Search
216 results for “denzilferreira”
-
Mad props to pinokio.computer for how streamlined and easy they made installing and configuring ComfyUI with compatibility for 5700XT ROCm combo!
-
The best hack you can do in life is to move. #AntiDesk #Recovery #Sports #ActiveAging
-
The best hack you can do in life is to move. #AntiDesk #Recovery #Sports #ActiveAging
-
The best hack you can do in life is to move. #AntiDesk #Recovery #Sports #ActiveAging
-
The best hack you can do in life is to move. #AntiDesk #Recovery #Sports #ActiveAging
-
So, good news. ROCm 6.3.4 and PyTorch 2.4.0 seems stable enough with gfx1103 if I use HSA override for 11.0.0, using latest firmware blobs and kernel 6.13.10 on Fedora 41.
In your Dockerfile, build your AI app from:
```
FROM rocm/pytorch:rocm6.3.4_ubuntu24.04_py3.12_pytorch_release_2.4.0
``` -
Last evening, I spent some time investigating how to generate video from a text prompt and created this short clip. I was thinking it would require more GPU power, but all worked out OK. Took around 100 seconds to create this clip. The model used is Wan2.1, a 1.3b parameter model that fits a 8GB VRAM GPU (AMD/Nvidia). I used ComfyUI running on my home lab. Impressive for a "tiny" model, but comes to demonstrate that model size does not equate to results. Give the little models a go. You may surprise yourself with the results.
-
Last evening, I spent some time investigating how to generate video from a text prompt and created this short clip. I was thinking it would require more GPU power, but all worked out OK. Took around 100 seconds to create this clip. The model used is Wan2.1, a 1.3b parameter model that fits a 8GB VRAM GPU (AMD/Nvidia). I used ComfyUI running on my home lab. Impressive for a "tiny" model, but comes to demonstrate that model size does not equate to results. Give the little models a go. You may surprise yourself with the results.
-
Last evening, I spent some time investigating how to generate video from a text prompt and created this short clip. I was thinking it would require more GPU power, but all worked out OK. Took around 100 seconds to create this clip. The model used is Wan2.1, a 1.3b parameter model that fits a 8GB VRAM GPU (AMD/Nvidia). I used ComfyUI running on my home lab. Impressive for a "tiny" model, but comes to demonstrate that model size does not equate to results. Give the little models a go. You may surprise yourself with the results.
-
Last evening, I spent some time investigating how to generate video from a text prompt and created this short clip. I was thinking it would require more GPU power, but all worked out OK. Took around 100 seconds to create this clip. The model used is Wan2.1, a 1.3b parameter model that fits a 8GB VRAM GPU (AMD/Nvidia). I used ComfyUI running on my home lab. Impressive for a "tiny" model, but comes to demonstrate that model size does not equate to results. Give the little models a go. You may surprise yourself with the results.
-
Folks with AMD RX 9070 and Linux: enable overdrive by adding this to your kernel arguments: amdgpu.ppfeaturemask=0xffffffff. Reboot so that the next steps are available.
Install CoreCtrl and in the System > Software tab, make sure you are on Mesa 25.1.0 and kernel 6.13.5+. In the Profiles tab, change Performance mode to Advanced, change the Power profile to 3D Fullscreen. Set voltage offset to -10mV and reduce the maximum memory MHz by 100MHz (yes, do it).
With the lower voltage, the GPU core clock will get closer to XT levels, and reducing the memory clock a tiny bit will make it stable. Your non XT will perform better, cooler and stable. Saw about 10-20fps more in Returnal with these settings!
PS: tested on Fedora 41, Gigabyte RX 9070 Gaming OC card.
-
My T495 modded with Oculink adapter is now complete! I got this from Etsy and it was perfect for the adapter I had!
#thinkpad #t495 #oculink #egpu
E-GPU holder / stand / dock / OCulink x4 / OCulink x8 by SimonsPrintSolution on Etsy https://www.etsy.com/fi-en/listing/1677870345/e-gpu-holder-stand-dock-oculink-x4
-
My T495 modded with Oculink adapter is now complete! I got this from Etsy and it was perfect for the adapter I had!
#thinkpad #t495 #oculink #egpu
E-GPU holder / stand / dock / OCulink x4 / OCulink x8 by SimonsPrintSolution on Etsy https://www.etsy.com/fi-en/listing/1677870345/e-gpu-holder-stand-dock-oculink-x4
-
My T495 modded with Oculink adapter is now complete! I got this from Etsy and it was perfect for the adapter I had!
#thinkpad #t495 #oculink #egpu
E-GPU holder / stand / dock / OCulink x4 / OCulink x8 by SimonsPrintSolution on Etsy https://www.etsy.com/fi-en/listing/1677870345/e-gpu-holder-stand-dock-oculink-x4
-
My T495 modded with Oculink adapter is now complete! I got this from Etsy and it was perfect for the adapter I had!
#thinkpad #t495 #oculink #egpu
E-GPU holder / stand / dock / OCulink x4 / OCulink x8 by SimonsPrintSolution on Etsy https://www.etsy.com/fi-en/listing/1677870345/e-gpu-holder-stand-dock-oculink-x4
-
First image created using CPU and took 20 minutes. Second one with a 5700XT GPU and took 20 seconds.
-
First image created using CPU and took 20 minutes. Second one with a 5700XT GPU and took 20 seconds.
-
First image created using CPU and took 20 minutes. Second one with a 5700XT GPU and took 20 seconds.
-
First image created using CPU and took 20 minutes. Second one with a 5700XT GPU and took 20 seconds.
-
I've now added Automatic1111 (Stable Diffusion) to my Ollama deployment running on my T495 + eGPU 5700XT by using moophlo/stable-diffusion Docker image. I initially tried to set this up but there are so many things that need to align. This worked OOTB.
https://hub.docker.com/r/moophlo/stable-diffusion
Create a folder, create Dockerfile, create docker-compose.yml file and paste their contents. Make sure to update the username and password you'd like to use to login on the UI. Add it to Ollama WebUI and bam. You got image generation running!
-
I've now added Automatic1111 (Stable Diffusion) to my Ollama deployment running on my T495 + eGPU 5700XT by using moophlo/stable-diffusion Docker image. I initially tried to set this up but there are so many things that need to align. This worked OOTB.
https://hub.docker.com/r/moophlo/stable-diffusion
Create a folder, create Dockerfile, create docker-compose.yml file and paste their contents. Make sure to update the username and password you'd like to use to login on the UI. Add it to Ollama WebUI and bam. You got image generation running!
-
I've now added Automatic1111 (Stable Diffusion) to my Ollama deployment running on my T495 + eGPU 5700XT by using moophlo/stable-diffusion Docker image. I initially tried to set this up but there are so many things that need to align. This worked OOTB.
https://hub.docker.com/r/moophlo/stable-diffusion
Create a folder, create Dockerfile, create docker-compose.yml file and paste their contents. Make sure to update the username and password you'd like to use to login on the UI. Add it to Ollama WebUI and bam. You got image generation running!
-
I've now added Automatic1111 (Stable Diffusion) to my Ollama deployment running on my T495 + eGPU 5700XT by using moophlo/stable-diffusion Docker image. I initially tried to set this up but there are so many things that need to align. This worked OOTB.
https://hub.docker.com/r/moophlo/stable-diffusion
Create a folder, create Dockerfile, create docker-compose.yml file and paste their contents. Make sure to update the username and password you'd like to use to login on the UI. Add it to Ollama WebUI and bam. You got image generation running!
-
I just modded my T495 to have an Oculink port so I can plug in an eGPU! Will do some benches soon. Now need to mod the bottom case so I don't have to use it nude 🤪
-
I just modded my T495 to have an Oculink port so I can plug in an eGPU! Will do some benches soon. Now need to mod the bottom case so I don't have to use it nude 🤪
-
I just modded my T495 to have an Oculink port so I can plug in an eGPU! Will do some benches soon. Now need to mod the bottom case so I don't have to use it nude 🤪
-
As Thinkpad T495 does not have Thunderbolt (because it's AMD Ryzen Zen+) I figured I could add Oculink to the free WWAN M.2 2248 slot to run an external GPU for some gaming and local LLMs.
This slot can be used with nvme SSDs and some people have done so. So I thought I could use an M2 Oculink there as well since it's nvme too. Nope. This slot is compatible only with Lenovo's firmware whitelisted WWAN cards or nvme SSD disks 😕
So the only way that this will work is by getting an Oculink adapter for the larger M2 slot where I currently have my SSD, and use a smaller sized 2248 slot SSD (same as Steam Decks) since there is no firmware whitelisting on that slot. This means sacrificing the SSD speeds (2 lanes), and Oculink will have more bandwidth (4 lanes). Starting to think this is not worth it.
Anyway, was a fun experiment.
#thinkpad #t495 #egpu #oculink #ryzen -
As Thinkpad T495 does not have Thunderbolt (because it's AMD Ryzen Zen+) I figured I could add Oculink to the free WWAN M.2 2248 slot to run an external GPU for some gaming and local LLMs.
This slot can be used with nvme SSDs and some people have done so. So I thought I could use an M2 Oculink there as well since it's nvme too. Nope. This slot is compatible only with Lenovo's firmware whitelisted WWAN cards or nvme SSD disks 😕
So the only way that this will work is by getting an Oculink adapter for the larger M2 slot where I currently have my SSD, and use a smaller sized 2248 slot SSD (same as Steam Decks) since there is no firmware whitelisting on that slot. This means sacrificing the SSD speeds (2 lanes), and Oculink will have more bandwidth (4 lanes). Starting to think this is not worth it.
Anyway, was a fun experiment.
#thinkpad #t495 #egpu #oculink #ryzen