RTX 4080 & 7900 XTX/XT benchmarks DCS

[xoxen]

Actually the 7900 XT is available in Germany for EUR 1.129,- (XFX and Powercolor at Mindfactory). I did not see the XTX listed anywhere.

 

Edit: Just found the XTX on ebay for EUR 1.399,-. So the scalpers are doing their job again. Shame on them.

Edited December 14, 2022 by xoxen

[ssg]

I got XTX for 1200 euro from AMD directly. Cheapest available 4080 is 1500 euro and cheapest 4090 is 2300 euro here.

If I were flatscreen gamer I think XTX would be obvious choice with these prices. VR however makes things bit different.

According to tests from the site that was posted above XTX loses to even 3080 in VR. I don't know how that translates to DCS. I will need to figure out some fairly consistent tests to test it myself. The issue could be driver issue, if I remember correctly 6000 series also has bad VR performance on launch but it improved later. You can't really bet on that however.

If XTX fixes either of 2 issues I currently have, screen tearing with ASW on or microstuttering when looking to sides with ASW off then I will probably end up keeping it. 

Edited December 14, 2022 by ssg

[coldViPer]

11 minutes ago, ssg said:

I got XTX for 1200 euro from AMD directly. Cheapest available 4080 is 1500 euro and cheapest 4090 is 2300 euro here.

If I were flatscreen gamer I think XTX would be obvious choice with these prices. VR however makes things bit different.

According to tests from the site that was posted above XTX loses to even 3080 in VR. I don't know how that translates to DCS. I will need to figure out some fairly consistent tests to test it myself. The issue could be driver issue, if I remember correctly 6000 series also has bad VR performance on launch but it improved later. You can't really bet on that however.

If XTX fixes either of 2 issues I currently have, screen tearing with AWS on or microstuttering when looking to sides with AWS off then I will probably end up keeping it. 

 

Can't wait your results, I have the same 3080 10Gb right now. What headset you have?

[Johnny Dioxin]

Well, it's a difficult decision. I just saw some 3080's available for £700 - that's a big saving over a 4000 series - but I don't know if it would be worth it, thinking of how much more VR performance I'd get from a 4080, and how much of a difference the 3080 will make over my 2080Ti.

It's likely to be my main card for at least 3 years, so I think maybe the extra layout for the 4080 is worth it.

Decisions, decisions...  a few months ago I was dead set on the 4080, now doubts start to creep in, if it's the right path.

[Andrei]

3 hours ago, ssg said:

According to tests from the site that was posted above XTX loses to even 3080 in VR. I don't know how that translates to DCS. I will need to figure out some fairly consistent tests to test it myself.

Please do share your findings. For me, the biggest question between 7900 XTX and 4080 is now how it's going to perform in DCS / MSFS VR.

While we might get some 'big boys' to test in MSFS, for DCS performance it's impossible to rely on anymore, but fellow community members

[Obic]

I am curious too 
I don't trust the youtuber one second (All they do is advertising) for who ever is pushing more.

I own a 3090 FTW3 Ultra.
It's probably one of the worst card, I've ever own as far as the layout on the PCB goes.
Precision X1 is a Abortion for OC compare to AMD onboard solution.
I have yet to turn one DLSS or RT so I could not care less about this non-sense.
Nvidia driver are far from that great 

The price after taxes is what makes the 7900xtx interesting to me in Canada.
(15% Taxes)

4090 Assuming you can find a Reference are 2550$ - AIB 2600 to 3300$ (4000$ if you like scalper)
4080 Assuming you can find a Reference are 1950$ - AIB 2100 to 2400$
7900XtX Assuming you can find a Reference are 1550$ - Most expensive AIB 1950$
 

[Strong05]

Tried the 7900xt last night and was very disappointed.  Frames were about the same or worse then my 6800xt.  BAR had a noticeable negative impact on frames.  Overall about 30-40% less fps then the 4080 I returned.

6800xt=50fps sitting on the ramp in F5 on Caucus however frames seemed to drop more during flight then the 7900 (relatively speaking, frames didn’t seem to drop in flight as much with the 7900 but they were still lower overall in flight as well with the 7900xt) to about 40-45.

7900xt = 42fps dipped into the high 30’s in flight  saw high gpu utilization throughout (high 90's with 99% most of the time I was looking at it).

4080 = 60fps  rock solid 60 or more in flight, gpu utilization was often below 90%, down to as low as 75%

 

I’m currently running a 5800x3d with 32 go ram at 3400 with g2 in open xr.  Cleaned up drivers with ddu after every gpu swap.  Also probably worth mentioning that my 6800xt is overclocked/undervolted.  I did not change anything for the 7900xt except BAR.

On a sub note, the low gpu utilization of the 4080 is weird because if I turned on FRS with my 6800xt in VR and ran a low resolution I could easily get 90fps, which seems to indicate on this map (same for all comparisons) that I am not CPU limited.

 

Edited December 14, 2022 by Strong05

[Andrei]

Ouch.. Wonder if the next release of the drivers will do it any good...

[EightyDuce]

Interesting overclocking results of the 7900XTX TUF. 

https://www.techpowerup.com/review/asus-radeon-rx-7900-xtx-tuf-oc/39.html

 

If this isn't a "golden sample" card could have some potential for those on water. 

Edited December 14, 2022 by EightyDuce

[CybrSlydr]

Do you folks think my 8700k is hampering my 6800XT in VR?  

Would getting something like a 13700k, 32GB DDR5 help out rather than getting a new gen GPU?

Edited December 14, 2022 by CybrSlydr

[Andrei]

13 minutes ago, CybrSlydr said:

Do you folks think my 8700k is hampering my 6800XT in VR?  

Would getting something like a 13700k, 32GB DDR5 help out rather than getting a new gen GPU?

 

With a high likelihood it does. 

Moving from 7700K to 5800X has raised FPS from 65 -> 80 in 2D scenario I've been benchmarking.
And from 5800X to 5800 X3D has improved CPU frame time from ~14.8 to 12.6 (- 15%). 

That is on the same graphics card 2070S. 

But the best thing for you would be to get an FPSVR for Stream VR or OpenXR Toolkit if you use Open XR and see where your frametime bottleneck is.

[CybrSlydr]

I do have FPSVR, never really used it.  I don't currently have OpenXR.

[CybrSlydr]

26 minutes ago, Andrei said:

With a high likelihood it does. 

Moving from 7700K to 5800X has raised FPS from 65 -> 80 in 2D scenario I've been benchmarking.
And from 5800X to 5800 X3D has improved CPU frame time from ~14.8 to 12.6 (- 15%). 

That is on the same graphics card 2070S. 

But the best thing for you would be to get an FPSVR for Stream VR or OpenXR Toolkit if you use Open XR and see where your frametime bottleneck is.

Here's a screenshot from a Liberation campaign in Syria.

CPU frame time is in the high 40ms-low50ms most of the time.  Reprojection ratio is in the mid 50%.  It's also chewing up and using all 32GB of my system RAM and 15.6 of the 16GB of my VRAM.

EDIT:  Sorry, wrong screenshot.

null

Edited December 14, 2022 by CybrSlydr

[Andrei]

Well, there you go. Your CPU is a limiting factor by a rather big margin. At least in CPU-heavy scenario, which DCS:Liberation is.

Now I guess we need to come back on topic

[Caysuyu]

2 hours ago, Strong05 said:

Tried the 7900xt last night and was very disappointed.  Frames were about the same or worse then my 6800xt.  BAR had a noticeable negative impact on frames.  Overall about 30-40% less fps then the 4080 I returned.

6800xt=50fps sitting on the ramp in F5 on Caucus however frames seemed to drop more during flight then the 7900 (relatively speaking, frames didn’t seem to drop in flight as much with the 7900 but they were still lower overall in flight as well with the 7900xt) to about 40-45.

7900xt = 42fps dipped into the high 30’s in flight  saw high gpu utilization throughout (high 90's with 99% most of the time I was looking at it).

4080 = 60fps  rock solid 60 or more in flight, gpu utilization was often below 90%, down to as low as 75%

 

I’m currently running a 5800x3d with 32 go ram at 3400 with g2 in open xr.  Cleaned up drivers with ddu after every gpu swap.  Also probably worth mentioning that my 6800xt is overclocked/undervolted.  I did not change anything for the 7900xt except BAR.

On a sub note, the low gpu utilization of the 4080 is weird because if I turned on FRS with my 6800xt in VR and ran a low resolution I could easily get 90fps, which seems to indicate on this map (same for all comparisons) that I am not CPU limited.

 

 

Thanks for the initial report. Having read your post, I was planning to wait but I pulled the trigger on a 7900 xtx when I could score one on amd.com. It will still be a significant improvement over my GTX 1080 and as the resolution of my headset (Samsung Odyssey +) is relatively low,  I expect to see some improvement. I hope AMD improves VR perf through driver updates soon. 

[Voyager]

@Johnny Dioxin Try to wait for January. There are rumors that nVidia may drop the price on the 4080, especially if 7900XT sales undercut it. The 7900XTX is a competitive card at its price point, even if it is under performing where we all expected it to. And the 4080 is very poorly positioned. 

On the XTX supply, MLID is saying that AMD is mostly shipping XT models, not XTX models, but there should be about 2-3x cards on shelves by the end of the year as the Lovelace cards. 

My suspicion is AMD is off-loading the XT dies early while supply is tight and they haven't solved what's going wrong with their cards.

[SkateZilla]

I'm waiting for drivers, MT and Vulkan before I put it in my main DCS System.
AMD bailed on DX11 long ago, so even though they did a DX11 Perf. Patch for the rDNA1/2 GPU Drivers, I doubt it applies to rDNA3,

I have been working with a local streamer that does OCing, and we were able to get his reference AMD 7900XTX to within 5-8% of a RTX4090 in non RT Synthetic benchmarks.

he asked for my help, as I was overclocking GPUs back when we had the 8800GT/GTS/GTX and they had a similar Core/Shader Clock setup, 
And we all know, you can jump shader clock pretty easily on those GPUs for a nice performance boost.

1st stage: OCing out of the box, we were able to get 10-15% higher shader clocks, with fan curves and no voltage changes.
2nd stage: additional 5% Core clocks with slight mV increase, however hot spot temps on GPU were concerning

3rd stage: removed heatsink, replaced thermal goo, with some better TIM, also removed the 3 MM thick Pads they had on the ram, and replaced with copper shims and better TIM. the RAM Pads were horrible, also, there was NO TIM on the rear VRM Array, the forward array had the same thick pads as the VRAM modules. (replaced it, and added TIM and copper shim to the rear array, also, the VRM only used a plate to pull heat, the plate was directly cooled by fans, and not attached to the vapor chamber, sinks or pipes.)

4th stage: increased mv, hotspot on vrm was cooler as result of cleaning up stock heatsink. at this point we were also getting a consistent 12% higher boost core clock at stock voltage, due to lower GPU temps (again, they had an insane amount of thermal goo on the Chiplet package, and those super thick pads on the VRAM modules.)

We got halfway into to that point before we realized we had to replace the cooling solution. as the 7900XTX with this cooler was only designed to dissipate 375w, and we were pushing 400w about -20% of the Rasterization of a RTX4090 (Stock/NonOC)

We replaced the air cooler with a AIO 360 water unit, and attached larger heatsinks to the VRM and VRAM arrays, and had to mount fans blowing on them.

5th stage, at this point we were able to increase power to the GPU some more and the boost clocks well over 3GHz on the shader clocks and 2.9~ on the core clocks, once we got that stable it was ~18% behind the RTX4090.

6th Stage, we finished this yesterday, we replaced the AIO 360 with a custom loop, GPU was water blocked, and we utilized 3 custom ram water blocks attached to the VRAM Array, (Modified MOSFET water blocks), and then we added MOSFet Blocks to the VRM Array as well. this entire loop was external, as it was simply built for testing. Loop was Reservoir, Pump, MOSFET/VRAM Blocks/ 240 RAD / GPU / 240 RAD / Reservoir. The Reservoir was also cooled (in a small desktop minifridge modified w/ tub ports.)

7th Stage, added more voltage and was able to push core clocks to 3.2 and shader clock to 3.4 ghz At this point we are 10~% behind the RTX4090.

8th stage, adjusted core and shader clocks to allow more rasterization while not req. more power, lowered shader to 3.25, increased core to 3.35 and gained 2% (-8% of RTX4090).

At this point the 4090 would be pulling 520~watts,
the RX7900XTX was pulling 425~w, 

Any further power increases caused stability problems, likely due to the limit of only 2x 8Pins, we were pulling 10% over the PCIe Connectors and PCIe Slot specs as well.

So, the performance is there, some AIB Partners are showing their 4090 sized coolers on their RX7900XTXs w/ an additional 8 Pin Power connector giving them up to 525w ceiling.

Younger system builders, if they weren't around during the nVidia G80/G90 days, dont understand how to manipulate the de-coupled frequencies.


Also Video w/ results is coming, it's being edited from 49 hours of footage.

we also told our findings to other reviewers, so 

Edited December 14, 2022 by SkateZilla

[Voyager]

@SkateZilla Did you two try it in any games or just synthetic benchmarks? 

From what I'm seeing in reviews, it seems to do as expected in the synthetic benchmarks, but the issues seem to crop up during live games instead. 

[SkateZilla]

Also, if your using Ryzen 3000+, Make sure you have Smart Access Memory Enabled, without it, your looking at 20-30% performance drop,

If you run Intel, you'll looking at the same drop.

Resizable BAR Enabled, Above 4G Decoding Enabled

Edited December 14, 2022 by SkateZilla

[Th3ChosenOn3]

I'll most likely be receiving my XFX RX 7900 XTX tomorrow and plan to try DCS. Don't really have comparable numbers to current gen offerings, as I'm currently GPU bound with GTX 1080 8GB.

I'm playing on a monitor (3440x1440), so won't be able to give VR numbers. Are there specific missions/tracks that are used for benchmarks?

[Andrei]

33 minutes ago, Th3ChosenOn3 said:

I'm playing on a monitor (3440x1440), so won't be able to give VR numbers. Are there specific missions/tracks that are used for benchmarks?

Try this one: https://cdn.discordapp.com/attachments/915347310567915541/1036740701196664875/plazma_torture_map_2022_flanker.miz

It was used in at least one Youtube video which was benchmarking 4090 and I've been using it as well for CPU benchmarks. It's pretty well done - short, but packed with action. Requires FC3 or Flanker.

P.S. Link to original benchmark by map's author.

Edited December 14, 2022 by Andrei

[jib]

5 hours ago, Strong05 said:

Tried the 7900xt last night and was very disappointed.  Frames were about the same or worse then my 6800xt.  BAR had a noticeable negative impact on frames.  Overall about 30-40% less fps then the 4080 I returned.

6800xt=50fps sitting on the ramp in F5 on Caucus however frames seemed to drop more during flight then the 7900 (relatively speaking, frames didn’t seem to drop in flight as much with the 7900 but they were still lower overall in flight as well with the 7900xt) to about 40-45.

7900xt = 42fps dipped into the high 30’s in flight  saw high gpu utilization throughout (high 90's with 99% most of the time I was looking at it).

4080 = 60fps  rock solid 60 or more in flight, gpu utilization was often below 90%, down to as low as 75%

 

I’m currently running a 5800x3d with 32 go ram at 3400 with g2 in open xr.  Cleaned up drivers with ddu after every gpu swap.  Also probably worth mentioning that my 6800xt is overclocked/undervolted.  I did not change anything for the 7900xt except BAR.

On a sub note, the low gpu utilization of the 4080 is weird because if I turned on FRS with my 6800xt in VR and ran a low resolution I could easily get 90fps, which seems to indicate on this map (same for all comparisons) that I am not CPU limited.

 

 

That is really disappointing news to hear, but thanks for sharing.
I think I will wait and see if Vulcan / AMD driver updates improve the situation in the next 6 months.
I cannot justify the price of a RTX 4090 as most are going for £2,000+ in the UK that's a months wages

[Th3ChosenOn3]

31 minutes ago, Andrei said:

Try this one: https://cdn.discordapp.com/attachments/915347310567915541/1036740701196664875/plazma_torture_map_2022_flanker.miz

It was used in at least one Youtube video which was benchmarking 4090 and I've been using it as well for CPU benchmarks. It's pretty well done - short, but packed with action. Requires FC3 or Flanker.

P.S. Link to original benchmark by map's author.

 

Thank you, will give it a go when I get my card, but might be bit CPU bound with a 5800X.

Edited December 14, 2022 by Th3ChosenOn3

[SkateZilla]

ASUS' TUF Card gets to the same marks as the OC'd we did,
7900XTX TUF runs 3 8 Pins, and is on par w/ 4090 NonRT

[3WA]

1 hour ago, jib said:

I cannot justify the price of a RTX 4090 as most are going for £2,000+ in the UK that's a months wages

I know right?  The GPU prices are getting INSANE.