MSFS forums claiming 64GB of Ram helps on 3090s

[Noel 4,496]

On 2/28/2021 at 8:11 AM, kand said:

LOL, 3090 owner, I went from 32gb DDR4 3200 to 64gb DDR4 3600...... and got lower FPS over London and 2020 uses exactly the same amount of CPU mem (15/64) and VRAM (8/24) 

save your money 

I can't believe people actually fall for this when the difference w/ likely be imperceptable, barely measureable, just like this well validated comment, "There also is a small performance benefit from using four sticks".   This 'small perf benefit' would be akin to throwing a few bricks in the Grand Canyon in your quest to fill it up.  Save your dollars unless it's clear you're getting anywhere near maxing out your 32Gb of RAM.  So far, in the most complex areas of planet earth the most I've seen w/ all settings on Ultra is 21Gb, and that only briefly.  Usually hanging around 11-15Gb in MSFS.

[Pilot53 544]

Well I have to say I was incredibly skeptical of this when I first read it, but decided I needed to test it out for myself since I have a 3090 and like to fool around with PC hardware for fun.  I am surprised to report that upgrading from 32GB to 64GB keeping speed and timings identical DID indeed make a very noticeable difference.  The fps numbers were close, I only gained 1 FPS, however the sim is much more fluid especially when using VR.  Its like the difference between having ASW enabled on an oculus headset below 90 FPS. I wish i had looked at the frame times and FPS minimums instead.  The sim would also get really stuttery when looking at clouds, or when inside them as well as looking at large cities, this is no longer the case.  

I also noticed that when exiting VR without the headset asleep before with 32GB I was unable to do other things like browse the web until the HMD went to sleep.  This is no longer the case with 64GB.  

So in the task manager I was using about 75-90% of virtual memory and 20GB of physical memory with 32GB, now I use the same amount of physical memory but virtual memory is down to 40%.  The other thing I noticed was that the page file was showing 4% usage with 32GB, now its at 0.0% the entire time. 

So, I guess there is something to this that I dont quite understand, I dont know if it makes a difference with a 3080 and only 10GB of VRAM, but I can say with the 24GB 3090 upgrading to 64GB did infact lead to a noticeably better experience in MSFS.  

Edited March 2, 2021 by Pilot53

[Ianrivaldosmith 5,317]

 Brilliant. Mine arrives this week (extra 32GB), so, I hope to have the same experience 

[RXP 2,338]

18 minutes ago, Pilot53 said:

I am surprised to report that upgrading from 32GB to 64GB keeping speed and timings identical DID indeed make a very noticeable difference.

Thank you for sharing! I'm glad it is helping.

[Pilot53 544]

32 minutes ago, RXP said:

Thank you for sharing! I'm glad it is helping.

Yessir!  Thank you for bringing this to our attention, I still wish i fully understood why it works though lol!

[CYXR 257]

On 2/27/2021 at 1:30 AM, G-RFRY said:

Some car production has had to shut down for lack of silicon.

Cars... chips and other beautifull things made of silicone; like err... muffin trays.  

[Glenn Fitzpatrick 3,798]

1 hour ago, Pilot53 said:

Well I have to say I was incredibly skeptical of this when I first read it, but decided I needed to test it out for myself since I have a 3090 and like to fool around with PC hardware for fun.  I am surprised to report that upgrading from 32GB to 64GB keeping speed and timings identical DID indeed make a very noticeable difference.  The fps numbers were close, I only gained 1 FPS, however the sim is much more fluid especially when using VR.  Its like the difference between having ASW enabled on an oculus headset below 90 FPS. I wish i had looked at the frame times and FPS minimums instead.  The sim would also get really stuttery when looking at clouds, or when inside them as well as looking at large cities, this is no longer the case.  

I also noticed that when exiting VR without the headset asleep before with 32GB I was unable to do other things like browse the web until the HMD went to sleep.  This is no longer the case with 64GB.  

So in the task manager I was using about 75-90% of virtual memory and 20GB of physical memory with 32GB, now I use the same amount of physical memory but virtual memory is down to 40%.  The other thing I noticed was that the page file was showing 4% usage with 32GB, now its at 0.0% the entire time. 

So, I guess there is something to this that I dont quite understand, I dont know if it makes a difference with a 3080 and only 10GB of VRAM, but I can say with the 24GB 3090 upgrading to 64GB did infact lead to a noticeably better experience in MSFS.  

So just to confirm , using a VR headset the 64GB makes a big difference.   

What about just a normal 2D monitor,  presumably 32Gb is fine and all you need if your not using VR?

[Pilot53 544]

11 minutes ago, Glenn Fitzpatrick said:

So just to confirm , using a VR headset the 64GB makes a big difference.   

What about just a normal 2D monitor,  presumably 32Gb is fine and all you need if your not using VR?

It was hard to tell, because without VR MSFS runs pretty smoothly and my fps are in the 40-50's in my test scenario vs 28-30 in VR, and with a headset you can really detect stutters with the slightest movement of your head.  I believe it helps too without VR.  It seems this has something to do with system allocating an amount of system ram proportional to the amount of VRAM you have on your graphics card.  That's why this seems to make a difference to those with the 3090 and its massive 24GB of VRAM, causing the system to use the page file with only (lol) 32GB of system ram.

Edited March 3, 2021 by Pilot53

[Noel 4,496]

On 3/1/2021 at 2:07 PM, RXP said:

Of course this is just speculation, albeit from technical understanding of how video card drivers are working internally in terms of resource management, but there are a lot of assumptions as well. In practice though, there are a number of people with a 3090 reporting much shorter load times and smoother flights with 64GB RAM instead of 32GB RAM, and this can only be attributed to me to having more RAM to accommodate the RAM needs of the simulator, the same RAM being otherwise amputated by some amount because of the 24GB and the DX11 driver.

Sounds plausible but still begs this question: in the most complex areas on the planet (in my 200h of exploration to some pretty busy areas), all settings on Ultra at 3440x1440, 4XSSAA usually, but sometimes 8X, the most sys ram I see in use (MSI Afterburner OSD) is 21Gb in total, that's everything running alongside.   Average in moderately busy areas is maybe 12-15Gb or so.  I can load my 5Gb Garritan Abbey Roads VST and play piano while the sim runs and usually still have 10-11Gb free of sys ram.   And...I've only got an 8Gb VRAM GPU.  Why doesn't the driver utilize much more of the 32Gb of sys ram already installed?  Why does doubling that change anything?

[RXP 2,338]

@Noel most likely because the amount of reserved RAM by DX11 is proportional to the amount of VRAM, regardless of use. This pretty much is the same way the Win10 Paging File size is computed in "AUTO": proportionally to your RAM size, and actually the memory DX11 could be allocating for backing the buffers is the one you'd get with VirtualAlloc and this type is backed by the paging file.

It begs the question of reserved vs committed, but in any case, if the OS is reserving lot's of RAM to back VRAM just in case, and commits some significant amount of this to keep copies of the buffers DX11 is managing in VRAM, you might be consuming paging file and this is way slower than RAM, which is way slower than cache, etc...

Latency Comparison Numbers (~2012)
----------------------------------
L1 cache reference                           0.5 ns
Branch mispredict                            5   ns
L2 cache reference                           7   ns                      14x L1 cache
Mutex lock/unlock                           25   ns
Main memory reference                      100   ns                      20x L2 cache, 200x L1 cache
Compress 1K bytes with Zippy             3,000   ns        3 us
Send 1K bytes over 1 Gbps network       10,000   ns       10 us
Read 4K randomly from SSD*             150,000   ns      150 us          ~1GB/sec SSD
Read 1 MB sequentially from memory     250,000   ns      250 us
Round trip within same datacenter      500,000   ns      500 us
Read 1 MB sequentially from SSD*     1,000,000   ns    1,000 us    1 ms  ~1GB/sec SSD, 4X memory
Disk seek                           10,000,000   ns   10,000 us   10 ms  20x datacenter roundtrip
Read 1 MB sequentially from disk    20,000,000   ns   20,000 us   20 ms  80x memory, 20X SSD
Send packet CA->Netherlands->CA    150,000,000   ns  150,000 us  150 ms

https://colin-scott.github.io/personal_website/research/interactive_latency.html

[Pilot53 544]

7 minutes ago, RXP said:

@Noel most likely because the amount of reserved RAM by DX11 is proportional to the amount of VRAM, regardless of use. This pretty much is the same way the Win10 Paging File size is computed in "AUTO": proportionally to your RAM size, and actually the memory DX11 could be allocating for backing the buffers is the one you'd get with VirtualAlloc and this type is backed by the paging file.

It begs the question of reserved vs committed, but in any case, if the OS is reserving lot's of RAM to back VRAM just in case, and commits some significant amount of this to keep copies of the buffers DX11 is managing in VRAM, you might be consuming paging file and this is way slower than RAM, which is way slower than cache, etc...

Latency Comparison Numbers (~2012)
----------------------------------
L1 cache reference                           0.5 ns
Branch mispredict                            5   ns
L2 cache reference                           7   ns                      14x L1 cache
Mutex lock/unlock                           25   ns
Main memory reference                      100   ns                      20x L2 cache, 200x L1 cache
Compress 1K bytes with Zippy             3,000   ns        3 us
Send 1K bytes over 1 Gbps network       10,000   ns       10 us
Read 4K randomly from SSD*             150,000   ns      150 us          ~1GB/sec SSD
Read 1 MB sequentially from memory     250,000   ns      250 us
Round trip within same datacenter      500,000   ns      500 us
Read 1 MB sequentially from SSD*     1,000,000   ns    1,000 us    1 ms  ~1GB/sec SSD, 4X memory
Disk seek                           10,000,000   ns   10,000 us   10 ms  20x datacenter roundtrip
Read 1 MB sequentially from disk    20,000,000   ns   20,000 us   20 ms  80x memory, 20X SSD
Send packet CA->Netherlands->CA    150,000,000   ns  150,000 us  150 ms

https://colin-scott.github.io/personal_website/research/interactive_latency.html

This explanation makes sense, and would explain why I noticed a huge increase in smoothness despite the sim consuming the same amount of VRAM and system ram, because the page file was no longer being used because the system had much more virtual memory available.

[Noel 4,496]

2 hours ago, RXP said:

@Noel most likely because the amount of reserved RAM by DX11 is proportional to the amount of VRAM, regardless of use.

I see, so really it's an essentially defective way to assess and use resources such that the more you install, the more you need.  Stupid is as stupid does.  Maybe DX-12 fixes this weirdness?

Edited March 3, 2021 by Noel

[in2tech 588]

People also said during the beta testing to upgrade to 32 GB RAM and a high end graphics card just to run MSFS. And that is not true for a lot of people, like myself with what would be considered a low end system, and it runs just fine for me. Granted in large airports I might have to turn my settings down, but I also have too for other sims at times. Now VR and Ultra settings all the time is a different story. Just saying I am VERY happy with my performance on my existing Low to Medium system. Also, it appears to pick out certain systems Low or Extremely High End and give them problems such as constant CTD, which I have only seen a few times, since launch. And in most cases it was ME trying to run Ultra on a Nvidia 1060 3 GB RAM, which is just stupid, but hey I can try, it's my computer 🙂 Also learned my lesson.

i7-8700
16 GB RAM
Nvidia 1060 3 GB RAM
1 TB HDD ( I have a 256 SSD also, but it is not big enough to run MSFS, Windows, etc... and MSFS )
24" 1080 Monitor

Now I do wish I had an SSD to run MSFS on. Might buy one or just wait till I need a NEW PC!

I have no clue why it works so well for me on High End most of the time. But me happy camper as I though for sure, it was not gonna run at all!

I do hope one day that it runs smoothly for everyone though. It's so much fun if you enjoy flight simulation of any kind, from GA, to Jets, Missions, Multiplayer, etc... etc...

Yes I know a lot has to be done or 3rd parties fill the void for certain people. Patience Grass Hopper 🙂 ( not even sure that is right? )

Edited March 3, 2021 by in2tech

[RXP 2,338]

@in2tech Yes you have to always chase the balance between visuals and perf no matter what.

But this topic is about a specific question only: whether a RTX 3090 system runs better with 64GB or 32GB with FS2020 in DX11.

This might change with DX12 but I doubt for the very same reasons: FS2020 instead of the DX11 driver will maintain in RAM copies of buffer data it sends to the video card VRAM. The purpose of DX12 is for the application to directly handle this, and therefore leverage opportunities to optimize this, vs the driver forced to adopting a generic and conservative approach, but if the first 6 months of beta and the last 6 months of release are any indication on the subject of optimizations, I'd rather wait to see the final outcome.

Besides, how much RAM and VRAM an Xbox is having? These might be very good metrics about what FS2020 will be optimized for...

Edited March 3, 2021 by RXP

[Noel 4,496]

33 minutes ago, RXP said:

@in2tech Yes you have to always chase the balance between visuals and perf no matter what.

Even w/ a 3090 on board? The lowly 2070 Super 8Gb delivers liquid smooth MSFS thru really complex scenery and still have most all sliders and Ultra and about the only thing that threatens this is Ultra clouds in some scenarios, so it's already close to good enough.  My sense is when I upgrade to a better GPU I'll never need to adjust settings if the current behavior is any indicator.  I do need more VRAM for sure and more processing power for the GPU, but really performance per se is very close to a non issue, very thankfully.  And especially the main thread which in the worst places hovers around 70% (in the TBM930), but usually is bouncing around 20-40% utilization.   I hope when DX-12 arrives hopefully there will be even a little more headroom in the main thread for PMDG's debut in MSFS.  I'm astounded at how much better total performance is in MSFS over P3D 4.5.

Edited March 3, 2021 by Noel