Did anyone see an improvement with the AMD RTX 7900 XTX?

With this new version. Not seeing it so far. Had no idea the AMD chip didn’t fully utilize TensorRT. I am going to compare it with a Nvidia RTX 4080.

I’m also interested about the AMD 7900XTX performance with TVAI v3.1.x. Can anyone test the card and share with us?

If someone is looking for NV 4090/3090 performance, I got some data can share with you guys.

Upscale from 1080p to 4k(using default setting, v3.1.0.4b)
NV 4090 Proteus ~5.6fps
NV 3090 Proteus ~2.9fps
NV 4090 Artemis ~7.4fps
NV 3090 Artemis ~3.8fps

3 Likes

Thanks twpotato, for the Nvidia data, very much appreciated.

I just recently came into some good fortune and built a new machine w/ a 7900X3D & 7900 XTX. Initial testing was in 2.6.4. 640x360->1920x1080 w/ Proteus v3 is running at 0.18 sec/frame (5.56fps), running at ~30% processor, 20% GPU load, 6gb VRAM usage.

3.1.8 runs at 6.69fps, with roughly the same load across both CPU and GPU.

Using Gaia on the same same video, 2.6.4 achieved 9.09 fps, with GPU load at 38% and CPU at 13%, using 8.3gb of VRAM.

Gaia in 3.1.8 achieved ~16.7fps, with GPU load at 61%, CPU load at 3%, and VRAM utilization at 8.1gb.

4 Likes

Thanks for your sharing, and just let you know the tvai newest version has embeded performance test, please go to below link and post your result if you could.
https://community.topazlabs.com/t/performance-benchmarking/41416

seems topaz labs is not yet using the new matrix cores in the new rdna3 series.

nod.ai has updated stable diffusion to use them on the 7xxx and performance is very good. so it seems topaz labs has some work to do :slightly_smiling_face:

1 Like

link seems not working.

I recently upgraded from a 5700 XT to a 7900 XTX and my FPS during processing doubled as a result. However, there is still ALOT of optimization to be done with AMD GPUs. The 7900 XTX should be able to compete with a 4080 but instead performance is currently less than half.

1 Like

I second your request!#AMD optimisation

2 Likes

About one week ago I swapped out my Nvidia 3090 for AMD 7900 XTX. I used the Topaz benchmark tool before/after the change. The video card was the only change I am aware of to the system.

Comparison using the 1920x1080 (FHD) Benchmark:


Comparison using the 3840x2160 (4K) Benchmark:


As shown, the 7900 XTX is not as fast, but there may be other tradeoffs not yet considered, E.G. power consumption and heat (of which the 3090 has plenty, and the XTX seems to not sweat it too much no matter what it’s doing). Maybe/maybe not, I just don’t know. I am glad I switched because it is significantly faster in some VR games I use my PC for, and so that was worth it. I’ll just have to be a little more patient with my video work until optimizations are made. Oh, the tradeoffs! :slight_smile:

5 Likes

You could try to use the Adrenalin 23.5.2 driver in Video AI.
AMD is currently reworking their compute APIs and it seems that there performance improvements but also regressions with the newer 23.7.x drivers.

Artemis 1X at 1080p is 16 fps with 23.7.x and 20 fps with 23.5.x

I also think that the big issue is that Topaz Video AI isn’t fully utilizing the full potential of the RDNA2/RDNA3 chips. The 7900 XTX is hovering around ~2-2.3 GHz while using 150-170 watts and 50-60 percent of utilization in Task-Manager. There is definitely room for improvements.

2 Likes

I have both cards a 7900xtx and a 4070.

The 4070 is some how faster in Topaz for some unknown reason.

1 Like

Tensor