MSFS forums claiming 64GB of Ram helps on 3090s

[Noel 4,499]

12 minutes ago, MrFuzzy said:

I put additional 32 GB and could not see any tangible difference honestly, but I am digging the extra RGB lights in the case so I will keep them.

The whole thing is a crock IMO.  You triple your VRAM, and the software is so screwed up it needs more system ram when you hardly see the top 1/2 of your mega expensive 3090's VRAM ever used.  Oh, yeah you can dial up render scaling until it squeals.  Look at the two posts below, and tell me which one was done w/ render scaling at 130, versus 100.  Much ado about nonsense.  Wait for DX-12 before wasting your time, effort and $$ with this.  Just MO of course, and I'm sticking with it.

Edited March 3, 2021 by Noel

[spacedyemeerkat 421]

10 minutes ago, MrFuzzy said:

...but I am digging the extra RGB lights in the case so I will keep them.

I love that! The older I get (48), the more I like my case to be lit up like a Christmas tree. Looking forward to getting a new PC built later this year and going full on with lighting. People are going to think UFOs are landing, I tell ya!

[Noel 4,499]

17 minutes ago, RXP said:

I believe if this is how it runs on your system then don't bother. However if someone could lend you 32GB to try out why not just trying?

My build done a year ago is rock solid, runs w/ or w/o HT enabled at 5ghz and cool, no crashes to desktop, etc.   I don't want to mess w/ memory timing when everything is alread perfect, and really, for what?  

[RXP 2,338]

1 hour ago, Noel said:

The whole thing is a crock IMO.  You triple your VRAM, and the software is so screwed up it needs more system ram when you hardly see the top 1/2 of your mega expensive 3090's VRAM ever used.

I understand it is hard to see how this could be possible. I don't know how experienced you are coding 3D engines in general or DX11 specifically, but it is not unusual allocating and keep RAM copies of vertex buffers for example, because this is the way the API is meant to be used. If the game is pushing mush more data to VRAM because it can and has space for it, inherently this will therefore also consume more RAM as well. And please don't forget any RAM<->VRAM transfer is done via DMA (or its more modern variants) and for this, it is very effective for the OS and the drivers to "reserve" a shared address space. This is really a simplification but it is just to point out there are both API use level incidence to using more RAM when using more VRAM, and there is also a driver/OS level incidence to it.

For example, here are some strategies to updating buffers (I just did a quick search for illustrations, I'm not saying these are the best practices):

Quote

The question I have is whether you require performance or the convenience of having the data in one location?

There are a few configurations you can try.

1. Set up your Buffer to have both CPU Read and Write Access. This though mean you will be pushing and pulling your buffer up and down the bus. In the end, it also causes performance issues on the GPU such as blocking etc (waiting for the data to be moved back onto the GPU). I personally don't use this in my editor.
2. If memory is not the issue, set up a copy of your buffer on CPU side, each frame map with Discard and block copy the data across. This is performant, but also memory intensive. You obviously have to manage the data partioning and indexing into this space. I don't use this, but I toyed with it, too much effort!
3. You bite the bullet, you map to the buffer as per 2, and write each vertex object into the mapped buffer. I do this, and unless the buffer is freaking huge, I havent had issue with it in my own editor.
4. Use the Computer shader to update the buffer, create a resource view and access view and pass the updates via a constant buffer. Bit of a Sledgehammer to crack a wallnut. And still doesn't stop the fact you may need pull the data back off the GPU ala as per item 1.

There are some variations on managing the buffer, such as interleaving you can play with also (2 copies, one on GPU while the other is being written to) which you can try also. There are some rather ornate mechanisms such as building the content of the buffer in another thread and then flagging the update.

At the end of the day, DX 11 doesn't offer the ability (someone might know better) to edit the data in GPU memory directly, there is alot shifting between CPU and GPU.

https://stackoverflow.com/a/57317534

 

PS: I'm not trying to convince anyone either! I'm just exposing an empirical finding which started from cross-testing with 2 other people with 3090 how and what to push on the 3090 vs the 3070, and this came up as an interesting discovery on their systems with their settings. Digging technical docs and knowledge, I devise a probable explanation only. If this doesn't make any difference on your system it is possible, if it does though, well, your gain!

Edited March 3, 2021 by RXP

[flightsimJim 3]

On 2/27/2021 at 8:40 AM, hanhamreds said:

What he said👍

I've ploughed 4.5k into this - no more!

64gb no way - the game is massively CPU bottlenecked and has real issues under the hood imo.

I'm now running at about 30fps at almost any setting other than low with render scaling at 1440p but set to minimum - low at 2k still 30fps.

Yet on any other game it benchmarks on par with any review or YT example, in addition it benchmarks above 99% of anything ever recorded in the latest 3D Mark. Lastly I can run my CPU fixed at 4.3ghz all cores, or 4.6 one core, cycle system memory speeds all which have no affect on frame rate.

 

 

What are your specs?

My specs: 2080ti, i9-9900K, 32GB Ram 

My sim can cruise at 60 FPS no problem on ULTRA HIGH settings.  I only fly VR now so have had to lower the settings a bit to medium...but medium in VR kicks word not allowed on ultra high settings with a monitor so I don't care. 

[Glenn Fitzpatrick 3,798]

4 hours ago, spacedyemeerkat said:

I love that! The older I get (48), the more I like my case to be lit up like a Christmas tree. Looking forward to getting a new PC built later this year and going full on with lighting. People are going to think UFOs are landing, I tell ya!

 

I actually turned off the RGB on my ASUS motherboard, Their Aura software had the Win10 lighting service continually sitting on 15% CPU use which for me was almost an entire core.

 

 

[Maxis 1,864]

18 hours ago, Glenn Fitzpatrick said:

 

Fair enough. I am sticking with 1440p for the time being myself.

It seems to me the people seeing genuine and worthwhile benefits out of 64GB ram are running a 3090 on a hi res VR headset like the Reverb G2 (with 2 displays at 2160x2160 each and WMR/3D software) .

 

Yep my system will never get to that level because i have no interest in VR at the moment so what im more interested in is if we will eventually get off the limited by mainthread problem and can get the sim to not be so bottlenecked by the cpu in scenery intensive areas at the 2k ultrawide resolution. That's where im hoping optimization and better core usage/scheduling and direct x 12 comes in. I don't yet think (for most people) we are at the point where the sim is loading up massive amounts of textures to fill up all that mem .. (unless of course you've hard modified the draw distance to the max and are running 4k ultra with multiple screens).

Of course we have yet to have a major PMDG level ac addon and also have yet to see maximum levels of AI traffic at major resource heavy airports. When those addons come out and i find the sim wanting is when i think ill need to re evaluate what next to buy. I would rather hope this system holds up at least till December though..

Previous config 3700x /32gb  ram / RTX 2080 Super / 3440x1440p ultra

Present config 5900x /32gb ram / 6800XT / 3440x1440p ultra/ render scaling increased by 20

I cant see the point if the sim is running smooth and is typically on song above 40-50 fps 80% of the time and mid 30s in the New York area. Whats left to improve until something major gets released ?

Edited March 4, 2021 by Maxis

[MrFuzzy 2,118]

8 hours ago, Glenn Fitzpatrick said:

 

I actually turned off the RGB on my ASUS motherboard, Their Aura software had the Win10 lighting service continually sitting on 15% CPU use which for me was almost an entire core.

 

 

With iCue I set the lights as I wanted (all set to "Rainbow") and uninstalled it. The lights stick to their settings.

[MrFuzzy 2,118]

10 hours ago, RXP said:

I understand it is hard to see how this could be possible. I don't know how experienced you are coding 3D engines in general or DX11 specifically, but it is not unusual allocating and keep RAM copies of vertex buffers for example, because this is the way the API is meant to be used. If the game is pushing mush more data to VRAM because it can and has space for it, inherently this will therefore also consume more RAM as well. And please don't forget any RAM<->VRAM transfer is done via DMA (or its more modern variants) and for this, it is very effective for the OS and the drivers to "reserve" a shared address space. This is really a simplification but it is just to point out there are both API use level incidence to using more RAM when using more VRAM, and there is also a driver/OS level incidence to it.

 

Yes but I guess that these RAM and VRAM usages would be detected by FS developer's mode overlay, by RTSS and by task manager, but that's not the case.

What happens is: you fly over NY or London with 11-12 GB of your RTX 3090's VRAM occupied (50% headroom) and 16-17 GB of RAM occupied (75% headroom), let's not even mention 14 extra CPU threads that sit at 0-10% on a 10850K (6 threads well used of which only one hammered), and with all that headroom you still can have stutters, objects popping in, terrain morphing as you fly over it, white/black seams that appear and disappear on mountains and water.

I am exaggerating, with my system all these issues do not manifest often and I'm having a smooth and almost flawless experience, but I feel that all the power at my disposal is not quite well used, and that is due to flaws in MSFS engine.

 

Edited March 4, 2021 by MrFuzzy

[fogboundturtle 2,708]

I wonder if AMD risizeable bar will help

[MrFuzzy 2,118]

19 minutes ago, fogboundturtle said:

I wonder if AMD risizeable bar will help

nVidia has it too. It will help in all GPU limited scenarios.

[G-RFRY 1,711]

On 3/3/2021 at 12:10 AM, CYXR said:

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

Samsung Has Troubles Restarting Texas Fab: Chip Shortages to Get Worse | Tom's Hardware

If your thinking of getting more storage SSD get in soon prices could start climbing.

Edited March 4, 2021 by G-RFRY

[akita 429]

42 minutes ago, MrFuzzy said:

 

Yes but I guess that these RAM and VRAM usages would be detected by FS developer's mode overlay, by RTSS and by task manager, but that's not the case.

What happens is: you fly over NY or London with 11-12 GB of your RTX 3090's VRAM occupied (50% headroom) and 16-17 GB of RAM occupied (75% headroom), let's not even mention 14 extra CPU threads that sit at 0-10% on a 10850K (6 threads well used of which only one hammered), and with all that headroom you still can have stutters, objects popping in, terrain morphing as you fly over it, white/black seams that appear and disappear on mountains and water.

I am exaggerating, with my system all these issues do not manifest often and I'm having a smooth and almost flawless experience, but I feel that all the power at my disposal is not quite well used, and that is due to flaws in MSFS engine.

 

Because DX11 can't do parallel rendering, you can have 100 cores, only one will send draw calls to your GPU.

Like when driving a car, it's nice to have 5 paths on your road, means more room for more cars and less dense traffic, but at the end of the road, all those 5 lines of cars will have to make it into a single one causing a traffic jam.

 

 

 

[Noel 4,499]

11 hours ago, RXP said:

PS: I'm not trying to convince anyone either!

You're a patient soul and thank you for the add'l explanation of the idea.  In the end however, I think it will turn out to be irrelevant, especially when DX12 arrives for MSFS.  

[spacedyemeerkat 421]

9 hours ago, Glenn Fitzpatrick said:

I actually turned off the RGB on my ASUS motherboard, Their Aura software had the Win10 lighting service continually sitting on 15% CPU use which for me was almost an entire core.

That's interesting. I have an ASUS motherboard but don't use Aura, and iCue doesn't appear to be using any monitorable CPU resource. Given ASUS's prominence in this particular market space, it's hard to see how they could release a product draining CPU time in the way you describe. (I am not saying it isn't true!)

Edited March 4, 2021 by spacedyemeerkat