---
title: "#cuda articles | Hashnode"
description: "Explore 118 articles tagged #cuda on Hashnode. Tutorials, guides, and how-tos from the community."
url: "https://hashnode.com/tag/cuda"
---

Tag

# #cuda

118 posts · 24 followers

[Viswaretas Kotra](https://hashnode.com/@viswa-kotra) · [4d ago](https://hashnode.com/posts/measure-gpu-startup-to-the-first-usable-result/6af8fd189b5550202719251f)

### [Measure GPU startup to the first usable result](https://nodus-compute.hashnode.dev/measure-gpu-startup-to-the-first-usable-result?utm_source=hashnode&utm_medium=feed)

A GPU worker can be reachable while the useful result is still seconds away. If a benchmark stops at “ready,” it needs to say what became ready. Here is a deliberately fictional example from our runna

3 min read · [nodus-compute.hashnode.dev](https://nodus-compute.hashnode.dev/measure-gpu-startup-to-the-first-usable-result?utm_source=hashnode&utm_medium=feed)

[K. L. Aadithya](https://hashnode.com/@ash-grey) · [4d ago](https://hashnode.com/posts/how-i-built-an-llm-inference-engine-from-scratch-in-cuda-235-tokens-s-on-a-6-gb-laptop-gpu/6af8ac25576765a570c8a730)

### [How I Built an LLM Inference Engine From Scratch in CUDA (235 tokens/s on a 6 GB Laptop GPU)](https://aadithya.hashnode.dev/how-i-built-an-llm-inference-engine-from-scratch-in-cuda-235-tokens-s-on-a-6-gb-laptop-gpu?utm_source=hashnode&utm_medium=feed)

I built tinyinfer, an open-source LLM inference engine in C++/CUDA with no PyTorch or cuBLAS. It runs TinyLlama at 235 tokens/s on a 6 GB RTX 3050 laptop GPU.

11 min read · [aadithya.hashnode.dev](https://aadithya.hashnode.dev/how-i-built-an-llm-inference-engine-from-scratch-in-cuda-235-tokens-s-on-a-6-gb-laptop-gpu?utm_source=hashnode&utm_medium=feed)

[Viswaretas Kotra](https://hashnode.com/@viswa-kotra) · [4d ago](https://hashnode.com/posts/a-gpu-partition-changes-the-baseline/6af8892f1e59258b5d32ef54)

### [A GPU partition changes the baseline](https://nodus-compute.hashnode.dev/a-gpu-partition-changes-the-baseline?utm_source=hashnode&utm_medium=feed)

A runnable fictional example shows why partition profiles, sibling load and timing uncertainty belong in a kernel benchmark baseline.

3 min read · [nodus-compute.hashnode.dev](https://nodus-compute.hashnode.dev/a-gpu-partition-changes-the-baseline?utm_source=hashnode&utm_medium=feed)

[Viswaretas Kotra](https://hashnode.com/@viswa-kotra) · [5d ago](https://hashnode.com/posts/three-gpu-timing-numbers-can-all-describe-different-boundaries/6af7a70ec5555a2d0eb2f191)

### [Three GPU timing numbers can all describe different boundaries](https://nodus-compute.hashnode.dev/three-gpu-timing-numbers-can-all-describe-different-boundaries?utm_source=hashnode&utm_medium=feed)

A runnable synthetic worksheet separates GPU timing boundaries, overlap and incompatible clocks. Eight cases retained; no hardware measurements.

4 min read · [nodus-compute.hashnode.dev](https://nodus-compute.hashnode.dev/three-gpu-timing-numbers-can-all-describe-different-boundaries?utm_source=hashnode&utm_medium=feed)

[Akshay Suryawanshi](https://hashnode.com/@akshaysuryawanshi) · [6d ago](https://hashnode.com/posts/resizing-vllm-s-kv-cache-without-restarting-the-server/6af533f9230ef56d408e4992)

### [Resizing vLLM’s KV Cache Without Restarting the Server](https://akshaysuryawanshi.hashnode.dev/resizing-vllm-s-kv-cache-without-restarting-the-server?utm_source=hashnode&utm_medium=feed)

How I built a vLLM plugin to resize KV cache without restarting, using drain and preserve strategies, CUDA VMM, and Qwen3-32B tests on an H100.

13 min read · [akshaysuryawanshi.hashnode.dev](https://akshaysuryawanshi.hashnode.dev/resizing-vllm-s-kv-cache-without-restarting-the-server?utm_source=hashnode&utm_medium=feed)

[Anshu Gondi](https://hashnode.com/@anshugondi) · [Nov 4](https://hashnode.com/posts/why-vllm-sglang-and-flashattention-are-just-1970s-embedded-systems-algorithms/6aeacbba787e0a1857bf3f58)

### [Why vLLM, SGLang, and FlashAttention Are Just 1970s Embedded Systems Algorithms](https://anshugondi.hashnode.dev/why-vllm-sglang-and-flashattention-are-just-1970s-embedded-systems-algorithms?utm_source=hashnode&utm_medium=feed)

Demystifying GPU inference primitives by tracing FlashAttention, PagedAttention, and KV-cache reuse back to classic operating system memory paging and hardware scratchpads.

3 min read · [anshugondi.hashnode.dev](https://anshugondi.hashnode.dev/why-vllm-sglang-and-flashattention-are-just-1970s-embedded-systems-algorithms?utm_source=hashnode&utm_medium=feed)

[Aleksei Romanov](https://hashnode.com/@gfactor) · [Nov 3](https://hashnode.com/posts/teaching-a-27b-model-to-write-triton-kernels-sft-rl-gym/6aea252b0aa43068615c154e)

### [Teaching a 27B Model to Write Triton Kernels: SFT + RL Gym](https://gfactor.hashnode.dev/teaching-a-27b-model-to-write-triton-kernels-sft-rl-gym?utm_source=hashnode&utm_medium=feed)

In Part 1, we explored how GPU kernels work under the hood — why operator fusion and memory tiling matter, and how benchmarks like KernelBench evaluate whether an LLM can write kernels that actually b

27 min read · [gfactor.hashnode.dev](https://gfactor.hashnode.dev/teaching-a-27b-model-to-write-triton-kernels-sft-rl-gym?utm_source=hashnode&utm_medium=feed)

[Aleksei Romanov](https://hashnode.com/@gfactor) · [Nov 3](https://hashnode.com/posts/gpu-kernels-explained-cuda-triton-cublas-torch-compile-and-kernelbench/6ae986128db764f898d76a12)

### [GPU Kernels, Explained: CUDA, Triton, cuBLAS, torch.compile and KernelBench](https://gfactor.hashnode.dev/gpu-kernels-explained-cuda-triton-cublas-torch-compile-and-kernelbench?utm_source=hashnode&utm_medium=feed)

When you run a neural network in PyTorch, Python is only the coordinator. Behind every tensor operation — from a simple activation to a multi-head attention block — sits one or more GPU kernels: speci

28 min read · [gfactor.hashnode.dev](https://gfactor.hashnode.dev/gpu-kernels-explained-cuda-triton-cublas-torch-compile-and-kernelbench?utm_source=hashnode&utm_medium=feed)

[Anshu Gondi](https://hashnode.com/@anshugondi) · [Oct 27](https://hashnode.com/posts/stripping-the-python-tax-fintally-native-engineomplete-publication-ready-hashnode-article-formatted-cleanly-in-markdown-you-can-paste-this-directly-into-hashnode-s-editor-stripping/6ae07cab375e3174907ac4e8)

### [Here is the complete, publication-ready Hashnode article formatted cleanly in Markdown. You can paste this directly into Hashnode's editor. Stripping](https://anshugondi.hashnode.dev/stripping-the-python-tax-fintally-native-engineomplete-publication-ready-hashnode-article-formatted-cleanly-in-markdown-you-can-paste-this-directly-into-hashnode-s-editor-stripping?utm_source=hashnode&utm_medium=feed)

Here is the complete, publication-ready Hashnode article formatted cleanly in Markdown. You can paste this directly into Hashnode's editor. Stripping the Python Tax: How We Ported vLLM Concepts direct

3 min read · [anshugondi.hashnode.dev](https://anshugondi.hashnode.dev/stripping-the-python-tax-fintally-native-engineomplete-publication-ready-hashnode-article-formatted-cleanly-in-markdown-you-can-paste-this-directly-into-hashnode-s-editor-stripping?utm_source=hashnode&utm_medium=feed)

[Aman Karki](https://hashnode.com/@amankarki) · [Oct 26](https://hashnode.com/posts/gpu-fix-faster-on-one-card-slower-on-another-waves/6adf75af7a752cdd73090b42)

### [My GPU fix was 13% faster on one card and 81% slower on another](https://amankarki.hashnode.dev/gpu-fix-faster-on-one-card-slower-on-another-waves?utm_source=hashnode&utm_medium=feed)

A vLLM int8 GEMM change, a reviewer with a smaller Ada GPU, and the one idea that explained it: waves

7 min read · [amankarki.hashnode.dev](https://amankarki.hashnode.dev/gpu-fix-faster-on-one-card-slower-on-another-waves?utm_source=hashnode&utm_medium=feed)
