The 4090 will perform beautifully provided you have a top-end AMD 7000 series CPU, allegedly.
(But then, what CPU have you teamed up with your 3090ti? It does make a difference)
Nvidia hire FDev to pushing 40 series?!!...
- Thread starter Lulzwhat
- Start date
(But then, what CPU have you teamed up with your 3090ti? It does make a difference)
I recently noticed what a difference the CPU can make as I upgraded from an i7 2600k to an i9 12900k while keeping my 1660Ti. On foot improved tremendously and in on foot CZ I went from low framerates with input lag to a smooth 30-40 frames while increasing settings from high (with ultra textures) to ultra at 1080p.
Riva Tuner Statistics Server together with MSI Afterburner can give a very nice on screen display giving access to a lot of information about the CPU and GPU performance. I'd say that if the GPU isn't running at say 98%+, then the game is CPU bound and changing the graphics card won't help much, except it might let you run better at higher resolutions.
Seemingly a 4090 is often under utilized (in most games) if playing at 1080p, but might make sense if playing at 1440p or 2160p. I assume that it would also be helpful for VR resolutions.
Riva Tuner Statistics Server together with MSI Afterburner can give a very nice on screen display giving access to a lot of information about the CPU and GPU performance. I'd say that if the GPU isn't running at say 98%+, then the game is CPU bound and changing the graphics card won't help much, except it might let you run better at higher resolutions.
Seemingly a 4090 is often under utilized (in most games) if playing at 1080p, but might make sense if playing at 1440p or 2160p. I assume that it would also be helpful for VR resolutions.
![BigBANGtheory [VR]](https://forums-cdn.zaonce.net/data/avatars/m/71/71385.jpg?1644873788){kind=avatar}
Ok PC games get more graphically demanding over time everyone accepts that and it's how progress is made.
If we apply your observation to Elite Dangerous which in fairness does seem to play out the way you descried... What we are saying is a game released in 2014 needed the GPU grunt from a 1080Ti to get VR working to a good standard, then along comes Odyssey 6/7yrs later with 'graphical improvements' but hardly a new engine from the ground up and certainly not massive gaming changing improvements mostly an extension to Horizons.
We now need a GPU that costs 3X the price and delivers 300% the performance to maintain the same framerate/frametime with what is subjectively small graphical enhancements and extended gameplay. We aren't talking about Unreal Tournament > Crysis graphics here this is an expansion to what is now quite an old game.
There is something not right with the Cobra engine, Odyssey has bolted on new features but its come at a tremendous performance penalty with little return on the higher screen resolutions you can't simply categorise it as the game evolving and being more demanding the scale is way off. Even if we ignore the on foot elements Odyssey is not VR friendly there are big issues still.
hardly a new engine?
The gfx is totally overhauled in Odyssey compared to Horizons.
(i may like more the slightly cartoony look of Horizons, but undeniably Odyssey is way more detailed)
Anyway, Now you are happy how Horizons (a game launched in 2015) plays on 2020 hardware.
Along the same line, you will be happy regarding how Odyssey will run on 2026 hardware
You are not wrong but you also need to consider that adding new features to a 6 years old game doesn't make it any easier. And that FDEV does some very unique things with their engine. And that it's not a proprietary engine like Unreal which gets developed with unlimited amounts of money.
If you consider these things it's no surprise that it doesn't run so fast. It would probably run much better if they could develop the entire game from scratch but I guess that's not an option.
but that is ridiculous situation, I mean I'd get it if say Odyssey was pushing photorealistic graphics using ray traced lighting effects i.e. there was a massive jump in visual fidelity but at best Odyssey is 'a bit more realistic' 'a little better' yet the performance penalty is massive. It all points to a problem with the game engine and how its taking too long to render frames.
Yes it probably is caused by Frontier building on something that was ultimately coded 10yrs ago and isn't leveraging or optimised on modern PC hardware properly I get that, its also not my problem to solve is how I see it. If it doesn't impact you then great no big deal, but if it does its borderline game breaking.
Saltiness aside if I end up getting one of these Big Bertha GPUs I'll report back on the facts and values, if it works then yes that's ultimately a good thing for the owners I guess.
Actually using an Intel Core i9-12900KF, so CPU certainly isn't the bottleneck. CPU is at about 15-20% while playing but the 3090 Ti runs at 85-90% average with 100% spikes when in graphically intensive areas. Gets nice and toasty in my office.
The performance decrease between Horizons (3.8) and Odyssey in VR is far beyond any increase in visual fidelity. Odyssey looks better than Horizons, in my opinion, but as we know Odyssey is even worse when it comes to anti-aliasing, and it looks awful when having to decrease quality just to be able to use VR at a non-nauseaiting framerate. I get that a Reverb G2 is challenging graphically, it is 2160x2160 x 2, so it's effectively 4K with additional VR overhead, but the performance delta between two versions of the same game is disappointing.
Other VR games don't have this problem, so it's not a "well, that's just VR for you". That you have to throw the latest and greatest hardware at a game for it to be playable points to serious game engine issues.
The performance difference for me (on good hardware) is between 20 & 30% (I don't have lower spec hardware to test on, so can make no comment) - better than at launch, certainly, but not as good as I feel it ought to be.
the only other VR game that comes close, from my collection, is NMS - which was 'fixed' by dropping resolution & textures to that on the PSVR apparently - I've not bothered much with NMS in VR (or pancake) much recently, but don't think it has improved noticeably. So, no, it isn't just VR... It is inefficient code (allegedly poorly allocating resources, leading to some threads becoming saturated and everything else waiting for it to clear)
Tried with a new RTX 4090. While it is better than the 3090 Ti, it's still not great and nowhere near Horizons performance.
I've bypassed Steam VR using OpenComposite to remove that overhead which has proven helpful. But even just using mid to high graphics settings with HMD 1.0/SS 1.0 it doesn't provide a steady 90 fps in more demanding areas. What is really annoying is when it drops frames and stutters away the GPU isn't even running at more than 60% load at any time. It's like it has hit a wall, similar to being CPU limited, but my CPU is at 5-10%. So it's not even making use of the hardware available to it. Seems that VR performance in the Odyssey graphics engine is bad no matter how much hardware you throw at it.
I did try it in Horizons 3.8 with everything cranked up, and it looks amazing and is buttery smooth. Oh well.
Note for those who haven't read the entire thread. This is on an HP Reverb G2 playing Odyssey, your mileage may vary on lower resolution headsets and/or Horizons.
I've bypassed Steam VR using OpenComposite to remove that overhead which has proven helpful. But even just using mid to high graphics settings with HMD 1.0/SS 1.0 it doesn't provide a steady 90 fps in more demanding areas. What is really annoying is when it drops frames and stutters away the GPU isn't even running at more than 60% load at any time. It's like it has hit a wall, similar to being CPU limited, but my CPU is at 5-10%. So it's not even making use of the hardware available to it. Seems that VR performance in the Odyssey graphics engine is bad no matter how much hardware you throw at it.
I did try it in Horizons 3.8 with everything cranked up, and it looks amazing and is buttery smooth. Oh well.
Note for those who haven't read the entire thread. This is on an HP Reverb G2 playing Odyssey, your mileage may vary on lower resolution headsets and/or Horizons.
Last edited:
it is cpu limited in certain areas
and it doesnt show because it's single thread performance that matters and the polling interval is not small enough to catch all the small blockers.
and since we're about 4090 and dlss3 was mentioned... it's not that pretty as nvidia was trying to imply
a very nice analysis below
Last edited:
Hoping to run some Odyssey tests on a 4090 tonight (although I've the hassle of replacing my power supply and reconnecting everything first.)
I don't have an absolutely top end CPU at the moment (I have a 10700k), so will see how that works out...
Hoping I can at least turn up super sampling a bit and maintain the same or higher frame rate as I'm currently getting.
I don't have an absolutely top end CPU at the moment (I have a 10700k), so will see how that works out...
Hoping I can at least turn up super sampling a bit and maintain the same or higher frame rate as I'm currently getting.
Ah, right, that makes sense, I did not check individual core usage.
So even with an i9-12900KF and an RTX 4090 you can't run Odyssey VR anywhere close to Horizons with much lesser hardware. Oh well, that's as much performance as I can throw at it.
Last edited:
@Morbad wrote in a few threads about CPU bottlenecking, I hope pinging them here will get a reminder which they were.
The game is generally limited by one or two threads and you can easily be CPU limited at only 5-10% total utilization of a part with 20 logical cores.
Also, due to the nature of the loads involved and Windows' propensity to rapidly cycle threads between logical cores, no single core may show sustained peak loads, even if it is the bottleneck, unless tools with sufficiently fast polling rates are used to monitor utilization.
Furthermore, even when there are gaps in actual CPU utilization, it can be due to the game waiting on main memory; EDO is extremely sensitive to cache/memory performance and often sees significant gains from CPUs with larger caches or well tuned system memory, all other things being equal.
In general, unless GPU utlization is pegged at a very high level (upper 90s percentage), you can be fairly certain of a CPU or memory subsystem bottleneck.
Supersampling should be wholly GPU dependent. So, whatever your fps floor is due to other limitations, it likely won't get worse until you start maxing out GPU utilization with more pixels.
ok big bertha GPU incoming tomorrow hopefully, lets see if we can brute force Odyssey into a decent VR experience
NVIDIA didn't think about Frontier Developments once while thinking about the marketing and sales of the 4000 series.
That graphics card costs more than my entire gaming rig, LOL. I guess it's a good thing I like older games and can still enjoy 1080p 60fps (though probably not in Odyssey).
I am glad for all the early adopters, however, as that's what makes today's $1000 card become tomorrow's $100 card. So thank you everyone!
I am glad for all the early adopters, however, as that's what makes today's $1000 card become tomorrow's $100 card. So thank you everyone!
I have some good news regarding the 4090 in VR...
Absolute Game Changer
Absolute Game Changer
What can it do? I'm considering getting one for my 12900 system just to try to run EDO on my index 