NNathanael FETUE21h ago · 4 min readAttaching XDP without leaving a trace: bpf_link, kill -9 and a signal mask nobody testsBuilding AgentXDP, part 9. Previous: part 8, a lock-free queue broken on purpose. Next: part 10, a metrics endpoint a slow client cannot stall. An observability agent must do no harm. If it crashes, opegasus-nathanael-tech.hashnode.dev
LLantern Works1d ago · 5 min readGPU Rental Price Report — Week of Oct 26, 2026Disclosure: some vendor links on Lantern Works are referral/affiliate links (currently RunPod, Vast.ai, Thunder Compute, Novita AI, Ucartz, 99Stack Cloud and Clore.ai). Lantern Works may earn a commislanternworks.hashnode.dev
LLantern Works1d ago · 4 min readH100 for 100 hours: $320 cheapest tracked, plus GPU price alerts (Oct 29, 2026)Most "cheapest GPU" lists compare an H100 PCIe at one vendor with an H100 SXM at another and call it a deal. This post keeps the variants apart. All prices below were checked on Thursday, 2026-10-29 alanternworks.hashnode.dev
SSanskriti Harmukh1d ago · 15 min readOptimizing GPU Resource Planning with NVIDIA DynamoGPU resource allocation for LLM inference presents complex optimization challenges. Workloads exhibit variable request patterns, unpredictable sequence lengths, and strict latency requirements, makingvultr.hashnode.dev
SSanskriti Harmukh1d ago · 5 min readBuilding a PyTorch Container ImagePyTorch is a flexible machine learning library for building deep learning models that can perform a wide range of tasks such as image recognition and language processing. This guide walks through clonvultr.hashnode.dev
CCloudGPU1d ago · 10 min readWan 2.2 A14B at 720p on a 24 GB GPU: why it runs out of memory and what to changeIf you try Wan 2.2 T2V A14B at 1280×720 on a 24 GB card and get torch.OutOfMemoryError, you are not alone. A public bug report, reported in InvokeAI issue #9831 (2026-10-12) on an RTX 4090 24GB, lays cloudgpu.hashnode.dev
LLantern Works2d ago · 4 min readDescribe your agent, get the cheapest sourced infra stackLike-for-like saving: for the same 3,000 web-search queries a month, the cheapest plan with a cited price is Exa at $4/mo, against Firecrawl Hobby at $24 and SearchApi Developer at $40. That's $20 to lanternworks.hashnode.dev
LLantern Works2d ago · 3 min readGPU rental prices compared: H100, A100, RTX 4090 and more (Oct 2026)Like-for-like saving: 100 hours on one on-demand H100 (not spot or interruptible, not reserved) cost $199.00 at the cheapest tracked rate, RunPod H100 PCIe, On-demand (Community Cloud) at $1.99/hr. Thlanternworks.hashnode.dev
NNathanael FETUE3d ago · 4 min readA lock-free queue, broken on purpose to prove it worksBuilding AgentXDP, part 8. Part 7 built the histogram that the kernel and the agent fill the same way. This part builds the road events take inside the agent: a queue between two threads, with no lockpegasus-nathanael-tech.hashnode.dev
WWeiwei Zhang4d ago · 7 min readHow to Design an 800G InfiniBand XDR Network for Large-Scale GPU ClustersBuilding a large GPU cluster involves more than connecting servers to high-speed switches. Distributed AI training depends on fast, predictable communication between GPUs, and the network must maintaiaicplight.hashnode.dev
