Comparison of 5070ti and 5080 in VR

[Qcumber]

I have been lucky enough this week to be able to compare the 5080 FE with the 5070ti (MSI Ventus x3 OC).

Setup

A 9800x3d with a x870 motherboard (Asus), 64Gb DDR5 RAM at 6000Mhz, a 2 Tb Lexar EQ790 NVME drive (game drive). The CPU has a liquid, three-fan cooler.

When comparing the two cards all settings are the same. DCS settings are at the bottom.

The cards were compared at stock clock speeds and with overclocking (MSI afterburner).

	Stock			OC
	Core	Boost	Mem	Core	Boost	Mem
5070ti	2295	2482	1750	2695	2882*	2000
5080	2295	2617	1875	2695	3017*	2125

*Actual clock speeds were close to 3200MHz for both cards during testing.

 

For comparison here are the 3DMark scores:

	5070ti (stock)	5070ti (OC)	5070ti (%)	5080 (stock)	5080 (OC)	5080 (%)
Steel Nomad*	6273	6745	107.5%	8050	9020	112.0%
Fire Strike#	67955	72704	107.0%	78293	83752	107.0%
*Average of 4-6 runs, #graphics score

Quest Pro with Meta link cable at 72Hz and base resolution (1808x1856 per eye).

QVFR

smoothen_focus_view_edges=0.1

sharpen_focus_view=0.9

horizontal_focus_section=0.25

vertical_focus_section=0.25

peripheral_multiplier=0.8

focus_multiplier=2.2

Equivalent resolutions (if applied to the full screen)

	Base	2.2	0.8
Horizontal	1808	3978	1446
Vertical	1856	4083	1485

 

The test run was the instant action mission “Free flight over Syria” in the F-16.

 

I used “XRFrameTools by Fred Emmott to capture single frames over a 1-minute period (4300 frames).

XRFrameTools/README.md at main · fredemmott/XRFrameTools.

Data recorded:

CPU render time

GPU render time

VRAM use (current)

FPS

The results were extracted to a spreadsheet and used to calculate average values (plus min/max) and the number of missed frames (those that exceeded 13.9 ms; (1000/72 = 13.89 ms)). This is shown as the red dotted line in the charts.

The charts represent the number of events occurring in 0.1 ms bins from 0 to 20 ms, displayed as a percentage of the total frames. What this shows is the distribution of frame times and the consistency across the track. Ideally, these should look tall and narrow to represent consistent frames times.

Results

The blue traces show the results for the 5070ti. Base clock (open). Overclocked (filled).

The green traces show the results for the 5080. Base clock (open). Overclocked (filled).

 

GPU render times were 2 to 2.6 ms faster (approx. 28-32%) for the 5080 vs the 5070ti. Overclocking for each card produced a 7-8% reduction in GPU render times, and a more consistent distribution (narrow and taller). The stock 5070ti was pushing the 13.9 ms boundary which could lead to missed frames and stutter.

The CPU render times were essentially the same for each run.

 

	f-16 Syria	f-16 Syria	f-16 Syria	f-16 Syria
	5070ti (base)	5070ti (oc)	5080 (Base)	5080 (OC)
CPU mean	4.12	4.04	4.09	4.06
CPU median	4.09	4.01	4.04	4.01
CPU min	3.37	3.16	3.31	0.24
CPU max	7.48	6.80	6.74	8.93
GPU mean	10.65	9.49	8.05	7.47
GPU median	10.58	9.42	7.97	7.34
GPU min	8.03	7.07	5.85	0.21
GPU max	14.00	13.22	11.21	13.66
VRAM mean	7.11	7.26	6.93	7.60
VRAM median	6.88	7.27	6.97	7.52
VRAM min	6.13	7.12	6.54	7.29
VRAM max	7.83	7.32	7.01	7.85
Events	4229	4229	4214	4214
Missed	1.00	0.00	0.00	0.00
% missed	0.02%	0.00%	0.00%	0.00%
FPS	72.2	72.1	72.1	72.0

 

 

 

 

Edited March 23 by Qcumber

[dock999]

So what is your conclusion?   The 5080 can give you a 28%-uplift in VR DCS in comparison with the 5070Ti.

[Qcumber]

1 hour ago, dock999 said:

So what is your conclusion?   The 5080 can give you a 28%-uplift in VR DCS in comparison with the 5070Ti.

Yes. A 32% increase at stock speeds and a 28% increase when overclocked. Of course the overclock potential will vary between cards. Also, this is a single test. I need to compare the two in a more demanding situation. 

[Qcumber]

Does anyone have a 9800x3d and a 4090 using a Quest Pro? I'm interested in comparing this with the 5070ti and 5080. 

[Qcumber]

I have done a comparison of more demanding track. This is a free flight of BF-109 low over Cairo. I did not use XRFrametools for the 5070ti test so I have had to use OXRTK so not every frame is captured. 

The mean GPU frame time is 9.1ms for the 5080 and 12.7 ms for the 5080. A 39% difference. All settings are as per the tests above. 

[RoboHackfeld]

Well here just my two cents to the 5080 performance compared to my old AMD 6900XT.

 

Short? Its massive 

Long? I havent tested much so far (No DLSS 4 or something) but just out of the Box I jumped at my Benchmark Scene which is A10 II over Center of Aleppo in Syria and my FPS jumped from 50 (GPU Bound) up to 90 which is my max Hz, so there should be still Headroom and I am pretty sure the bottleneck will most likely the 5800x3d and not the GPU.

On my Carrier Test with a lot of static object, like Planes and vehicles, nothing changed so far. When I look to the "Mother" its 55fps (cpu bound) like before, but thats like I said a problem of my cpu. If I am moving to the CATs or just looking up front its stable 90 fps.

 

I am pretty amazed and will come back here when I found some time to set it all up with the DLSS 4 workarround and stuff. 

 

 

Edited April 4 by RoboHackfeld

[Qcumber]

27 minutes ago, RoboHackfeld said:

Well here just my two cents to the 5080 performance compared to my old AMD 6900XT.

 

Short? Its massive 

Long? I havent tested much so far (No DLSS 4 or something) but just out of the Box I jumped at my Benchmark Scene which is A10 II over Center of Aleppo in Syria and my FPS jumped from 50 (GPU Bound) up to 90 which is my max Hz, so there should be still Headroom and I am pretty sure the bottleneck will most likely the 5800x3d and not the GPU.

On my Carrier Test with a lot of static object, like Planes and vehicles, nothing changed so far. When I look to the "Mother" its 55fps (cpu bound) like before, but thats like I said a problem of my cpu. If I am moving to the CATs or just looking up front its stable 90 fps.

 

I am pretty amazed and will come back here when I found some time to set it all up with the DLSS 4 workarround and stuff. 

 

 

Great to hear you are getting good results. Which card have you got (manufacturer) and have you tried overclocking yet?

[RoboHackfeld]

Got the Gigabyte 5080 AERO which first was some kind of a downer.

To explain that: I really wanted to have that card, because every test is positive about that card but what no one tells us is, that there is some kind of magnetical FanStop which makes some noises when the Fans switch to FanStop, not when the computer is shutting down, only when it is active software forced FanStop because Temperature is under 50°C or what ever you set up.

Since I know its not a problem, first I thought it is related to bad or poor bearings, yeah it is really not a bug its a feature and I dont know why the hell they doing something like that, just let the fans run out on their own.....

And it is not only related to this card, if you search for Gigabyte GPU FanStop noise it seems that all Gigabyte Cards have this "Feature" doesnt matter if Aero, Aorus, 5080, 5090, 5070 ti, .... even 10xx Cards from the past have this. 

But yeah it only happens after I quit a game. If I am at the desktop the temperatures never went up so far that the fans start spinning. 

A lot OT but maybe important for some other pilots trying to get a Gigabyte Card. 

Havent tried OC yet, only and thats what I do with every card, I raised the power target. 

 

Edited April 4 by RoboHackfeld

[Qcumber]

1 minute ago, RoboHackfeld said:

Got the Gigabyte 5080 AERO which first was some kind of a downer.

To explain that: I really wanted to have that card, because every test is positive about that card but what no one tells us is, that there is some kind of magnetical FanStop which makes some noises when the Fans switch to FanStop, not when the computer is shutting down, only when it is active software forced FanStop because Temperature is under 50°C or what ever you set up.

Since I know its not a problem, first I thought it is related to bad or poor bearings, yeah it is really not a bug its a feature and I dont know why the hell they doing something like that, just let the fans run out on their own.....

And it is not only related to this card, if you search for Gigabyte GPU FanStop noise it seems that all Gigabyte Cards have this "Feature" doesnt matter if Aero, Aorus, 5080, 5090, 5070 ti, .... even 10xx Cards from the past have this. 

But yeah it only happens after I quit a game. If I am at the desktop the temperatures never went up so far that the fans start spinning. 

 

Havent tried OC yet, only and thats what I do with every card, I raised the power target. 

 

Hopefully that won't affect performance. You should be able to get a decent overclock with that. I can manage +420MHz core clock and +2000MHz memory. That gives me a base if 3320MHz clock boosted and about a 10-11% increase in synthetic benchmarks. 

[RoboHackfeld]

Still not sure if I send that card back to the shop but like I said, it seems to be a Gigabyte Feature which should not affect performance. Gigabytes statement to that was "Its not a problem and It wont be in the future."

Wow, thats alot for the FE. Not bad!

Your doing the OC with Afterburner? 

 

 

Edited April 4 by RoboHackfeld

[Qcumber]

Just now, RoboHackfeld said:

Thanks thats alot for the FE. Not bad!

Your doing the OC with Afterburner? 

 

 

Yes. MSI Afterburner. The review by TechPowerUp managed to get a 12% boost with Time Spy Extreme but they managed to overclock the core to +500 and memory to +3000. 

I can get the core clock stable to +447 with synthetic benchmarks but it's not stable in DCS. It's probably related to DLSS4 and preset K as these are supposed to be temperamental l to overclocking. 

6 minutes ago, RoboHackfeld said:

Still not sure if I send that card back to the shop but like I said, it seems to be a Gigabyte Feature which should not affect performance. Gigabytes statement to that was "Its not a problem and It wont be in the future."

Wow, thats alot for the FE. Not bad!

Your doing the OC with Afterburner? 

 

 

It depends on how much you paid and if you can get a better deal. I was lucky to get the FE at MSRP. I had bought a 5070ti for £799 but then managed to get the 5080 for £949 which is what I really wanted in the first place. So the 5070ti went back. 

[RoboHackfeld]

I will check OC maybe next weekend. Than I have a lot of time to make some tests. 

 

Yeah I got the gigabyte card for  1400€ which is just a bit over the MSRP Gigabyte announced back in January, so its okay for me. Ii I can get another Card like an ASUS Prime for a good price within the next 27 days I think I will replace it but if not I ll keep it. 

 

 

@Qcumber How did you manage to get 3320MHz with only a 447MHZ OC on a 5080 FE? Isnt the the Base clock at 2640MHz?

My Gigabyte runs out of the box with 2850 MHz and 1875MHz memory clock in Steel Nomad which I am pretty fine with. 

Edited April 4 by RoboHackfeld

[Qcumber]

5 hours ago, RoboHackfeld said:

How did you manage to get 3320MHz with only a 447MHZ OC on a 5080 FE? Isnt the the Base clock at 2640MHz?

I'll post some me screenshots of my settings later. 

With GPU-Z the o/c is about 3060. However the actual boost clock speed running on MSI Afterburner is 3320. This drops to about 3150-3220 when running DCS missions. I don't fully understand the reasoning so perhaps others can comment. 

[whitav8]

Please keep testing and tweaking this combination of 9800X3D and RTX 5080 or similar equipment. The concept of running in VR without motion reprojection (ASW,SSW,smoothing,...) would be such a wonderful state for VR. Maybe not for all of us, but we have lived through lots of VR stuttering for the last five or six years and it would be great to finally get over that hurdle. Obviously, it wouldn't be perfect - some temporary extra load of the GPU or terrain data loading,etc - but close to "fluid".

Qcumber - could you please post your histogram code/script for a spreadsheet. I am doing similar testing with XRFrametools and think your histogram plots tell us a lot

Thanks

Dave

[Qcumber]

I have just posted my spreadsheet and workings here. I am very happy to collaborate. Try to find some way of standardising the benchmarking in VR. Any suggestions on developing and streamlining the workflow would be very much appreciated. 

 

[whitav8]

Thanks so much for posting your amazing efforts in this Qcumber. This type of analysis helps us find the best solutions and maybe find some work arounds for code that wasn't actually written with "real-time programming" methods in mind - although I am very pleased that it (DCS) almost works well. When I helped code the Boeing 777 model in the early 1990s, we decided on a 10 millisecond frame time since that was the frame time of the real Fly-By-Wire system. We had an 8 physical CPU system with a shared memory backplane ( functionally similar to our 9800X3D) that cost $350,000 each - and we bought 50 of the systems (test pilot training and systems test) before we certified the 777 in 1995. We did use a background thread for the Flight Management System which controlled route flying.

Please keep us posted on your results. I will get started running similar tests even now with my 3080ti before I get the 5080.

 

[whitav8]

I have had success with Qcumber's analysis plots and find them fine grain detailed especially when comparing between two different variations. I also decided to try CapFrameX as a means to try to measure just basic total frametime and get plots a little easier even though it doesn't have as detailed a result. If you haven't tried it...

Also, I wondered if "real" frametime and therefore "real" FPS is something like 1 / (MAX(CPU_render, GPU_render) + SubmitCPU) > In other words, it seems like we have to add in 1-2 milliseconds in order to get what is really our "worst" numbers that we have to keep less than 13.88 msec for us 72Hz Quest3 folks.

I received my 5080 and will install it later this week - very heavy (support needed??) and somewhat cumbersome with the three 6 pin power cable adapter!

[Qcumber]

 

23 minutes ago, whitav8 said:

I have had success with Qcumber's analysis plots and find them fine grain detailed especially when comparing between two different variations.

Thanks. I am glad it's useful but it still needs some work.

24 minutes ago, whitav8 said:

I also decided to try CapFrameX as a means to try to measure just basic total frametime and get plots a little easier even though it doesn't have as detailed a result. If you haven't tried it...

Not yet but funnily enough I was thinking about this today. 

24 minutes ago, whitav8 said:

Also, I wondered if "real" frametime and therefore "real" FPS is something like 1 / (MAX(CPU_render, GPU_render) + SubmitCPU) > In other words, it seems like we have to add in 1-2 milliseconds in order to get what is really our "worst" numbers that we have to keep less than 13.88 msec for us 72Hz Quest3 folks.

Not sure as such but if I understand correctly XRFrametools shows quite a large variations in FPS for each frame. I am guessing that these are captured based on an amalgamation of frametimes (!?). I should ask Fred Emmott about this. Maybe it would be useful to calculate FPS averages over 10 or so frames. It is some I am working but need to make sure I am doing this correctly.

It might be useful to create some standards to test across different systems. We can't account for variations in PC builds but we can look at relative performance on each build. I was thinking about 3 tracks. A light one, an intermediate one and a heavy one. It's probably best that I post the specific suggestions in my other thread relating to methods. 

26 minutes ago, whitav8 said:

I received my 5080 and will install it later this week - very heavy (support needed??) and somewhat cumbersome with the three 6 pin power cable adapter!

The FE does not come with a support bracket and the official information is that you do not need one. However there is noticeable "sag" so I installed a bracket just in case. The three pin connector is a PITA. I ended up ordering a new power cable as i only had two that would fit the bill. I initially had to daisy chain one of them. All OK now though. 

[Qcumber]

I am getting some amazing performance in  VR at 90 Hz with the 9800x3d and 5080 combination. I'll try to post some performance tests and a video at the weekend. 

[whitav8]

Thanks for the quick response and suggestions, Qcumber. This is quite interesting. Even at 72 or 80,90 - I am interested in how many msec. over the frame budget (13.8, 12.5, 11.1)   and how many consecutive frames that are, for example, 3 msec. different before we "easily" see a stutter. Obviously there will be scenery load stutters that we have to live with but just what constitutes an objectionable stutter?

[JesterIsDead]

how do we get our hands on one of these- sold out everywhere

[Qcumber]

A 5080? It's depends where you live. In the UK availability is not too bad if you keep a close eye on local stocks. It might be worth using a price check bot.  

[Rainer_B.]

Got lucky. Ordered a MSI RTX5080 Shadow 3x at Amazon for €1169. Got it yesterday. Order date was the 5th with unknown delivery Date. Will replace my 3080 at Friday.

[Qcumber]

The new NVIDIA driver apparently gives a significant performance boost to the 5080. I have not yet tried it. 

https://wccftech.com/nvidia-gpu-driver-576-02-results-in-higher-rtx-5080-performance/

 

[Qcumber]

I am getting an average of +5% improvement in Steel Nomad. I'm not sure if that will translate to DCS. With my overclock of +10% that puts this at +15% above the previous driver baseline. I need to test if it can cope with a higher overclock now.