 7950 - will I need some specific AMD software in terms of tweaking CCD’S or is all down via the bios?

1 Like

cl30 6000mhz is the sweet spot. 2x32 kit for 64gb is where i’d personally go
For tuning, the amd chipset drivers and game mode is a working combo to park cores. Myself have done it in the BIOS as it’s easy enough. This way I have ccd0 always looking to be run

thanks
1)will go cl30 and 6000mhz then - note you go for 64gb but whats the logic and will it get used and improve performance ? (also assume if going 64gb you should go x2 slots otherwise might slow the RAM?)

2)by game mode assume this is the microsoft game bar which you can enable on the desktop? also for amd chipset drivers - will these not come preinstalled or is there a seperate control software(guessing lasso is not really required)?

1 Like

You don’t need 64GB, I haven’t seen more than 32GB used yet, even with navigraph, Fenix, and other stuff running. Maybe 2024 will even use less. Having more RAM than the sim wants to use will not improve performance.

I do have 64GB in 2 sticks and the reason I did that is because if you add 4 sticks to the ryzen 7/9 then that reduces your RAM speed, so either you buy 64 to start with or if you later want to upgrade you have to ditch the 32 and buy 64 new, can’t really just add 32 like we would before. So it just depends how over the top and future proof you want to be.

Note that 32GB is likely to be easier to OC than 64, if you want to try that.

1 Like

The chipset drivers will be available from your motherboard manufacturer and amd website.
Yes, game mode in graphics settings and game bar on

That is not always the case with higher tlods, and some heavy third party scenery. I have seen my system use way more. There are city photogrammetrys’ on flightsim.to now that are into 30GB. Then, don’t forget the possibility of a 16gb ramdisk for rolling cache is out the window too. With this enabled I have seen up to 80% of my 64GB used.

I have seen higher usage than this too

1 Like

OK. I haven’t tried any city sized photogrammetry other than what Asobo provides and that streams rather than all loads into ram together. In my testing flying all over London pg with inibuilds EGLL Orbx EGLC and Orbx city landmarks it’s not gone above about 20GB.

For sure if you want rolling cache in RAM then 64GB would be recommended. I’ll try that when I get time, at the moment I have rolling cache on SSD which seems to work well enough. I looked at the lifetime writes specification and cache on SSD is no concern for me.

1 Like

You are totally right though, for normal usage of the sim. 32GB gets it done. :smiley:

I would have agreed with this a few days ago, but @MrTonySM has been helping me get my system sorted after all the troubles I have had, and has given me hours of his time and the depth of his knowledge in this area for which I am extremely grateful (and I hope commenting this here doesn’t open him up to HOURS MORE helping others do the same, sorry mate!).

Through his encouragement, I tried adding my other 2 16gb “Fury” sticks back in to the board and doing some tests. This is not EXPO ram, but it is on the QVL list for the Asus board (X670E-F). I had problems posting with all 4 sticks, and the SECOND kit of G.Skill I bought does not post at all with the 2nd 32gb stick in. I am not saying the G.Skill isn’t good for the system (obviously it is rated highly for a reason), but for me it does not work. I wanted to find a solution without sending the machine back - as all tests do show it is working correctly - just something with that ram kit it doesn’t like.

So through some tweaks and setting DOCP 2 on my Kingston Fury, I managed to get 64gb in 4 slots working, and it’s fully stable, but I did have to slow it down a LITTLE bit otherwise it was throwing some errors in OCCT test software. Just done a 1 hour stress test at 5800Mhz and it’s totally fine. Now I think the issues with using 4 slots would be (as you mentioned above), if you want to overclock it more.

I think my next step is to tweak the voltages and try to get the CL time down (even below the CL32 it’s rated for) and this should compensate for the 200Mhz loss and hopefully still withstand a memory stress test as below:

Upshot is, through my new found actual experience at the hands of an expert, 4 slots is NOT slower, it is just potentially more tricky to get stable at those speeds. But clearly even with something being weird with my board in it not liking TWO separate bought kits of G.Skill even at STOCK speeds, this Fury Renegade XMP 4x 16gb kit is now working just fine (likely since the 1415 bios update things have changed) and I am more than happy to stick with using 4 slots to get my 64gb. Ramdisk is on the agenda over the weekend now! :smiley:

1 Like

You’re very welcome. Very happy we got that second kit running for ya :smiley:

1 Like

I’ll take this article with a grain of salt too :smiley: Reviewer states 4 sticks is faster than 2 on Ryzen 5000 series. Updated May 2022

My comment on 4 sticks wasn’t based on random internet hearsay but instead from the manufacturer’s datasheet.

Max Memory Speed
2x1R DDR5-5200
2x2R DDR5-5200
4x1R DDR5-3600
4x2R DDR5-3600

https://www.amd.com/en/product/12741

I’d actually bought 32GB knowing I could just buy 32GB more if I ever wanted to, but then while waiting for the other parts I looked into this and realised the limitation, so I took it back to the shop and swapped it for 64GB.

Now of course, you can overclock it and you may or may not be lucky. These are guaranteed speeds not maximums, but given that 4 sticks comes with substantially lower rating that shows there’s a lot less headroom. I’ve read some decent articles where 4 sticks have been unable to overclock to 6000. YMMV

At first I was a bit surprised that 4 sticks had this lower speed but then it’s clear that adding extra sticks places greater load on the output drivers for the bus from the CPU, also it will affect the electrical characteristics and probably cause reflections, etc. Given the tolerances and frequencies modern hardware runs at it kind of makes sense. It’s been a long time since I put an oscilloscope on a CPU and looked at it, we were still in MHz back then!

1 Like

How did you get your CPU score that high in 3DMark? I can only get mine to maybe 16000 on a good day…

1 Like

We went through that, and had to settle at 5800mhz. Even bumping voltage wouldn’t get it to 6000. We could get it posting, but not stable

My memory is highly tuned as well. Might be my undervolt and mobo settings help. I also have an EK Nucleus 360 AIO. Keeps things on the cool. I’m trying to crack 21000. When I finish off my oc cpu and ram the rest of the way, im just about positive it will

I’m still running mine at manufacturer settings. It’s still so much better than the xbox I don’t really care to waste hours messing around with it rather than flying. :grin:

3 Likes

I keep finding myself doing more tinkering than anything lol. I can’t help myself

1 Like

We’re hopeful we can chase away the 200mhz loss with tightening timings. What ya can’t get in speed, get it in reduced latency. Same thing. I’m sure you know, but a lot of people don’t. RAM’s value is based on bandwidth vs latency. a 7000mhz+ kit or w/e doesn’t mean it’s faster. I’m putting this up for others that might read

Careful you aren’t confusing 1500VA with 1500W … They are different.

For example: I have a similar (maybe same) Cyberpower UPS rated for 1500VA @ 1050W max.
My previous UPS - A very similar model (now with my parents) was rated: 1500VA @ 900W max.

2 Likes

My bad, too much to always remember. Thank-you, and smart choice You are very right. 1050W 1500VA