I have this error : Expected all tensors to be on the same device, but got tensors is on cpu, different from other tensors on cuda:0 (when checking argument in method wrapper_CUDA_cat)
I think there might be something mis-configured in the default workflow. I just opened the t2v workflow and tried a simple generation without changing anything and I hit OOM errors on a RTX 6000 Pro with 96gb of VRAM.
EDIT:
If anyone else runs into OOM errors which they think they shouldn't be getting, check your ENV. I accidentally created my ENV with python=3.14 when I changed the env to use python=3.11 the workflow worked fine.
Local workstation or 6000 pro on the cloud? I’ve noticed with Runpod, even on a “secure” instance, it hit OOM on a 6000 pro in the closest US servers, where a 5090 in Europe did not with the same workflows. This leads me to believe their servers are not secure and we could be booted off in the middle of a job, or run into an OOM, because someone is logging into the same server. They may be charging as secure but could be community. Who would know without auditing the company to see? Anyone else noticing this?
You might want to consider renting GPUs on a cloud service. Runpod is very popular, but there are tons of options, and Comfy offers their own cloud service as well. Do your research before investing a lot of money.
I have this error : Expected all tensors to be on the same device, but got tensors is on cpu, different from other tensors on cuda:0 (when checking argument in method wrapper_CUDA_cat)
Error.
SamplerCustomAdvanced
mat1 and mat2 shapes cannot be multiplied (466816x1 and 128x3)
Change the live preview method in Settings > Execution to 'none'. Check out here:
https://github.com/Comfy-Org/ComfyUI/issues/11656
it works for me
I did a test with my 4080 GPU and give memory errors. Which are the models minimum requirements?
I think there might be something mis-configured in the default workflow. I just opened the t2v workflow and tried a simple generation without changing anything and I hit OOM errors on a RTX 6000 Pro with 96gb of VRAM.
EDIT:
If anyone else runs into OOM errors which they think they shouldn't be getting, check your ENV. I accidentally created my ENV with python=3.14 when I changed the env to use python=3.11 the workflow worked fine.
Local workstation or 6000 pro on the cloud? I’ve noticed with Runpod, even on a “secure” instance, it hit OOM on a 6000 pro in the closest US servers, where a 5090 in Europe did not with the same workflows. This leads me to believe their servers are not secure and we could be booted off in the middle of a job, or run into an OOM, because someone is logging into the same server. They may be charging as secure but could be community. Who would know without auditing the company to see? Anyone else noticing this?
I cant speak much for how it works at the cloud vendors,. I am only working locally, since I have a RTX 6000 Pro in my workstation.
Thanks for the tip.
Does it work with AMD cards?
ResizeImageMaskNode.execute() missing 1 required positional argument: 'resize_type'
i get that error when trying to use the i2v workflow
Error
LTXVGemmaCLIPModelLoader
No files matching pattern 'tokenizer.model' found under C:\ComfyUI_windows_portable\ComfyUI\models
I’m sure there are FP8 and gguf models available. Which is what I’ve been having to do since the last two updates, even with a 3090.
I've this laptop with i9 13th gen and integrated graphics. It wouldn't work so great, right?
You might want to consider renting GPUs on a cloud service. Runpod is very popular, but there are tons of options, and Comfy offers their own cloud service as well. Do your research before investing a lot of money.
I'm building a pc and curious about comfyUi. What specs should i be looking for?
In general you want the best GPU you can afford with the most VRAM possible. For instance 3090 24GB > 5080 16GB
You also want a decent amount of RAM since Comfy can offload models to your system as well. I think 64GB is a good target.
Speed certainly matters, and is a worthwhile investment, but if you don't have enough memory, then you won't be able to run certain workflows at all.
There isn’t a specific spec depends on what you can afford. ComfyUI isn’t a video game with a minimum hardware requirement.