Error first try SD3 directml RX580

[KillyTheNetTerminal]

Error occurred when executing KSampler:

Expected all tensors to be on the same device, but found at least two devices, privateuseone:0 and cpu!

File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\nodes.py", line 1355, in sample
return common_ksampler(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise=denoise)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\nodes.py", line 1325, in common_ksampler
samples = comfy.sample.sample(model, noise, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\sample.py", line 43, in sample
samples = sampler.sample(noise, positive, negative, cfg=cfg, latent_image=latent_image, start_step=start_step, last_step=last_step, force_full_denoise=force_full_denoise, denoise_mask=noise_mask, sigmas=sigmas, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 794, in sample
return sample(self.model, noise, positive, negative, cfg, self.device, sampler, sigmas, self.model_options, latent_image=latent_image, denoise_mask=denoise_mask, callback=callback, disable_pbar=disable_pbar, seed=seed)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 696, in sample
return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 683, in sample
output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 662, in inner_sample
samples = sampler.sample(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 567, in sample
samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comf\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\k_diffusion\sampling.py", line 137, in sample_euler
denoised = model(x, sigma_hat * s_in, **extra_args)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 291, in call
out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 649, in call
return self.predict_noise(*args, **kwargs)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 652, in predict_noise
return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 277, in sampling_function
out = calc_cond_batch(model, conds, x, timestep, model_options)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\samplers.py", line 226, in calc_cond_batch
output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\model_base.py", line 103, in apply_model
model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float()
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comf\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comf\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
return forward_call(*args, **kwargs)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\ldm\modules\diffusionmodules\mmdit.py", line 961, in forward
return super().forward(x, timesteps, context=context, y=y)
File "C:\Users\WarMa\OneDrive\Escritorio\SD\comfyuai\ComfyUI\comfy\ldm\modules\diffusionmodules\mmdit.py", line 937, in forward
x = self.x_embedder(x) + self.cropped_pos_embed(hw, device=x.device).to(dtype=x.dtype)

[timesqueezer]

Can confirm the same error on a RTX 3050 / Intel Core i7-11800H notebook. The only difference is this line:
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

[KillyTheNetTerminal]

exactly cause you have Nvidea and Cuda

[KillyTheNetTerminal]

working on CPU but slow as hell. i3-9100f

[Cremesis]

Same problem for me

Exception during processing!!! Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!
Traceback (most recent call last):
  File "C:\Users\<myuser>\Downloads\comfyui\ComfyUI\execution.py", line 151, in recursive_execute
    output_data, output_ui = get_output_data(obj, input_data_all)

[jtyszkiew]

Seems to work for me on:

│Total VRAM 11980 MB, total RAM 64140 MB
│pytorch version: 2.3.0+cu121
│Set vram state to: NORMAL_VRAM
│Device: cuda:0 NVIDIA GeForce RTX 4070 : cudaMallocAsync
│VAE dtype: torch.bfloat16
│Using pytorch cross attention

Nvidia + CUDA

[kuldp18]

Can confirm the same error on a RTX 3050 / Intel Core i7-11800H notebook. The only difference is this line: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

I have the same error too..

[15Litrov]

A fresh manual install with nightly pytorch (other not tested) helped me overcome this problem.
1050ti 4gb + 32gb RAM

[kuldp18]

A fresh manual install with nightly pytorch (other not tested) helped me overcome this problem. 1050ti 4gb + 32gb RAM

can we just update the pytorch in the current install? and how is 4gb vram handling sd3 btw?

[15Litrov]

A fresh manual install with nightly pytorch (other not tested) helped me overcome this problem. 1050ti 4gb + 32gb RAM

can we just update the pytorch in the current install? and how is 4gb vram handling sd3 btw?

Maybe? I did not test.
About performance: 30 s/it for 1024x1024 with dualCLIP.

[kuldp18]

Guys the issue is fixed, please do an update!

[AlexBenjarmin]

Guys the issue is fixed, please do an update!

update Comfy UI?

[KillyTheNetTerminal]

yes it works now, update comfyui (I use manager) very slow per it. There's is a way to speed up this?

[KillyTheNetTerminal]

[ltdrdata]

You should not use dpmpp_2m, karras.
Just use euler, sgm_uniform.

karras is bad for SD3.

[kuldp18]

You should not use dpmpp_2m, karras. Just use euler, sgm_uniform.

karras is bad for SD3.

the official recommendation is dpm though, isn't euler too random according to sd3 architecture?

[ltdrdata]

You should not use dpmpp_2m, karras. Just use euler, sgm_uniform.
karras is bad for SD3.

the official recommendation is dpm though, isn't euler too random according to sd3 architecture?

https://comfyanonymous.github.io/ComfyUI_examples/sd3/

Official example is suggesting euler, sgm_uniform.
In my test.
dpmpp_2m sampler is ok.
but the scheduler must be one of normal, simple, sgm_uniform, ddim_uniform.

[KillyTheNetTerminal]

the same, the image is still noisy

[KillyTheNetTerminal]

[ltdrdata]

Try on cpu mode.

[Wallboy]

Same issue with just getting noisy generated images. 7900 XTX also running using DirectML.

Perhaps SD3 is not working with AMD GPUs/DirectML yet.

[kopaser6463]

Same issue, i nail it down a little bit to variable named out in sampling_function in samplers.py being different on cpu/directml.
Here a crazy path to it.
nodes.py -> samplers.py -> KSampler.sample -> sample(diferent one) -> CFGGuider.sample ->CFGGuider.inner_sample (sampler.sample(self, sigmas...)) -> sampler = sampler_object(self.sampler just a name) -> sampler_object -> ksampler -> KSAMPLER.sample(self, model_wrap, sigmas, extra_args...) -> model_k = KSamplerX0Inpaint(model_wrap, sigmas) ->
model_wrap is self in sampler.sample so CFGGuider() call return self.predict_noise() -> sampling_function(model) -> cfg_function(model) -> out.
It is different on cpu/directml.
Why? I don't know.

[Wallboy]

If anyone wants to get a working SD3 with AMD GPUs in the mean time, look up the ComfyUI Zluda fork and use that instead. Working great.

Just be warned that the first generation takes a little while as a bunch of databases are being processed. Similar to if you've ever used A1111 and Zluda, you had that same wait time for your first generation after installing it.

[KillyTheNetTerminal]

I never try and set up zluda for comfyui, this speed ups generations? compared to directml? how can I test this?