GPU Benchmark! 4070 4080 4090 3080 3090 #2970
Replies: 68 comments 81 replies
|
2070 rtx mobile |
|
Nvidia RTX 4090 FE 100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:03<00:00, 6.17it/s] (edit: used default resolution first go around) |
NVIDIA RTX5060TI 16GB |
|
CPU vs GPU 1024x1024 2 steps Intel 13900KS 24,25 sec. https://files.catbox.moe/f4wb53.mp4 CPU.vs.GPU.test_2.mp4 |
|
Not quite a benchmark, you Gorilla. |
|
RTX 3090 |
NVIDIA RTX 5060TI 16GB |
|
Just for laughs ... |
|
My 1070 is my favorite, was my longtime friend still in good use, pretty good for what it was, 3060 is 2 faster with only half as many additional cores, but with the 4090 I have now, uncomparable. |
|
RTX 3070 OC Drivers:~528.xx 100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:08<00:00, 2.26it/s] |
I don´t now, where to change settings for "Sage attention" ?! Now, the settings are correct, as in the posted Picture above... 
|
|
Thanks for the clarification. The settings I mentioned are CLI args so you need to install sage attention and then start comfyui with Thinking about it would actually be better if we would know the startup logs per benchmark, too or better the CLI args + the startup log. Even different versions of pytorch can have impact on the performance, especially for newer cards. |
|
@BBCreativo I have the same video card, so I was confused. Here is my test with sd_xl_base_1.0.safetensors and this is from v1-5-pruned-emaonly-fp16 --fast +xformers --use-sage-attention But I compile xformers, Flash Attention, and SageAttention for myself. |
I´ve runned it, with the settings from the picture, on top of this thread? |
|
I thought that the first time you tested with a small v1-5-pruned-emaonly-fp16 and 512x512, which is why you got such high speed. I was wrong, I apologize. |
|
My poor old GPU 😢 |
|
RTX 4080 |
|
Would love to get some 3060 / 4060 data to better complete the regression... Then I'll post proposed for every 3000/4000 card.
|
|
RTX 3090 driver:551.61 RTX 3090 driver:531.18 |
|
RTX 4060ti 16GB |
|
RTX 4070 12Gb got prompt |
|
WSL2 (6.1.x kernel) 3060Ti Tesla P40 No xformers 3060ti Tesla P40 |
|
4090 FE, Ryzen 7900X, Samsung 980 2TB M.2 got prompt
|
|
не менее 9,1 должно быть. скиньте дрова моложе 2024г. совсем другая карта будет. 4090 будет как изначально с 94TF на fp16 и 32. |
|
Моложе 2024г это как? Этот пост итак из апреля 2024. Ну и тут ни у кого не вижу на 4090 чтобы было 9 it/s |
|
RTX 3070 Laptop GPU
|
|
R7 5700X3D got prompt |
|
3080TI [Crystools INFO] Crystools version: 1.22.1 1024x1024 SDXL BASE 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:04<00:00, 4.01it/s] |
|
Intel B580 Steel Legend OC 12 GB Total VRAM 11874 MB, total RAM 32679 MB When you use Pytorch(as well as I could manage to get it working at all) you have to live with some warnings now and then Standard operation SDXL BASE 1.0 2025-05-12 17:01:33,326 - _logger.py - IPEX - INFO - Currently split master weight for xpu only support sgd Slightly overclocked GPU For 12gb card that didnt cost me NVIDIA money I am reasonable happy.
|
(ailibuv) C:\Users\mrced\ComfyUI>python main.py ^
|
|
Above was with OC on the card, without it the decoding takes double the time. iterations is almost the same but look at this image: |
|
I have dual intel gpu(ACER A770 16GB, B580 TRI 12GB) installed, so I use pytorch version: 2.7.0+xpu 9700X+B580 TRI(PCIe 4.0 x4) 9700X+ACER A770(PCIe 4.0 x16) Don't know why, but my A770 is just not even close to B580. |
|
i installed a conda env with latest pytorch. went from 4.18 it s to 3.60 it s using python main.py --disable-xformers --disable-ipex-optimize --bf16-unet --bf16-vae --bf16-text-enc start args |
|
After the changes I made to implement Before, the fastest speed I saw using the right environments and driver versions and etc. was around 1.31it/s so this is a good ~12-13% improvement. This is only marginally slower than the results above for the A770 which I think is running headless without the desktop environment overhead. Would like to see if anyone wants to test with the changes now in. |
|
I updated all elements, settings and drivers, it became faster |
|
Can this be that the B580 is faster than the RTX 5060 TI 16GB? |
|
Nvidia is better than all other cards, they simply work with all AI applications, maybe in some cases B580 is better, but I doubt it's worth it |
|
Let me know, a month has passed. Maybe new drivers have been released and the speed has changed? |
|
Hardware:
Software:
Seed 0 & 1 Average speed of 16 random seeds: 10.50 it/s | 2.26 seconds |
|
GTX 1080 + Intel 4790K |
|
GTX 970 + i7-8700k |
|
RTX 4060 + i7-8700k |
|
6700 XT user here, just wanted to share my numbers. SDXL Gen: |
|
6800XT, 5800X3D, Bazzite Linux, kernel Linux 6.13.7-107, ROCm [START] Security scan Prestartup times for custom nodes: Checkpoint files will always be loaded safely. Loading: ComfyUI-Manager (V3.33.8)[ComfyUI-Manager] network_mode: public ComfyUI Revision: 3645 [c5de495] *DETACHED | Released on '2025-07-08'Starting server To see the GUI go to: http://0.0.0.0:8188 got prompt |
|
1080 TI, 3700X, TrueNAS Scale, kernel 6.12.15 [START] Security scan ComfyUI-Manager: installing dependencies done.** ComfyUI startup time: 2025-07-10 16:11:44.737 Checkpoint files will always be loaded safely.
|
|
start with: B580 TRI(PCIe 4.0 x4) ACER A770(PCIe 4.0 x16) Today I also tried comfyui_openvino. All the offload is on cpu. It takes forever to compile, and if you change model or selected device without enable cache, you have to compile it again. |
|
очень хороший результат! |
|
Recently built my first PC, major upgrade from an old Dell G7 laptop that was constantly thermal throttling. Intel 245k NON-OC OC using MSI Afterburner, +330core/+2600mem (real core speed is ~3150-3200MHz) SageAttention 2++, NON-OC SageAttention 2++, OC SageAttention 2++, Fast FP16 Accumulation, Non-OC SageAttention 2++, Fast FP16 Accumulation, OC Nice little card. The image difference using both Sage and Fast FP16 is unnoticeable with this SDXL Benchmark, but with my standard NoobAI + Euler A workflow, the difference can be pretty significant though without noticeable quality loss. Seems Intel may have the better AI card in terms of raw speed (per older posts). Can't wait to see what the B60 and Celestial can do. |
|
System: Windows 11 Pro x64 24H2 (Build: 26100.6584) ComfyUI ROCm v0.3.59 (stable) via Stability Matrix v.2.15.0 (stable) Workarounds used: and added two lines in main.py: to get rid off for getting stuck in 1024x1024 only, can not create larger or smaller than 1024x1024, a big part of memory management issue as well. I has no idea why... but combining two workarounds worked great for me. Result: |
|
No problem! |
|
3080 Ti 12GB 
First couple of passes are slower due to loading/caching. Then it's 5 it/s. Faster than 5060 Ti at least xD. Quickly went through comments and made this little chart based on them: It's not very fair because software versions are different, but better than nothing I guess. Would be interesting to see Wan2.2 comparison. System: Ryzen 7700, 2x32GB 6000 CL32, ADATA 960 1TB. Edit: Tried pytorch attention, got the same speeds: 4.43 sec for generation. 
Btw. I had to manually remake the workflow from the topic starter comment. Because default SD XL workflow seems to have changed. Here is what I ran: |
and others
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
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:
All reactions