Say, "No" to the RTX 5090 card - Definitely the generation to skip over for DCS VR

[Tensorial_Architect]

Been fooling around with the ASUS TUF 5090 card in our labs and on average, it only beats the 4090 by 21% across most major titles. (And I don't mean the hand-picked titles that Nvidia showed at their reveal presentation.) As many of you know, this card requires up to 600 watts of power (the 4090 requires up to 450) and if you are overclocking everything else, you safely need a 1200 watt PSU. It requires a 33% increase in power over the RTX 4090 to provide a 21% improvement at the screen, ... which is pure s$%! for a flagship card. And that does not even take into account the very large price bump. (The ASUS TUF we were able to secure was almost $2500 US dollars compared to the $1780 ASUS TUF RTX 4090 we procured back in Nov, 2022.)

Specifically, for DCS while using the Reverb G2, the Quest Pro, the Quest 3, the Varjo Aero, the Pimax Crystal, and the Somnium VR Ultimate Edition:

(the G2 and Q3 are using Mbbuchia's Fixed Foveated rendering mode, while the Pro, Aero, Crystal, and Ultimate are using Mbbuchia's true eye-tracked quad views)

G2 - 11.3% increase over the RTX 4090 with Fixed Foveated

Q3 - 14.8% increase over the RTX 4090 with Fixed Foveated

Q Pro - 17.3% increase over the RTX 4090 with true Quad Views/eye tracking

Aero - 14.4% increase over the RTX 4090 with true Quad Views/eye tracking

Crystal - 11.8% increase over the RTX 4090 with true Quad Views/eye tracking

Ultimate - 13.4% increase over the RTX 4090 with true Quad Views/eye tracking

Forum member Flextremes below mentioned that the 5090 should respond well to the 4k+ pixel heavy demands of VR with the card really showing its muscle with the high-end headsets and resolutions (think Aero, Crystal, and Somnium). I expected the same but sadly, I have not found that to be true.

Of course, DCS World is not a game which currently offers DLSS 4 which the 5090 is custom built to made to take advantage of. Still, imho, this is the worst generation of flagship cards in the last decade. Although we don't have one in our labs, it seems the 5080 will not beat the 4090 in anything. I do not recommend the RTX 5090 for DCS with VR.

(Some will ask what settings I am using - In short, ... everything maxed to the hilt. Mbbuchia's Quad Views does not work well with Lens Effects or Bloom, so those are off. Except for that, I've got everything, ... shadows, water, terrain, clouds, etc ... maxed at high or ultra. I am using DLSS with AF 16X. I am able to fly with the Quest Pro at 72 fps. To get that or higher with the Aero, Crystal, or Somnium, I have to dial back some things if I run at those respective headset's native resolutions. The very, very best combination of detail  and definition and color gamut is through the Somnium VR1 Ultimate but very, very, few are ever going to be able to shell out $4300 for a headset. A tiny, tiny, smidgen down from there for absolute best is the Aero. Overall though, ... and quite surprisingly given its low native resolution, ... the high-end sweetspot imo is the Quest Pro. By dialing up the resolution (through the desktop Link software) and using Quad Views, you get 95% of the eye candy of the Aero, Crystal, or Ultimate at a far cheaper price. Temperature wise, the 5090 tends to run about 4 deg C higher on average than the 4090. After a good hour of flying the Hornet in Syria or Dubai, I had GPU temps of 74 deg C and memory temps of 84 to 85 deg C (the RTX 4090 sitting in my second rig under my office desk averaged about 69.5 deg C on the GPU and about 77 deg C on the memory in the same situation with the room temperature set at 72 deg F). The actual work being done by the card (wattage is electrically speaking, work) averaged 570 watts over this one hour. From a noise perspective, the 5090 was the ASUS TUF card I used was almost identical to the ASUS 4090 TUF version. I have not overvolted the 5090 card yet, nor have I overclocked it.)

(Edit - Had time to today to add the Aero and Crystal to the results)

Edited February 7 by Tensorial_Architect

[EvilBert VR]

Hello Architect,

I really do think your post is confusing a little bit misleading and partially off from general thinking, addressing "only" current 4090 users.
It might be true for people considering about upgrading from a 4090 to the 50x cards but (like in my case when looking at my signature) it is going to be a HUGE gain of performance! Even if it was "...only by 21%..." compared to the 4090.

The same words were spoken back then when the 4090 was about to be released. But it turned out the 4090 was the thing to go for. Back then I decided to skip the 4090 and wait for the next gen so the 5090 is the thing to go.

So: Is it "worth" it? Moneywise? Performancewise? That is up to every single individual. Is it worth it for DCS? No doubt about it!

I was lucky to get a 5090 and I am looking so forward to its arrival within the next few days. Also with the MeganeX and the Somnium Ultimate inbound by March I am so excited about what I might experience compared to my current card. 
My XTal has pushed my card beyond its capabilities and I am very confident that is soon being changed by utilizing the 5090.

So saying "No" to the 5090 is a "No"!

Cheers!
 

[Tensorial_Architect]

You are speaking of using VR headsets priced in the thousands and then combining it with a card priced at roughly $2400 or more. Congratulations, ... you are either a surgeon, another speciality of medicine practitioner, a dentist, a pro athlete, or a highly paid lawyer, or a trust fund baby. You Sir have succeeded! .... Congrats

(Clarification note - When the 4090 was released, it was outperforming the 3090 Ti by roughly 70% across many titles. It was apparent early on that the RTX 4090 was going to deliver the "goods." In some titles it doubled the frame rates offered by the 3090 Ti, ... in others, it beat the 3090 Ti by only 45%. It was a true jump up. The 5090 is the complete opposite. For the massive electricity bill it will take to run it, it is worse than the 4090.)

Coming back down to reality, ..... your situation specifically EvilBert, ... even for VR users, ... is less than one in 10,000. You can take a look at the Steam VR headset usage for December, 2024 here:

 

As you can see, the Xtal is not even listed.

(Please note as well that only about 2.3% to 3% of U.S. based Steam users are even reporting a VR headset. An Xtal owner is like finding a 3 carat diamond in a coal field.)

My only request, ... when you drive by in your Bugatti with your Xtal sitting nicely on the passenger seat, .... could you throw a Benjamin or two out the window? 

Oh, ... and why EvilBert? ... Why not Good Bert? .... The world is filled with bad people and far too few Nelson Mandelas and Jonas Salks.

Edited February 3 by Tensorial_Architect

[freehand]

Off topic are you posting under another user name ?

[Tensorial_Architect]

I post across several forums and you can always find the real me by my signature; A wonderful method for appreciating the beauty of the Multiverse is to learn the language in which it was written, .... mathematics.

And that is the truth. Want to understand our reality at its deepest levels, ... look first to particles, ... then picture fields (quantum fields), .... then Information Theory renormalizations of quantum fields (and/or equivalently, ... stress-energy tensors from GR), ... then past ennui (along with strings and branes), ... and finally, ... as T.S. Eliot so eloquently wrote, .... and the end of all of our exploring, we will arrive at where we started, ... and know the place for the first time.

I leave or join forums often epheremally, .... usually to try and help others as my own unusual neural connectomes dictate.

Edited February 3 by Tensorial_Architect

[Phantom711]

@Tensorial_Architect

You got a pretty beefy system yourself…including that latest 5090. making assumptions on someone else‘s financial background based on the VR headset he chose/chooses, is pretty silly in my opinion. 
Fair enough, it was probably meant in good fun.

Even with (just) a QuestPro and a 4090 it‘s easy to run into ASW occasionally. The 20-30% performance increase of a 5090 might eliminate that. High res VR headsets would benefit even more so. 
 

Having a 4090 already, I will skip the 5090, but @EvilBert VR has a valid point there, that the 5090 is still a great upgrade for people with weaker 40-series cards or below.

[Tensorial_Architect]

As many have observed, it seems that it is every 2nd generation of Nvidia cards that gets it "right." 1080 ti, ... meh, .... 2080 ti, ... right. 3090 ti, ... meh, .... 4090, ... right. 

Following in that pattern, .... 5090, .... pass. 

And you are perfectly correct Phantom, .... I often joke but using sarcasm. Sarcasm is a much misunderstood thing in the Marvel movie and literal era of the present United States. That is why my passage through any forum has a measured half-life

Edited February 3 by Tensorial_Architect

[EvilBert VR]

1 hour ago, Tensorial_Architect said:

You are speaking of using VR headsets priced in the thousands and then combining it with a card priced at roughly $2400 or more. Congratulations, ... you are either a surgeon, another speciality of medicine practitioner, a dentist, a pro athlete, or a highly paid lawyer, or a trust fund baby. You Sir have succeeded! .... Congrats

(Clarification note - When the 4090 was released, it was in many titles performing at 170% of the rate of the 3090 Ti. It was apparent early on that the RTX 4090 was going to deliver the "goods." In some titles it beat the 3090 Ti by 200%, ... in others, by only 145%. It was a true jump up. The 5090 is the complete opposite. For the massive electricity bill it will take to run it, it is worse than the 4090.)

Coming back down to reality, ..... your situation even for VR users is less than one in 10,000. You can take a look at the Steam VR headset usage for December, 2024 here:

 

As you can see, the Xtal is not even listed.

My only request, ... when you drive by in your Bugatti with your Xtal sitting nicely on the passenger seat, .... could you throw a Benjamin or two out the window? 

Oh, ... and why Evil Burt? ... Why not Good Burt? .... The world is filled with bad people and far too few Nelson Mandelas and Jonas Salks.

Architect,

I'd appreciate it you were focusing more on topic and on what I stated than presuming what I do for a living or being cynical about cars driven etc. (Can't find the occasion in my post giving you a reason to do so.) Like @Phantom711 assumed, it might has been a funny approach.

Frankly, your post is giving your opinion that I respect. But as I quoted I do not agree with your title stating "NO".

I deeply respect your technical approach and I am glad having guys like you around that give scientific views but you just disregard the personal needs and views of random users along their subjective points of view being in (systemwise) situations like mine.

As I said: Your chosen title of the thread (along with your post) claims that your assumptions are a fact. And that is what am not consent with.

Cheers
 

[Tensorial_Architect]

But why Evil? 

Wouldn't it be better to be known as GoodBert?

Edited February 3 by Tensorial_Architect

[Phantom711]

@Tensorial_Architect

I do speak „sarcasm“ but it‘s hard to tell from a comment from a guy on a forum that I don‘t know. But I gave you the benefit of doubt…

[Tensorial_Architect]

Thanks Phantom!

Also guys, ... don't read too much into anything other than the results I measured.

If you have a 3080 card and it is not doing it for you or you always hated not being able to max out X or Y headset, ... and you now have $2400, .... then great, .... get a 5090! My only point is that if you already have a 4090, .... mehh, ... not much of an upgrade.

I think many were hoping that the 5090 would be this monster, ... just like the 4090 was. A card that would outperform the 4090 by 50% or more. Sadly, Nvidia's cycle continues on ....

 

Edited February 3 by Tensorial_Architect

[EvilBert VR]

5 minutes ago, Tensorial_Architect said:

But why Evil? 

Wouldn't it be better to be known as GoodBert?

Ya know, sometimes names from the past stick 

[RealDCSpilot]

@freehand Yeah, you are right. I don't remember the name, but his posts are exactly the same "quality". aka misleading... not from planet earth... etc.

However, everyone upgrading from anything lower than a 4090/4080 definitely benefits from the gain of raw performance when buying the 5090.

Edited February 3 by RealDCSpilot

[warmachine79]

So you got a 5090 working with a Pimax Crystal Light? There are reports in Discord that Pimax HMDs do not show a picture when connected to a Blackwellcard. Also confirmed by Quorra which is some Pimax official.

Edited February 3 by warmachine79

[Melv428]

thanks for the testing and results. As 4090 owner and enthusiast i tried all the time to find a reason to buy the 5090 but yes its totally unnecssesary for that price

[okopanja]

3 hours ago, Tensorial_Architect said:

You are speaking of using VR headsets priced in the thousands and then combining it with a card priced at roughly $2400 or more. Congratulations, ... you are either a surgeon, another speciality of medicine practitioner, a dentist, a pro athlete, or a highly paid lawyer, or a trust fund baby. You Sir have succeeded! .... Congrats

Actually, I do believe the most likely to buy it are people with more limited income (credit pit).

The people you mentioned spend their free time in a different ways, usually more quality time.

[Aapje]

2 hours ago, Tensorial_Architect said:

I think many were hoping that the 5090 would be this monster, ... just like the 4090 was. A card that would outperform the 4090 by 1.5 to 2X. Sadly, Nvidia's cycle continues on ....

I was hoping that the extra bandwidth would do more. However, it was always unlikely to be such a big gain, because the 30-series was made on a very poor (but cheap) node, and the 40-series benefited a lot in raw performance and efficiency from going back to TSMC. In contrast, the 50-series does not even benefit from a single node improvement.

[Hippo]

17 hours ago, Tensorial_Architect said:

I do not recommend the RTX 5090 for DCS with VR.

A bit of a blanket statement there.  I have a 3080Ti, what would you recommend I buy?  4090s are virtually unavailable and extremely expensive.  It'll be two years until the next gen.  5080 or anything below not enough of an increase to be worth bothering with.  I will probably wait around six months and hope for 5090 supply and prices to stabilise.
 

Quote

 

"it only beats the 4090 by 21%"

"it (4090) was in many titles performing at 170% of the rate of the 3090 Ti"

 

For consistency, shouldn't you say that it beat the 3090 Ti by 70%?  Or that the 5090 performs at 121% the rate of the 4090?  Otherwise you run the risk of appearing to present numbers in a way that supports a particular narrative.

Thank you for your post and the performance numbers on VR headsets, very informative.  It's nice to finally see 5090 perf numbers for DCS / VR.

 

Edited February 3 by Hippo

[Tensorial_Architect]

Yes, Hippo, you are right. The 4090 often beat the 3090 Ti by 70%. (In this same manner, the 5090 is beating the 4090 by about 21% (averaged).)

To my mind, I am always used to setting the previous flagship card (in this case, the 3090 Ti) to 100%, and measuring from there. 

But yes, you are exactly (mathematically) correct. Let me fix that in a few hours at lunch.

Edited February 3 by Tensorial_Architect

[SharpeXB]

Seems like a fair observation that upgrading to a 5090 isn’t too worthwhile for someone that already owns a 4090. But for anything less it could be. Another thing to consider is that the market for used 4090s is quite robust. I see bids that are actually higher than what I paid for mine although I doubt that will continue when the 5090s become more available. Still the real upgrade cost includes your trade-in. 

[Flextremes]

20 hours ago, Tensorial_Architect said:

Been fooling around with the ASUS TUF 5090 card in our labs and on average, it only beats the 4090 by 21% across most major titles. (And I don't mean the hand-picked titles that Nvidia showed at their reveal presentation.) As many of you know, this card requires up to 600 watts of power (the 4090 requires up to 450) and if you are overclocking everything else, you safely need a 1200 watt PSU. It requires a 33% increase in power over the RTX 4090 to provide a 21% improvement at the screen, ... which is pure s$%! for a flagship card. And that does not even take into account the very large price bump. (The ASUS TUF we were able to secure was almost $2500 US dollars compared to the $1780 ASUS TUF RTX 4090 we procured back in Nov, 2022.)

Specifically, for DCS while using the Reverb G2, the Quest Pro, the Quest 3, the Varjo Aero, the Pimax Crystal, and the Somnium VR Ultimate Edition:

(the G2 and Q3 are using Mbbuchia's Fixed Foveated rendering mode, while the Pro, Aero, Crystal, and Ultimate are using Mbbuchia's true eye-tracked quad views)

G2 - 11.3% increase over the RTX 4090 with Fixed Foveated

Q3 - 14.8% increase over the RTX 4090 with Fixed Foveated

Q Pro - 17.3% increase over the RTX 4090 with true Quad Views/eye tracking

Aero - 14.4% increase over the RTX 4090 with true Quad Views/eye tracking

Crystal - 11.8% increase over the RTX 4090 with true Quad Views/eye tracking

Ultimate - 13.4% increase over the RTX 4090 with true Quad Views/eye tracking

Of course, DCS World is not a game which currently offers DLSS 4 which the 5090 is custom built to made to take advantage of. Still, imho, this is the worst generation of flagship cards in the last decade. Although we don't have one in our labs, it seems the 5080 will not beat the 4090 in anything. I do not recommend the RTX 5090 for DCS with VR.

(Some will ask what settings I am using - In short, ... everything maxed to the hilt. Mbbuchia's Quad Views does not work well with Lens Effects or Bloom, so those are off. Except for that, I've got everything, ... shadows, water, terrain, etc at high or ultra. I am using DLSS with AF 16X. I am able to fly with the Quest Pro at 72 fps. To get that or higher with the Aero, Crystal, or Somnium, I have to dial back some things if I run at those respective headset's native resolutions. The very, very best combination of detail  and definition and color gamut is through the Somnium VR1 Ultimate but very, very, few are ever going to be able to shell out $4300 for a headset. Overall, the high-end sweetspot imo is the Quest Pro. By dialing up the resolution (through the desktop Link software) and using Quad Views, you get 95% of the eye candy of the Aero, Crystal, or Ultimate at a far cheaper price. Temperature wise, the 5090 tends to run about 4 deg C higher on average than the 4090. After a good hour of flying the Hornet in Syria or Dubai, I had GPU temps of 74 deg C and memory temps of 84 to 85 deg C (the RTX 4090 averages about 69.5 deg C on the GPU and about 77 deg C on the memory in the same situation). The actual work being done by the card (wattage is electrically speaking, work) averaged 570 watts over this one hour. I have not overvolted the 5090 card yet, nor have I overclocked it.)

(Edit - Had time to today to add the Aero and Crystal to the results)

As a Q Pro owner with a 4090 really appreciate this data. Thank you!

I completely agree on the visual fidelity of the Quest Pro: I prefer it over my Pimax Crystal Super. 

However, I am super curious to the 1%, 0.1% lows and/or minimum framerates. I was hoping the 5090 would make a bigger than 30% difference over the 4090.

I'm surprised by your 5090 results btw: I expected much better results (even better than average 30% raster performance seen in 4k games), because we are rendering DCS in VR in massive resolutions, way higher than regular 4K: and I expected the 70% increase in GPU mem bandwidth to really help.  

 

Edited February 3 by Flextremes

[Tensorial_Architect]

4 hours ago, Flextremes said:

As a Q Pro owner with a 4090 really appreciate this data. Thank you!

I completely agree on the visual fidelity of the Quest Pro: I prefer it over my Pimax Crystal Super. 

However, I am super curious to the 1%, 0.1% lows and/or minimum framerates. I was hoping the 5090 would make a bigger than 30% difference over the 4090.

I'm surprised by your 5090 results btw: I expected much better results (even better than average 30% raster performance seen in 4k games), because we are rendering DCS in VR in massive resolutions, way higher than regular 4K: and I expected the 70% increase in GPU mem bandwidth to really help.  

 

Like you, I was hoping for at least a 30% improvement, .... especially for $2500. (Actually, for $2500, I was expecting a real-world, fully nude, 32-22-32, 5'8", 114 lb, cheerful, no-STD, attractive faced, intelligent, female to ring my doorbell and help "install" the card along with other "services." I managed to get through to Jensen, ... and he informed me that due to heavy demand, ... she is coming, ... but on Thursday, .... (m$%##^$#@$@!). .... I mean the nerve of these guys to pull a stunt like that. Three dreaded days of waiting, ... . Future 5080 buyers take note, ... you only get the bobblehead doll of her, ... not the "services" package.)

Joking aside, I had the same hopes. Sadly, that is not the results I measured. That said, the 5090 will pick up real benefits if like Cyberpunk 2077, DCS adds DLSS4 support down the road.

Edited February 3 by Tensorial_Architect

[Notarobot]

Not sure how much of this is true ... but wasn't there a similar thing with the 40 series (fixed by driver update)

 

[Tensorial_Architect]

 

13 hours ago, warmachine79 said:

So you got a 5090 working with a Pimax Crystal Light? There are reports in Discord that Pimax HMDs do not show a picture when connected to a Blackwellcard. Also confirmed by Quorra which is some Pimax official.

Perhaps I got lucky? My Crystal has been, ... shall we say, ... temperamental, ... to say the least. On Sunday, she was not. (Thank you Zeus and Athena, ... ohh, ... and Sotha Sil.) I usually sacrifice a small goat before I fire up my Crystal, ... no goat needed this time, ..... YAY!

(I will say for all the math, physics, and engineering nerds out there like myself who will get this joke, ... I am endowed with an anti-Pauli field. After many years of experience, ... things in a lab that rarely work, ... often work for me. The opposite of the well-noted Wolfgang Pauli (my 2nd favorite physicist) effect.)

Edited February 3 by Tensorial_Architect

[Dragon1-1]

6 hours ago, Tensorial_Architect said:

Yes, Hippo, you are right. The 4090 often beat the 3090 Ti by 70%. (In this same manner, the 5090 is beating the 4090 by about 21% (averaged).)

Are both of those numbers with or without framegen? Because if half of that 70% is fake frames, it'd look a lot less impressive. For what it's worth, DCS doesn't even support frame generation.