GPU Benchmark! 4070 4080 4090 3080 3090 #2970
Replies: 17 comments 1 reply
-
2070 rtx mobile |
Beta Was this translation helpful? Give feedback.
-
Nvidia RTX 4090 FE 100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:03<00:00, 6.17it/s] (edit: used default resolution first go around) |
Beta Was this translation helpful? Give feedback.
-
CPU vs GPU 1024x1024 2 steps Intel 13900KS 24,25 sec. https://files.catbox.moe/f4wb53.mp4 CPU.vs.GPU.test_2.mp4 |
Beta Was this translation helpful? Give feedback.
-
RTX 3090 |
Beta Was this translation helpful? Give feedback.
-
Just for laughs ... |
Beta Was this translation helpful? Give feedback.
-
RTX 3070 OC Drivers:~528.xx 100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:08<00:00, 2.26it/s] |
Beta Was this translation helpful? Give feedback.
-
My poor old GPU 😢 |
Beta Was this translation helpful? Give feedback.
-
RTX 4080
|
Beta Was this translation helpful? Give feedback.
-
Would love to get some 3060 / 4060 data to better complete the regression... Then I'll post proposed for every 3000/4000 card. |
Beta Was this translation helpful? Give feedback.
-
RTX 4070 12Gb got prompt |
Beta Was this translation helpful? Give feedback.
-
WSL2 (6.1.x kernel) 3060Ti
Tesla P40
No xformers 3060ti
Tesla P40
|
Beta Was this translation helpful? Give feedback.
-
4090 FE, Ryzen 7900X, Samsung 980 2TB M.2 got prompt |
Beta Was this translation helpful? Give feedback.
-
RTX 3070 Laptop GPU
|
Beta Was this translation helpful? Give feedback.
-
RTX 3060, Linux, Torch 2.2.2 (cu121)
Just for fun, here's a onediff-accelerated run after warmup with both the model and the VAE accelerated (it really is worth using, the speedup is massive and it handles changes in resolution just fine):
|
Beta Was this translation helpful? Give feedback.
-
4070 ti super got prompt |
Beta Was this translation helpful? Give feedback.
-
Could we Do a benchmark for GPU´s ?
Need to get new hardware .
To keep it simple we could use the default load worklow on 1024*1024 with sdxl 1.0.
Seed 1
I know that ram comes in play when workflows get more complex , but for the start simple ist best.
Please post the second run. first run ist a little bit corupted by loading data.
theory pudget:
toms hardware 1111:
Beta Was this translation helpful? Give feedback.
All reactions