Submissions from fergusfinn.com

		What happens when you run a CUDA kernel? (fergusfinn.com)
		294 points by mezark 3 months ago \| past \| 32 comments
		Adaptive speculative decoding: picking draft lengths at runtime (fergusfinn.com)
		5 points by hasheddan 3 months ago \| past
		InfiniBand, RoCE, and All That (fergusfinn.com)
		5 points by hasheddan 3 months ago \| past
		InfiniBand, RoCE, and All That (fergusfinn.com)
		4 points by kjeetgill 3 months ago \| past
		InfiniBand, RoCE, and All That (fergusfinn.com)
		5 points by kkm 3 months ago \| past
		UCCL-EP: DeepEP-style expert parallelism on any NIC, no GPU-initiated comms (fergusfinn.com)
		9 points by kkm 3 months ago \| past
		Anatomy of a high-performance EP kernel (fergusfinn.com)
		16 points by kkm 3 months ago \| past \| 1 comment
		The Economics of Speculative Decoding (fergusfinn.com)
		30 points by kkm 3 months ago \| past \| 6 comments
		Speculative KV coding: losslessly compressing KV cache by up to ~4× (fergusfinn.com)
		155 points by kkm 3 months ago \| past \| 48 comments
		70x faster cold(ish) starts for SGLang (fergusfinn.com)
		1 point by kkm 3 months ago \| past
		Bringing Up DeepSeek-V4-Flash on AMD MI300X (fergusfinn.com)
		120 points by kkm 3 months ago \| past \| 25 comments
		Pushing memory bound CUDA kernels past the speed of light with data compression (fergusfinn.com)
		2 points by somnial 4 months ago \| past
		Speculative KV coding: ~4× losslessly compressed KV cache using a small model (fergusfinn.com)
		2 points by somnial 4 months ago \| past
		In search of wasted bits: how much information do LLM weights carry? (fergusfinn.com)
		1 point by gmays 4 months ago \| past
		Redundant Information in LLM Weights (fergusfinn.com)
		5 points by mezark 4 months ago \| past
		Tans: Precomputing RANS (fergusfinn.com)
		3 points by mezark 5 months ago \| past
		Also-RANS: Asymmetric Numeral Systems for Entropy Coding (fergusfinn.com)
		25 points by mezark 5 months ago \| past
		70x faster cold(ish) starts for SGLang (fergusfinn.com)
		1 point by somnial 5 months ago \| past
		70x faster cold(ish) starts for SGLang (fergusfinn.com)
		4 points by mezark 5 months ago \| past
		Parallel Primitives for Multi-Agent Workflows (fergusfinn.com)
		1 point by mezark 8 months ago \| past
		LLM powered data structures: A lock-free binary search tree (fergusfinn.com)
		1 point by somnial 8 months ago \| past
		Parallel Primitives for Multi-Agent Workflows (fergusfinn.com)
		1 point by somnial 8 months ago \| past
		Scheduling in LLM Inference (fergusfinn.com)
		1 point by somnial 10 months ago \| past
		How fast can an LLM go? (fergusfinn.com)
		2 points by kkm 10 months ago \| past
		How fast can an LLM go? (fergusfinn.com)
		2 points by gmays 10 months ago \| past
		How fast can an LLM go? (fergusfinn.com)
		2 points by somnial 11 months ago \| past