Looking at RTX 3080 - Concerned about 10GB VRAM for DCS in VR

With the launch of the new Nvidia GPUs and impending AMD "Big Navi" GPU, I'm getting the itch to upgrade and tinker with my machine. I've already got a 2080 ti. However, I'd like use this upgrade opportunity to move that GPU to my living room HTPC.

Performance per dollar, the 3080 is looking excellent. However, the 10GB of VRAM concerns me. Especially since DCS (with Rift S) maxes out the 11GB VRAM on the 2080 ti. Also, at some point in the near future, I am likely going to upgrade my VR headset to something with a bit more clarity as more products become available. As such, I want to "future proof" (I know, that term sucks) for my next headset. For this reason, the 3090, with 24GB VRAM, looks appealing. But at 2x the price of the 3080, I just don't think it's worth it.

All that said, I'm now wondering if, for my use case (DCS in VR), going with a GPU that has lower VRAM, is potentially a step in the wrong direction. Every system has a bottleneck somewhere. Is maxed out VRAM a potential bottleneck, in this case? Or am I looking at VRAM usage, incorrectly?

...

On another note, supposedly there's a mystery 20GB 3080 SKU that's in the works to 'best' the 16GB Radeon 6900, which is still awaiting announcement. This 20GB 3080 has been leaked on a couple of different sites. Based on previous Nvidia GPU launches, I'd imagine the 20GB 3080 ("ti" or "Super") will get released sometime in Q1, 2021. Waiting for that would probably be the wisest move. But now I'm rambling and I'm curious to get thoughts from those here who are more knowledgeable on this.

Couple of threads in the VR and hardware sections on this.

Realistically I would only expect very modest performance gains going from the 2080ti to the 3080. It might be a bit better for VR. And again, the gains going from a 3080 to 3090 don't look stellar, ~10% or so on average. So far one guy has a 3080 and he claims little improvement coming from a 1080ti which is both contentious and dissapointing.

I was initially excited by the prospect of getting maybe a 50% frame boost with a 3090 vs 2080ti, but it doesn't look its gonna be the case. It might be in the 20-30% range or even less in DCS. So, I'm in holding pattern till things calm down and we see some actual DCS specific info.

I am pretty sure the 11GB is just "allocated" Whether it's all being used or not requires debugging tool to see.

If it's maxing out 11GB, majority of people are on 8GB and they'd be in trouble.

Couple of threads in the VR and hardware sections on this.

 

Realistically I would only expect very modest performance gains going from the 2080ti to the 3080. It might be a bit better for VR. And again, the gains going from a 3080 to 3090 don't look stellar, ~10% or so on average. So far one guy has a 3080 and he claims little improvement coming from a 1080ti which is both contentious and dissapointing.

 

I was initially excited by the prospect of getting maybe a 50% frame boost with a 3090 vs 2080ti, but it doesn't look its gonna be the case. It might be in the 20-30% range or even less in DCS. So, I'm in holding pattern till things calm down and we see some actual DCS specific info.

Ah crap. I figured I was likely failing at navigating this forum properly and posting in the wrong section. Apologies. I do appreciate your response, regardless.

I think I'm with you, though. When I first learned that the 3090 would have 24GB VRAM, I assumed I'd inevitably be spending the cash. But after seeing reviewers post disappointing real world improvement, compared to the 3080 on all the games I play (minus DCS), it just doesn't seem worthwhile.

I'll hold out for some videos of DCS in VR, where people will inevitably compare the 2080 ti vs 3080 vs 3090, and see what those results pan out to be.

I am pretty sure the 11GB is just "allocated" Whether it's all being used or not requires debugging tool to see.

If it's maxing out 11GB, majority of people are on 8GB and they'd be in trouble.

Solid point. I imagine 8gb vs 11gb vs 20gb will have minimal difference. But in the case of DCS with VR, where we are fighting to maintain 30fps, even with a watercooled, overclocked 2080 ti, I want to make sure my next move makes a consistent difference, even if it is minimal in the grand scheme of things.

Though, at this point I'm convinced that carrier FPS will remain at 30fps or below, even with quad 3090's in SLI.

Though, at this point I'm convinced that carrier FPS will remain at 30fps or below, even with quad 3090's in SLI.

Things like carrier, dense cities, multiplayer, and lots of AI are mostly CPU. If that's what you mostly do, and if you already have 2080Ti, GPU upgrade will not do much for you.

People who do single player campaign missions in VR would mostly see benefits from GPU upgrade.

Things like carrier, dense cities, multiplayer, and lots of AI are mostly CPU. If that's what you mostly do, and if you already have 2080Ti, GPU upgrade will not do much for you.

 

People who do single player campaign missions would mostly see benefits from GPU upgrade.

Interesting. Well maybe I should shift focus from the GPU to CPU. I've got a 3900x but the Ryzen 4000 CPU's are about to get announced and reviewed shortly after. If they provide a notable bump over the 3900x, that may be the smarter upgrade path. It would be cheaper and satisfy my desire to tinker, for now.

Yeah, DCS unfortunately is a CPU/thread hog. And its not multithreaded really so mainly you are chasing Ghz and single core performance, hence my choice for a 9700k 2 years ago. And looking a 10700 it doesn't seem like its worth it.

So one quick update, a credible source (Aurelius) does say that going from a 2080ti to a 3090 gives about a 40% boost in DCS. Not worth the price of admission for him, but hopefully a few folks will chime in once they get their cards.

Do not be wary of the 10GB VRAM.

The settings you will be using to run VR at good framerates will not be that demanding :lol:

The guy who tested his rig with a 3080 coming from a 1080ti has an older generation PCI-E 3.0 motherboard. The 3080 is a PCI-4 compatible card, but how much performance loss was due to that? Id like to know more about what his equipment might be because I think I have a similar setup.

Also, the 2080 was supposed to double as a Ti type, and the 2080Ti was supposed to double as a titan, correct? Thing with this is that the 1080Ti cards and below had more bandwidth dedicated to the PCI express lanes. Was that the same with the 2080 series non Ti cards? Thats all important when making a judgement about feasibility of upgrading...because from the sounds of it, tech from 2 years ago is severely outdated now even though it technically should be future proofed for at least 5-6 years. On that subject, DDR5 is supposed to be making an appearance soon.

The guy who tested his rig with a 3080 coming from a 1080ti has an older generation PCI-E 3.0 motherboard. The 3080 is a PCI-4 compatible card, but how much performance loss was due to that? Id like to know more about what his equipment might be because I think I have a similar setup.

 

Also, the 2080 was supposed to double as a Ti type, and the 2080Ti was supposed to double as a titan, correct? Thing with this is that the 1080Ti cards and below had more bandwidth dedicated to the PCI express lanes. Was that the same with the 2080 series non Ti cards? Thats all important when making a judgement about feasibility of upgrading...because from the sounds of it, tech from 2 years ago is severely outdated now even though it technically should be future proofed for at least 5-6 years. On that subject, DDR5 is supposed to be making an appearance soon.

From what I saw in that thread, his optimizations were non existent and testing methodology was shoddy at best. When testing card vs card, you need to be strict with ensuring all settings are correct between both, otherwise the data is useless

I thought I read somewhere that though the 3080 has 10GB's of RAM, it's GDDRX6. With compression it comes out to around 12-14GB's when comparing to a turing...

PCIe 3 vs PCIe 4 is irrelevant. The bandwidth is there and any differences ( +/- 1 FPS) are due to bus timing or measuring issues/errors.

https://www.techspot.com/review/2104-pcie4-vs-pcie3-gpu-performance/

FWIW wrt 10GB... the 2070 Super in my system is constantly pinned at 8GB in Syria and the Channel map. I very, very rarely will see stutters. The code is managing the VRAM fine it seems. When there is a stutter it is a one-off due to a terrain "scope change" and is fine I guess.

All this said, I'm still waiting for a 3080Ti/Super with 20Gb for my next card. Depending on timing I may do a new Tiger Lake CPU when it hits the desktop platform - preliminaries look good.

Yeah, don't go with 10GB for DCS World, even if DCS adds support for PCIe 4.0 DirectStorage relatively fast it'll still be like 1-2 years, but even with DirectStorage, that's just speed, when you want to have a ton of stuff visible all at once that speed won't help you, need capacity for that.

When viewing thing from afar I guess each of those would be low quality LODs so it wouldn't take up as much but I still, why go smaller when you can go bigger.

But this is just my quick opinion, I'd have to look and think about this deeper.

PCIe 3 vs PCIe 4 is irrelevant. The bandwidth is there and any differences ( +/- 1 FPS) are due to bus timing or measuring issues/errors.

 

 

https://www.techspot.com/review/2104-pcie4-vs-pcie3-gpu-performance/

 

I dont know man, the more PCIE lanes you have in use like sound cards, NVme cards, USB 3.0 cards and the like, the more performance is noted to drop. Now Im not saying you are wrong, Im just stating Id like to see a comparison between this AND a 3080 on updated hardware. And I dont mean like in the link you provided, Id rather it be DCS. DCS isnt Death Stranding.

Gamers nexus did a whole analysis of the pcie 3 vs 4 as have several other people... All come out with the same, less than margin of error differences... Yes if you have those additional peripherals then you might encounter a difference,.. I don't think they tested what would happen for example if you stuck it in an 8x slot or diecided to use all the NVME drive ports available etc...

I see where you are going but the primary PCI-e on a motherboard (slot 1) is 16 lanes/channels direct to the CPU specifically to make sure the primary graphics card is direct connected to the CPU.

All other PCI-e channels could be going through the chipset for sound, nvme, whatever. You need to look at the specific architecture of the CPU and motherboard to know where lanes/channels are going where beyond the 16 on PCI-e slot 1.

There's caveats around multi GPU and using PCI-e slot 2 as well but we aren't talking about that.

FWIW, bandwidth:

PCI-e 3.0: 8.0 GT/s, 15.8 GB/s (x16 channels/lanes)

PCI-3 4.0: 16.0 GT/s, 31.5 GB/s

HDMI 2.1 is rated 48 Gbit/s (6 GBytes/s)

Also, look at the charts in the in the article I linked. Differences are timing issues.

From what I saw in that thread, his optimizations were non existent and testing methodology was shoddy at best. When testing card vs card, you need to be strict with ensuring all settings are correct between both, otherwise the data is useless

We also need to see one run a higher screen rez standard. Not adding PD.

He's still running the OG vive. Like to see and Reverb, Reverb 2 or index have a go at the 3080.

The guy who tested his rig with a 3080 coming from a 1080ti has an older generation PCI-E 3.0 motherboard. The 3080 is a PCI-4 compatible card, but how much performance loss was due to that?

Makes virtually zero difference for GPUs, including RTX3080.

Please keep on using your PCs with PCIe 3.0 and be happy, and let that rumour die.

While it's great for most recent NVME storage (for those that want the ultimate fastest drive speeds), for now PCIe 4.0 makes no real difference for graphics or gaming performance.

While it's great for most recent NVME storage (for those that want the ultimate fastest drive speeds), for now PCIe 4.0 makes no real difference for graphics or gaming performance.

You can argue it makes no difference for NVME storage as well. Playing games on SSD vs NVME is a barely noticeable difference; 3.5s to load something vs 2.3s... doesn't matter.

4DKLA7w9eeA

If/when DirectStorage to load textures directly into VRAM becomes a thing in DCS you could consider that being a reason to upgrade if not already onto PCI-e 4 by then.

That said, that implies there isn't already buffering/caching of "next texture" in place in the code already given sytems have access to lots of RAM nowadays. <not_being_ironic_at_all />

its confirmed there will be a 20gb version of the 3080, im interested in that

Math says no difference...

With the latest MSI Afterburner you can check Vram usage per process but I don't know how accurate this is for real usage. I tested this yesterday with Rift S and my 1070Ti 8Gb Vram was full allocated but DCS used ~6Gb when sitting in Mig-21 on ground (Syria) and ~5Gb in F-18. Not tested multiplayer yet. Settings was:

Textures: High

Terrain textures: High

Civ. Traffic: Low

Water: High

Visib range: High

Heat blur: Off

Shadows: Medium

Res. of Cockpit Displays: 1024

MSAA: Off

Depth of Field: Off

Lens effect: Off

Grass: Off

Trees: 70%

Preload radius: max

Chimney smoke: 1

Anisotropic filtering: 4x

Terrain object shadows: Flat

Cockpit global illumination: Off

With low Terrain textures Vram process usage was ~1Gb less. I have ordered 3080 and it have 2Gb more vram than my 1070 + it's more faster and compressed so can't wait ;)

on my settings (below) 9.5gb

I never had a issue with my 8GB Vram either. :D

Don't know if the 20 GB version will give us extra.

I think we must wait again another 2 years, or ED will do magic and wizardry! :D

I was overwhelmed that DCS took 32 GB from 64 GB ram... I don't get that. DCS non optimized coolest game ever. Title for the book of records. :D

But i am waiting now for my RTX 3080 gaming x trio to test, play and have a lot of fun! Besides that next month is my HP reverb G2 coming in! :D

its confirmed there will be a 20gb version of the 3080, im interested in that

Unless I missed an announcement, I wouldn't say it's "confirmed". Nvidia and board partners have not officially confirmed it. But based on all the reputable leaks, I think it's safe to say it is definitely coming. With that said, I'm with you on the wait.

While I do think 10GB VRAM is adequate for almost all games, including DCS, waiting a bit longer for a 20GB 3080 seems like the smarter move. I'm going to hold off on the GPU upgrade, which is out of my control anyway, considering the complete lack of stock.

For now, I'm going to turn my focus to the 4000 series Ryzen CPUs. I know I'm leaving a bit of performance on the table with the 3900x, compared to something like the Intel 10900k.