Any tips from AMD 7900XT/XTX users?

[Grodin]

Sorry if i'm on the wrong forum, had trouble finding more specific hardware thread.

 

So i have 3080ti & Reverb G2 on one rig, this works well, it's just lacking the power for stable 90 fps in busy scenes, but that's expected.

 

On my second rig i have 7900XT and boy is it a mess doing DCS VR with it. Apparently there are some issues with the drivers in VR.

But the real problem is i can't get reprojection to work at all (DCS freezes in main menu if reprojection is enabled) and while flying without reprojection my FPS ok-ish and stable when i don't look around, but moving my head makes it stutter quite a bit. Sometimes the headset just decides to turn off midflight, never have this on the nvidia.

 

Any tips to help with AMD. Is SteamVR better than openxr with amd, any settings i should try?

Edited April 4, 2023 by Grodin

[brain_delay]

I know the reverb G2 in particular has issues on amd moreso than any other headset. What's the cpu on the 2nd system? Have you done a full driver uninstall with DDU? 

[VirusAM]

AMD drivers and reprojection are a no-go.

I tried a 6900xt last year, it was a great card but only with very old drivers (21.5.2 iirc) it was possible to have a good working motion reprojection, so I sent the card back and picked a 3090.

I don't know about the new 7xxx generation but I read that many users have issues in VR so I think AMD didn't solve or did not care.

I also filed  bug reports and posted on AMD forums but no one cared to reply me (https://community.amd.com/t5/graphics/vr-motion-smoothing-not-working-with-recent-drivers/m-p/537998).

other relevant posts from other people here (https://community.amd.com/t5/drivers-software/amd-drivers-and-vr/m-p/538001#M159193)

Also @mbucchia (openxr toolkit creator) gave a very good explanation on why the reprojection doesn't work well with AMD cards and it has to do with encoder support and performance, don't know if  he has any news regarding AMD's 7th series family cards

Edited April 4, 2023 by VirusAM

[Grodin]

It has i7 9700. I've tried DDU too, no joy.

Even if i limit my fps to 50 for example it will stay very stable there, GPU is not having any trouble with that - but even then it stutters (while maintaining steady FPS) if i turn my head at all. Its kind of a jitter, for every 5 degrees i turn my head right it randomly goes back left 1 degree kind if thing, very annoying.

[Thinder]

On 4/4/2023 at 4:11 PM, VirusAM said:

AMD drivers and reprojection are a no-go.

I tried a 6900xt last year, it was a great card but only with very old drivers (21.5.2 iirc) it was possible to have a good working motion reprojection, so I sent the card back and picked a 3090.

I don't know about the new 7xxx generation but I read that many users have issues in VR so I think AMD didn't solve or did not care.

I also filed  bug reports and posted on AMD forums but no one cared to reply me (https://community.amd.com/t5/graphics/vr-motion-smoothing-not-working-with-recent-drivers/m-p/537998).

other relevant posts from other people here (https://community.amd.com/t5/drivers-software/amd-drivers-and-vr/m-p/538001#M159193)

Also @mbucchia (openxr toolkit creator) gave a very good explanation on why the reprojection doesn't work well with AMD cards and it has to do with encoder support and performance, don't know if  he has any news regarding AMD's 7th series family cards

 

it appears that AMD drivers are NOT the issue, Microsoft Windows update are.

AMD issued a warning through AMD Software asking us to check on our drivers because Mricrosoft updates replaces them with their own (see my signature), some updates contains a Multi-Plane Overlay (MPO) setting that prevents the GPU to boost, and if you try it manually you end up with crashes with mention "Driver Time Out".

The solution is to disable Multi-Plane Overlay and Microsoft automatic update before playing, check the signature of your drivers, if necessary use DDU in safe mode to clean GPU Drivers registry and reinstall it.

I had the chance to test all AMD drivers for the RX 7900 XTX thanks to a link to a pool of AMD drivers provided by Overclockers who were helping me try to sort performances issues after I lost <> 10% in 3DMark Pro scores, there is no difference in performances between them.

Overclockers tested my GPU and didn't find any problem with it.

Test results at 4K MSAA X2 were near 20.000 before this "incident" occured, now my GPU reaches 19.629 but I have to O.C it manually, to leave the card enough headroom to boost leaving me to think there still are performances hurdles in O.S.

The GPU runs stable and cool (max 81°C at junction recorded under heavy load) in DCS as well, Maximum Frequency is 3050MHz, maximum VRAM Frequency is 2650MHz.

The problem is not 100% resolved as I am sure there are other things holding this card in O.S, I saw what it was capable of when I first tested it before the update and know its true potential, I'm sure its issues are not coming from the hardware and perhaps very little from its drivers.

I understand that people doesn't have the patience to try to figure out why their RX 7900 XTX doesn't perform as advertised, they didn't experience its true potential, but frankly, I wouldn't swap it for a 3090 or 4080.

As for concluding that performances and drivers are at fault it is simply shortsighted, lack of proper information doesn't help, if I didn't pay attention to AMD warning or saw the GPU speed before O.S update I might have come to the same conclusions, I'm happy I didn't.

The funny part of this all is that this app was created to sort out NVIDIA drivers issues but works perfect with AMD drivers. See the article below.

After updating to NVIDIA Game Ready Driver 461.09 or newer, some desktop apps may flicker or stutter when resizing the window on some PC configurations

Edited April 25, 2023 by Thinder

[Thinder]

I just conducted a RAM kit test on my AMD Ryzen 7 5800X 3D to demonstrate how Ram/CPU bounding can affect GPU performances.

It's all down to the controller and what load it can manage, if it struggle to manage the amount of data, it will throttle down and reduce the GPU channel bandwidth, it is particularly visible with Ryzen 3/3D CPUs.

First, the maximum number of rank is 4, frequency 3200MHz (unless you use B.die/Cl14 which offers a much tighter range of timings, in which case you can set up your RAM to 3600MHz).

Second, interleaving (available only with 8GB sticks) can also make a little difference.

What I noticed is that the difference between 3200MHz and 3600MHz is negligible, one advantages the CPU scores while to other boosts the GPU score, with the 3600MHz kit you can chose which one you prefer.

Using a 4 X 16 GB kit will result in a penalty in performance of <> 24% over the same sticks used in a X 2 configuration.

Here are the scores obtained with both 4 X 8 and 2 X 16GB kits: 3DMark Pro Fire Strike Custom 4K 2 X MSAA.

Graphics score:

19 666 vs 19 603

Graphics test 1:

97.95 FPS vs 97.41 FPS

Graphics test 2:

75.87 FPS vs 75.76 FPS

Physics test:

77.16 FPS vs 75.82 FPS

Combined score:

51.36 FPS vs 51.16 FPS

>>>

As you can see for yourself, the best combo is the 4 X 8GB 3600MHz Cl14, GPU settings are the same as above and the card ran stable and cool.

Thermal is also a factor, the cooler, the higher the GPU will go in terms of frequency once you set up the maximum frequencies, it will limit itself when it reaches the thermal limit of <> 80/81°C, so room temperature and fan settings will play a role there and can explain some inconsistency in your test results, best case cooling is better.

In short if some experience issues with a particular headset, it might not come from the GPU potential performance or its driver.

[EightyDuce]

So at best, the difference is ~2% and at worst, sub 1%....ie. Margin of error. 

5 hours ago, Thinder said:

I just conducted a RAM kit test on my AMD Ryzen 7 5800X 3D to demonstrate how Ram/CPU bounding can affect GPU performances.

It's all down to the controller and what load it can manage, if it struggle to manage the amount of data, it will throttle down and reduce the GPU channel bandwidth, it is particularly visible with Ryzen 3/3D CPUs.

First, the maximum number of rank is 4, frequency 3200MHz (unless you use B.die/Cl14 which offers a much tighter range of timings, in which case you can set up your RAM to 3600MHz).

Second, interleaving (available only with 8GB sticks) can also make a little difference.

What I noticed is that the difference between 3200MHz and 3600MHz is negligible, one advantages the CPU scores while to other boosts the GPU score, with the 3600MHz kit you can chose which one you prefer.

Using a 4 X 16 GB kit will result in a penalty in performance of <> 24% over the same sticks used in a X 2 configuration.

Here are the scores obtained with both 4 X 8 and 2 X 16GB kits: 3DMark Pro Fire Strike Custom 4K 2 X MSAA.

Graphics score:

19 666 vs 19 603

Graphics test 1:

97.95 FPS vs 97.41 FPS

Graphics test 2:

75.87 FPS vs 75.76 FPS

Physics test:

77.16 FPS vs 75.82 FPS

Combined score:

51.36 FPS vs 51.16 FPS

>>>

As you can see for yourself, the best combo is the 4 X 8GB 3600MHz Cl14, GPU settings are the same as above and the card ran stable and cool.

Thermal is also a factor, the cooler, the higher the GPU will go in terms of frequency once you set up the maximum frequencies, it will limit itself when it reaches the thermal limit of <> 80/81°C, so room temperature and fan settings will play a role there and can explain some inconsistency in your test results, best case cooling is better.

In short if some experience issues with a particular headset, it might not come from the GPU potential performance or its driver.

What were the timings on the two sets of RAM, primary and secondary? 

Edited April 25, 2023 by EightyDuce