Setting up an AI/ML environment from scratch?
We just published a comprehensive 7-step guide on configuring an Ubuntu bare-metal NVIDIA GPU server.

We cover the exact bash commands for:
Installing proprietary NVIDIA drivers
Setting up Miniconda
Installing PyTorch & TensorFlow with full CUDA support

Read the full technical tutorial here:
https://www.eservers.uk/tutorials/howto/set-up-ai-ml-environment-gpu-server/

#MachineLearning #AI #PyTorch #TensorFlow #Ubuntu #Linux #DataScience #DevOps #GPUComputing #OpenSource

Unlock the full power of virtualization with GPU passthrough in Microsoft Hyper-V.
Give VMs direct GPU power for AI, VDI, and graphics-intensive workloads.

Learn how it works.
https://zurl.co/K3uXp

#HyperV #GPUComputing #Virtualization #VDI #AIInfrastructure #Microsoft #ITInfrastructure

Small making-of of my “True Beauty Is So Painful” piece (listening to “True Beauty Is So Painful” by Oomph! in the background), because “AI art = just pressing a button” is still a thing.

Here I’m showing briefly (15 MB max file upload) my SDXL workflow in ComfyUI, from node structure to model choice to parameters.

LoRAs in this setup are only linked to the positive prompt, because I wanted to fine-tune their weights there specifically, without affecting the negative prompt.

During rendering, I ran in parallel:
- GPU load with radeontop, you can clearly see how on RDNA2 everything (matrix multiplications, convs, etc.) runs over the shaders
- Temperatures & power states briefly shown with corectrl

Peak at 187 W, hotspot briefly at 97 °C
RDNA2 doing RDNA2 things…

Video workflow:
- recorded with OBS
- edited in Kdenlive
- transcoded with VAAPI (H.264)

No cloud, just decisions, iteration and real hardware.
Everything runs on Linux + ComfyUI (FOSS), so anyone can set this up.
No GPU? No problem, you can also run it using PyTorch’s CPU backend, just much slower.

#AIArt #ComfyUI #SDXL #stablediffusion #LoRA #FOSS #Linux #AMD #RDNA2 #GPUComputing #OpenSource #AIWorkflow #OBS #Kdenlive #VAAPI #DigitalArt #MakingOf #AIProcess #NoCloud

🔬 Breaking research shows how AI labs are revolutionizing computational efficiency! Token warehousing strategy could dramatically reduce GPU processing waste in large language models. Researchers uncover innovative techniques that might reshape machine learning infrastructure. Fascinating insights into cutting-edge AI optimization! #AI #MachineLearning #GPUComputing #LargeLanguageModels

🔗 https://aidailypost.com/news/ai-researchers-reveal-token-warehousing-strategy-cut-gpu

Người dùng đang tìm cách triển khai suy luận cục bộ cho mô hình lớn Qwen2.5-72B trên 2 GPU L40 (48GB VRAM mỗi chiếc) nhưng gặp trở ngại. Khi dùng Huggingface, quá trình bị treo, còn vLLM thì báo lỗi khởi tạo WorkerProc. Anh ấy đang tìm kiếm các gợi ý để giải quyết vấn đề phân chia mô hình và tăng tốc suy luận trên hệ thống đa GPU.
#LLM #AITech #vLLM #Huggingface #LocalInference #GPUComputing #Qwen2_5_72B

https://www.reddit.com/r/LocalLLaMA/comments/1q7gr9w/local_inference_with_big_model_shared_

New in the #VirtualObservatory: “Order Computational and Storage Resources at FAI” by Fesenkov Astrophysical Institute
https://dachs.fai.kz/soft_order_sims/q/compres/info
#AstronomicalInstrumentation #ComputationalAstronomy #GpuComputing #AutomatedTelescopes
Information on Service 'Order Computational and Storage Resources at FAI'

So sánh chi phí khi fine-tune Llama 3 70B:
- **AWS H100**: $4.50/giờ, setup 45 phút (cài driver + tải dữ liệu)
- **Cụm RTX4090s phân tán**: $2.00/giờ, setup 5 phút
Giả định: Cụm chậm hơn 1.6x do WAN.
📊 Kết quả:
• Chạy một lần dài → AWS nhanh hơn.
• Vòng nghiên cứu (3-4 lần chạy nhỏ) → Cụm RTX4090s rẻ hơn và cạnh tranh về tổng thời gian nhờ giảm chi phí "setup" lặp lại.
#AI #GPUComputing #CostOptimization #Llama3 #TríTuệNhânTạo #MáyTínhGPU #TốiƯuChiPhí

https://www.reddit.com/r/Loc

llama.cpp trên llama-server gặp vấn đề hiệu suất lớn khi dùng eGPU qua Thunderbolt 4. Tốc độ prefill (xử lý prompt) giảm từ ~2500 t/s (1 GPU) xuống ~150 t/s (2 GPU, 1 qua TB4). Có phải độ trễ của TB4 là thủ phạm chính? Liệu Oculink có tốt hơn?

#llama_cpp #llama_server #eGPU #Thunderbolt4 #LLM #AIPerformance #GPUComputing #HiệuSuấtAI #TínhToánGPU #PhầnCứngAI #MôHìnhNgônNgữ

https://www.reddit.com/r/LocalLLaMA/comments/1q08h2t/llamaserver_massive_prefill_cliff_2500_ts_150_ts/

Se você já montou ou pelo menos configurou um PC na vida, provavelmente tem essa imagem na cabeça:

CPU: aquela parte pequena e quadrada, onde se encaixa direto na placa-mãe, no soquete, coloca pasta térmica e prende o cooler em cima.

#technology #tecnologia #computer #computador #computadores #CPU #GPUs #GPUcomputing #nvidia #GeForce #AMD #Ryzen #Server #servidores

https://www.moprius.com/2025/12/cpu-e-gpu-de-computadores.html

CPU e GPU de computadores

CPU: aquela parte pequena e quadrada, onde se encaixa direto na placa-mãe, no soquete, coloca pasta térmica e prende o cooler em cima. GPU: uma placa

Moprius - Explorando o conhecimento
CUDA 13.1 Reinvents GPU Development — The Biggest Leap in Two Decades

CUDA 13.1 introduces the new CUDA Tile model, the biggest overhaul in 20 years, redefining GPU programming for AI, HPC, and future Blackwell GPUs

BuySellRam