Trending

Nvidia to drop CUDA support for Maxwell, Pascal, and Volta GPUs with the next major Toolkit release

VRAM requirements for all Qwen3 models (0.6B–32B) – what fits on your GPU?

[D] Are Two 5090 GPUs Sufficient for AI R&D into Novel Architectures?

RTX 5060 Ti 16GB sucks for gaming, but seems like a diamond in the rough for AI

We fit 50+ LLMs on 2 GPUs — cold starts under 2s. Here’s how.

[D] usefulness of learning CUDA/triton

LLM GPU calculator for inference and fine-tuning requirements

ubergarm/Qwen3-30B-A3B-GGUF 1600 tok/sec PP, 105 tok/sec TG on 3090TI FE 24GB VRAM

[D] Are weight offloading / weight streaming approaches like in Deepseek Zero used frequently in practice? (For enabling inference on disproportionately undersized GPUs)

NVIDIA

Save item

k8s-device-plugin

NVIDIA device plugin for Kubernetes

ROCm

Save item

ROCm

AMD ROCm™ Software - GitHub Home

NVIDIA

Save item

open-gpu-kernel-modules

NVIDIA Linux open GPU kernel module source

NVIDIA

Save item

warp

A Python framework for high performance GPU simulation and graphics