New rig - RTX5090 or AMD 9070?

[Boberro]

Hello, 

I am going back to DCS after like 13 years of break  I want to build a new PC, this is my initial setup:

RTX 5090/AMD 9070XT

64 GB RAM

AMD 7950X3D CPU

SSD of course

2K or 4K 27 inch screen + additional one for MFD panels - touchscreen.

Money is not an issue I think, I want to have a setup for longer time - preferably for years as I don't like to buy it often.

Does 5090 offer significant improvement in terms of performance over AMD 9070XT? RTX has 32 GB VRAM, AMD only 16 GB, so I wonder if it would be enough for extreme settings, reshade, mods etc?

Is it better to buy 9070XT and in two years newer RTX 6xxx? 5090 is not a milestone in nVidia family, it is just a gap-filler to be honest - the strongest on the market but no breakthrough.

[LucShep]

I don't agree with everything that Techspot (Hardware Unboxed website) posts or opiniates, but I tend to agree with their recent analysis of the current GPU market:

https://www.techspot.com/bestof/gpu-25-26/

I'd say that, regardless of "money no object", it's also important to have "notion of priorities and needs" and "common sense".

And in that matter (of common sense) the Nvidia RTX 5090 32GB is everything but - it's a nonsense halo product for a select few (like buying a Ferrari if you're a car enthusiast), with a horrible price tag even if considering the high performance and big VRAM.
And, no less important, the possible reliability issues with the very controversial 12V-2x6 power cable/plug.

The AMD RX 9070XT 16GB is currently the "best purchase" overall, considering the current market.
It's a good choice if gaming on a 1440P or 4K resolution screen.
The downside with it - or any AMD GPU for that matter - is that if you're considering going VR later it won't be the best choice.

The Nvidia RTX 5070Ti 16GB and RTX 5080 16GB are better choices for DCS and especially for VR (but at higher prices), so could be better purchases in the long run.

Regarding the VRAM - a GPU with 16GB VRAM is enough for DCS in general but, depending on module+map combo (if in MP) and if in VR or 4K screen at high/max settings, that limit can indeed be hit.

Unfortunately, the Nvidia RTX 4090 24GB is no longer produced and sold. That one was, pretty much, the "perfect" DCS high-end GPU - ironically so, as it was once so criticized but later became "the last great Nvidia release".

I'd say to not wait for the Nvidia RTX 60 series, or plan anything around them, as they won't be out before 2017 - and how good or bad they'll be it's all unknown.

Likewise, and for those waiting for the rumoured Nvidia RTX 50 SUPER series (refresh of the current line up) with higher VRAM, also notice that these are now in the back burner - postponed, if not canceled - due to the ongoing memory price hike.

Treat yourself with a GPU right now, as prices are expected to increase and stock to decrease (again) in coming weeks for an unknown period.

BTW, if that system is aimed for gaming, and not productivity, for the CPU I'd seriously consider the AMD 9800X3D instead of the AMD 7950X3D.
 

Edited Friday at 02:20 AM by LucShep

[SharpeXB]

13 hours ago, Boberro said:

2K or 4K 27 inch screen

If money is no object I’d get a bigger screen. And a 5090 is overkill for 2K

[Dragon1-1]

Note that a "breakthrough" in component performance is going to be harder and harder to come by. The whole reason for NVIDIA making such big deal of frame generation and DLSS is that chip density is hitting physical limits imposed by quantum tunneling effects. So the next generation might involve an even more incremental jump, mostly achieved by drawing more power.

Either way, you don't need a 5090 unless you're going for VR or 4K+ screen. As a GPU it's going to be pretty future-proof, I imagine.

[diego999]

16 hours ago, LucShep said:

I don't agree with everything that Techspot (Hardware Unboxed website) posts or opiniates, but I tend to agree with their recent analysis, regarding the GPU market as of right now:

https://www.techspot.com/bestof/gpu-25-26/

I'd say that, regardless of "money no object", it's also important to have "notion of priorities and needs" and "common sense".

And in that matter (of common sense) the Nvidia RTX 5090 32GB is everything but - it's a nonsense halo product for a select few (like buying a Ferrari if you're a car enthusiast), with a horrible price tag even if considering the high performance and big VRAM.
And, no less important, the possible reliability issues with the very controversial 12V-2x6 power cable/plug.

The AMD RX 9070XT 16GB is currently the "best purchase" overall, considering the current market.
It's a good choice if gaming on a 1440P or 4K resolution screen.
The downside with it - or any AMD GPU for that matter - is that if you're considering going VR later it won't be the best choice.

The Nvidia RTX 5070Ti 16GB and RTX 5080 16GB are better choices for DCS and especially for VR (but at higher prices), so could be better purchases in the long run.

Regarding the VRAM - a GPU with 16GB VRAM is enough for DCS in general but, depending on module+map combo (if in MP) and if in VR or 4K screen at high/max settings, that limit can indeed be hit.

Unfortunately, the Nvidia RTX 4090 24GB is no longer produced and sold. That one is, pretty much, the "perfect" DCS high-end GPU - ironically so, as it was once so criticized but later became "the last great Nvidia release".

I'd say to not wait for the Nvidia RTX 60 series, as they won't be out before 2017.

Likewise, and for those waiting for the rumoured Nvidia RTX 50 SUPER series (refresh of the current line up) with higher VRAM, also notice that these are now in the back burner - postponed, if not canceled - due to the ongoing memory price hike.

Treat yourself with a GPU right now, as prices are expected to increase and stock to decrease (again) in coming weeks for an unknown period.

BTW, if that system is aimed for gaming, and not productivity, for the CPU I'd seriously consider the AMD 9800X3D instead of the AMD 7950X3D.
 

 

These are good suggestions.

[Boberro]

22 hours ago, LucShep said:

And in that matter (of common sense) the Nvidia RTX 5090 32GB is everything but - it's a nonsense halo product for a select few (like buying a Ferrari if you're a car enthusiast), with a horrible price tag even if considering the high performance and big VRAM.
And, no less important, the possible reliability issues with the very controversial 12V-2x6 power cable/plug

Well they may enforce such a price because they literally have no competitors. When you think of the top performance it is simply nVidia. Every corp will be greedy in such a situation.

Regarding the power plug - it is funny as I read people tend to undervolt the GPU so they lose like 10% of the frames and 5090's power usage drops from 575W to 380W-450W which is much less demanding for the power cord and socket. In any case it is just fail but fortunately the occurrence of melted socket is less than in case of 4090 and undervolted 5090 is still much better than AMD here by the 4x more money of course.

Quote

The AMD RX 9070XT 16GB is currently the "best purchase" overall, considering the current market.
It's a good choice if gaming on a 1440P or 4K resolution screen.
The downside with it - or any AMD GPU for that matter - is that if you're considering going VR later it won't be the best choice.

Yes I think I will use a VR set but it is not 100% certain yet.

Quote

Treat yourself with a GPU right now, as prices are expected to increase and stock to decrease (again) in coming weeks for an unknown period.

How so? Isn't time after Christmas and January good time for such purchases?

Quote

BTW, if that system is aimed for gaming, and not productivity, for the CPU I'd seriously consider the AMD 9800X3D instead of the AMD 7950X3D.

I read it is related to core parking, but it should be solved? Yes, the set will be used mainly for gaming, sometimes for Photoshop or Excel.

 

12 hours ago, SharpeXB said:

If money is no object I’d get a bigger screen. And a 5090 is overkill for 2K

You mean bigger in terms of resolution? I consider 2K and 4K, but I guess 27-32 inches are mainly in 4K now.

 

9 hours ago, Dragon1-1 said:

Note that a "breakthrough" in component performance is going to be harder and harder to come by. The whole reason for NVIDIA making such big deal of frame generation and DLSS is that chip density is hitting physical limits imposed by quantum tunneling effects. So the next generation might involve an even more incremental jump, mostly achieved by drawing more power.

Either way, you don't need a 5090 unless you're going for VR or 4K+ screen. As a GPU it's going to be pretty future-proof, I imagine.

Raw progress from 4090 to 5090 wasn't impressive to be honest, but indeed DLSS and other stuff saved this generation. I had hoped also for less power usage, 575W default is a lot.

[Lange_666]

54 minutes ago, Boberro said:

You mean bigger in terms of resolution? I consider 2K and 4K, but I guess 27-32 inches are mainly in 4K now.

Personally (<--) i would not go 4K on a less than 32" monitor. Single pixels are way too small on a 27" and when visually searching for the enemy when no labels are used...  

[LucShep]

20 hours ago, Boberro said:

Well they may enforce such a price because they literally have no competitors. When you think of the top performance it is simply nVidia. Every corp will be greedy in such a situation.Regarding the power plug - it is funny as I read people tend to undervolt the GPU so they lose like 10% of the frames and 5090's power usage drops from 575W to 380W-450W which is much less demanding for the power cord and socket. In any case it is just fail but fortunately the occurrence of melted socket is less than in case of 4090 and undervolted 5090 is still much better than AMD here by the 4x more money of course.
 

The 12V-2x6 power cable/plug is still problematic on the RTX 5090 (and some RTX 5080 models), even if supposedly safer than the previous 12VHPWR on the RTX4090.

Of course, you can (and should!) undervolt it, but some facts seem to point that the 12V-2x6 cable/plug is (still) unable to provide more than 375W reliably at all times.
And since all the 5090 models (and some 5080 models) draw more than that, and use only one cable/plug...
You may power-limit it to 375W (so it never ever goes over that) and be "theoretically safe"... but I'm unsure who'd spend so much on a RTX5090 and power limit it so drastically.

Here, have a (long) read... draw your own conclusions:  

20 hours ago, Boberro said:

I read it is related to core parking, but it should be solved? Yes, the set will be used mainly for gaming, sometimes for Photoshop or Excel.
 

In the past two generations of X3D desktop processors, namely the 7800X3D and the 5800X3D, the 3D V-Cache die was stacked on top of the CPU complex die (CCD), which then required lower CPU clocks than their non-X3D CPU brethren, due to thermal constraints. 

With the 9800X3D, AMD has inverted the CCD+L3D stack (the 3D V-Cache die). The CCD is now on top, and the L3D is below it.
What this does is make the CCD's thermals behave like they do on the regular Ryzen 9000 series processors without 3D V-Cache (much improved now), which is how AMD was able to increase the base frequency significantly, therefore the performance. And it's considered to have the same overclocking capabilities as the regular 9000-series processors (not the case before).

If you need that higher core-count for work/productivity stuff, then go instead to the 9950X3D, not the older 7950X3D.
Otherwise, for gaming, there is no performance difference between 9950X3D vs 9800X3D, as they both will run games on 8 cores / 16 threads (on the CCD with X3D).

And yes, the 7950X3D and 9950X3D have two CCDs (one with 3D-V Cache, one without) so require affinity / parking set-up for the game executable, set to be run on the CCD with 3D-V Cache (9800X3D is single CCD, so it's good to go in that aspect, no FAFO).
 

Edited yesterday at 07:04 AM by LucShep

[SharpeXB]

5 hours ago, Boberro said:

Well they may enforce such a price because they literally have no competitors.

Well they do. AMD. And those cards are just as expensive I don’t think AMD has a card in the league of the 5090 though but that’s not a mainstream product. 

5 hours ago, Boberro said:

You mean bigger in terms of resolution? I consider 2K and 4K, but I guess 27-32 inches are mainly in 4K now.

No, just size really. If you’re spending that much on a PC why go small on the display? 

4 hours ago, Lange_666 said:

Personally (<--) i would not go 4K on a less than 32" monitor. Single pixels are way too small on a 27" and when visually searching for the enemy when no labels are used...  

4K is definitely perceivable on a 27” screen but again size just has a quality of its own. And if you’re older then it becomes rather vital.

[Dragon1-1]

11 hours ago, Boberro said:

Regarding the power plug - it is funny as I read people tend to undervolt the GPU so they lose like 10% of the frames and 5090's power usage drops from 575W to 380W-450W which is much less demanding for the power cord and socket.

In most cases, undervolting does not mean loss of framerate. You're not underclocking it, you're pushing it to get the same clock speed while drawing less power. The factory tuning on those GPUs assumes a generous stability margin, unless you really lost the silicon lottery, you can shave this margin down quite a bit and still maintain the same clock.

That said, even undervolted, the power draw is quite significant (and plug-melting). 

[diego999]

15 hours ago, Boberro said:

 

Regarding the power plug - it is funny as I read people tend to undervolt the GPU so they lose like 10% of the frames and 5090's power usage drops from 575W to 380W-450W which is much less demanding for the power cord and socket. .

 

That is undervolting done wrong. The idea is to slightly limit the power so it runs cooler without a noticeable effect on performance. 

I don't own one of those monster cards, but my 4070ti Super goes 8-10C° and around 50W below with undervolting. And I can't tell the difference unless I look at a benchmark.

There is absolutely no point in purchasing a 5090 only to take out ten percent of its performance.

 

[Boberro]

On 11/28/2025 at 1:27 PM, LucShep said:

You may power-limit it to 375W (so it never ever goes over that) and be "theoretically safe"... but I'm unsure who'd spend so much on a RTX5090 and power limit it so drastically.

I made some calculations with ChatGPT 5090 vs 9070XT when I would cap 5090 to 375W. GPT says that it would be still a bit faster but then it would be just madness to pay 4x or even 5x more (in case of Astral) just for 10% more FPS xD.

 

Quote

Here, have a (long) read... draw your own conclusions:  

I have read both topics, and the video. I am simply stunned to be honest. It is a pure fire hazard. Still, nobody sued nVidia, even the European Union remains silent.

 

Quote

If you need that higher core-count for work/productivity stuff, then go instead to the 9950X3D, not the older 7950X3D.
Otherwise, for gaming, there is no performance difference between 9950X3D vs 9800X3D, as they both will run games on 8 cores / 16 threads (on the CCD with X3D).

And yes, the 7950X3D and 9950X3D have two CCDs (one with 3D-V Cache, one without) so require affinity / parking set-up for the game executable, set to be run on the CCD with 3D-V Cache (9800X3D is single CCD, so it's good to go in that aspect, no FAFO).

I digged for more info and I see both 9800X3D and 9950X3D do the same job in games (so I presume core parking is done automatically now, or in process lasso manually), however there is even a significant difference in some apps. Unsure how this would relate to normal daily usage like ie. installing apps, unpacking some zips or streaming the same time playing.
Price difference is like 250 EUR between the two.
Tl;dr - 9950X3D in games it is more or less the same, and it may require manual core parking in case automatic fails. But for 250 EUR more offers also some performance boost in other jobs.
Will have to think about this as 250 EUR is still acceptable to me in the long run.
 

On 11/28/2025 at 4:46 PM, SharpeXB said:

 No, just size really. If you’re spending that much on a PC why go small on the display? 

I will look at the 32 inch to see how it goes then ^^

OK guys, I consider now this setup as alternative: 9950X3D (also 9800X3D), and 9070XT. What makes me sad is the VRAM amount on the GPU but there is literally no choice.
The only less bugged 5090 is from Asus, which added some monitoring stuff to detect the power unbalance and such cards are even more expensive than normal 5090s. This is pointless and just madness.
In this case 9950/9800 + 9070XT + 64 GB RAM should be fine to run DCS with mods at stable 60 FPS?
 

Edited yesterday at 05:39 PM by Boberro

[niru27]

6 minutes ago, Boberro said:

In this case 9950/9800 + 9070XT + 64 GB RAM should be fine to run DCS with mods at stable 60 FPS?

No.

I have a 9950x3D + 7900XT 20GB + 64GB RAM, and if there are unoptimized aircrafts like Chinook, Apache, etc, my fps will still drop below 60 @ 1440p. And I don't mean you are flying them, even if someone else is flying them near you, you will get affected. If you can afford 24/32 GB VRAM GPUs, go for it.

I avoided NVidia specifically due to the dodgy connector and the extra 4GB VRAM that AMD offered. For your use case, a 24GB 7900XTX would be ideal.

[Lange_666]

1 hour ago, Boberro said:

But for 250 EUR more offers also some performance boost in other jobs.

Depending on what those "jobs" are, but in general you won't notice if your browser opens 0.01 second slower. I would keep that 250€ in my pocket and use it for something else.

[Boberro]

18 hours ago, niru27 said:

No.

I have a 9950x3D + 7900XT 20GB + 64GB RAM, and if there are unoptimized aircrafts like Chinook, Apache, etc, my fps will still drop below 60 @ 1440p. And I don't mean you are flying them, even if someone else is flying them near you, you will get affected. If you can afford 24/32 GB VRAM GPUs, go for it.

I avoided NVidia specifically due to the dodgy connector and the extra 4GB VRAM that AMD offered. For your use case, a 24GB 7900XTX would be ideal.

It is sad then. The more I read about the performance of DCS the more it recalls my memories from 2010 era when I was still quite active in DCS. At that time DCS (Lock on) also struggled with performance. I remember ie water under the ground etc stuff which limited fps.

I can afford 5090, but I decided it is too risky and just spit in face from Jansen - I expected stuff fixed and less or even no burden anymore. I looked at 7900xtx, and it has good amount of VRAM - 24GB, but its performance is noticeably worse than 9070xt. Also, FSR is old and for what's worse, 9070XTX won't arrive either. Even 5080 is no alternative because it is also 16 GB.
I watched some DCS video and I see that after 15 years F2 bug which eats RAM to the atrocious levels is still present XD. Seems my 64GB of RAM plan must change to 128 GB which is insane.

[SharpeXB]

1 hour ago, Boberro said:

It is sad then. The more I read about the performance of DCS the more it recalls my memories from 2010 era when I was still quite active in DCS. At that time DCS (Lock on) also struggled with performance. I remember ie water under the ground etc stuff which limited fps.

I can afford 5090, but I decided it is too risky and just spit in face from Jansen - I expected stuff fixed and less or even no burden anymore. I looked at 7900xtx, and it has good amount of VRAM - 24GB, but its performance is noticeably worse than 9070xt. Also, FSR is old and for what's worse, 9070XTX won't arrive either. Even 5080 is no alternative because it is also 16 GB.
I watched some DCS video and I see that after 15 years F2 bug which eats RAM to the atrocious levels is still present XD. Seems my 64GB of RAM plan must change to 128 GB which is insane.

Well don’t talk yourself out of DCS altogether All these problems, while still terrible (ask me how I know about the 12VHPWR connector) are still rare and can be mitigated by things like making sure it’s seated fully and avoiding lower quality 3rd party products. Most of the performance “problems” with the game have to do with user-specific settings, not the game itself.

Edited 3 hours ago by SharpeXB

[Dragon1-1]

It is rather poorly optimized by modern standards, though. Which is why ED is moving to Vulkan, which, if done right, should wipe some technical debt they built up and allow better optimization for modern GPUs. VRAM consumption in particular really needs to be looked at.