korrents

On the map

Tap a claim on the ring to put it at the centre.

← For GPU performance, flops do not matter so much as the whole memory…

17 connected korrents · 17 moments from 11 Jun 2021 to 6 Nov 2026.

Everything filed under software performance software performance Everything filed under developer tools developer tools Everything filed under semiconductors semiconductors Everything filed under benchmarks benchmarks Everything filed under simplicity simplicity Same subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subject Read this korrent: For GPU performance, flops do not matter so much as the whole memory access pattern does, and torch.compile cannot find every optimization. For GPU performance, flops do not matterso much as the whole memory access patterndoes, and torch.compile cannot find everyoptimization. Last stated 2 years ago 9 Jun 2024 AK Andrej Karpathy — holds since 2024-06-09 — tap for who they are Same subject: Hand-optimizing hot spots barely matters any more, because the libraries are already fast and CPUs are extremely good at running bad code quickly; what is left is architecture. — tap to centre the map on it Hand-optimizing hot spots barelymatters any more, because thelibraries are already fast and CPUsare extremely good at running badcode quickly; what is left isarchitecture. Last stated 3 months ago 26 Aug 2026 CM Casey Muratori — holds since 2026-08-26 — tap for who they are Same subject: Criticism that a simpler, scaled-up architecture underperforms a heavily hand-optimized baseline actually demonstrates why that architecture-scaling research matters. — tap to centre the map on it Criticism that a simpler, scaled-uparchitecture underperforms a heavilyhand-optimized baseline actuallydemonstrates why thatarchitecture-scaling researchmatters. Last stated 5 years ago 11 Jun 2021 GW Gwern — holds since 2021-06-11 — tap for who they are Same subject: A tool's speed comes primarily from its design decisions rather than the programming language it is written in. — tap to centre the map on it A tool's speed comes primarily fromits design decisions rather than theprogramming language it is writtenin. Last stated 11 months ago 29 Dec 2025 AP Aaron Patterson — holds since 2025-12-29 — tap for who they are Same subject: Falling back to older 7-nanometre chips would not rescue AI compute, because the real gap between chip generations is twentyfold, not the threefold the flops suggest. — tap to centre the map on it Falling back to older 7-nanometrechips would not rescue AI compute,because the real gap between chipgenerations is twentyfold, not thethreefold the flops suggest. Last stated 8 months ago 13 Mar 2026 DP Dylan Patel — holds since 2026-03-13 — tap for who they are Same subject: Nvidia should build only what would not otherwise exist — twenty years of CUDA at a loss is the test, and the world already has plenty of clouds. — tap to centre the map on it Nvidia should build only what wouldnot otherwise exist — twenty yearsof CUDA at a loss is the test, andthe world already has plenty ofclouds. Last stated 7 months ago 15 Apr 2026 JH Jensen Huang — holds since 2026-04-15 — tap for who they are Same subject: PyTorch DDP's recommended no_sync pattern for gradient accumulation is super ugly, since it makes you copy-paste code into a context manager. — tap to centre the map on it PyTorch DDP's recommended no_syncpattern for gradient accumulation issuper ugly, since it makes youcopy-paste code into a contextmanager. Last stated 2 years ago 9 Jun 2024 AK Andrej Karpathy — holds since 2024-06-09 — tap for who they are Same subject: Graphics chips have far more room for future performance gains than CPUs, which are nearing physical limits. — tap to centre the map on it Graphics chips have far more roomfor future performance gains thanCPUs, which are nearing physicallimits. Last stated a year ago 8 Jun 2025 DC David Crawshaw — holds since 2025-06-08 — tap for who they are Same subject: A speed difference between languages that share the LLVM backend measures the benchmark author, not the languages. — tap to centre the map on it A speed difference between languagesthat share the LLVM backend measuresthe benchmark author, not thelanguages. Last stated 2 years ago 22 Mar 2025 TH ThePrimeagen — holds since 2025-03-22 — tap for who they are Same subject: A coding agent should probably open in your project folder rather than in the process's parent directory. — tap to centre the map on it A coding agent should probably openin your project folder rather thanin the process's parent directory. Last stated 2 weeks ago 6 Nov 2026 SH Scott Hanselman — holds since 2026-11-06 — tap for who they are Same subject: A coding AI that integrates only with its own vendor's tools will be very limited compared with one that draws on all the tools you use. — tap to centre the map on it A coding AI that integrates onlywith its own vendor's tools will bevery limited compared with one thatdraws on all the tools you use. Last stated 3 years ago 16 Mar 2024 QS Quinn Slack — holds since 2024-03-16 — tap for who they are Same subject: A developer using AI coding tools remains responsible for reviewing the code they ship. — tap to centre the map on it A developer using AI coding toolsremains responsible for reviewingthe code they ship. Last stated 8 months ago 10 Mar 2026 JS Jon Seager — holds since 2026-03-10 — tap for who they are Same subject: A software engineer who is not using a tool like Cursor is working at half productivity or worse. — tap to centre the map on it A software engineer who is not usinga tool like Cursor is working athalf productivity or worse. Last stated a year ago 18 Aug 2025 BT Bret Taylor — holds since 2025-08-18 — tap for who they are Same subject: AI coding agents work best with the same general-purpose tools built for humans, such as git and file utilities, rather than agent-specific interfaces. — tap to centre the map on it AI coding agents work best with thesame general-purpose tools built forhumans, such as git and fileutilities, rather thanagent-specific interfaces. Last stated a year ago 30 Sept 2025 HR Harper Reed — holds since 2025-09-30 — tap for who they are Same subject: AI coding assistants remove repetitive work without replacing human judgment. — tap to centre the map on it AI coding assistants removerepetitive work without replacinghuman judgment. Last stated 9 months ago 17 Feb 2026 EB Emanuele Bartolesi — holds since 2026-02-17 — tap for who they are Same subject: AI is a pair programmer rather than something to delegate all control to. — tap to centre the map on it AI is a pair programmer rather thansomething to delegate all controlto. Last stated a year ago 25 Aug 2025 LR Lee Robinson — holds since 2025-08-25 — tap for who they are Same subject: CMake tooling is missing a linter that could statically catch most CMakeLists.txt mistakes. — tap to centre the map on it CMake tooling is missing a linterthat could statically catch mostCMakeLists.txt mistakes. Last stated 2 months ago 20 Sept 2026 CW Chris Wellons — holds since 2026-09-20 — tap for who they are Same subject: Comparing code coverage between passing and failing test runs is a cheap, effective way to localize a bug quickly. — tap to centre the map on it Comparing code coverage betweenpassing and failing test runs is acheap, effective way to localize abug quickly. Last stated 2 years ago 25 Apr 2025 RC Russ Cox — holds since 2025-04-25 — tap for who they are
same subject or similar wordinga cloud: claims about one subject, named for itbar: when it was last stated, on a scale from 2015 to today — full is todaya face: someone who holds the claim — tap it for who they are

At the centre For GPU performance, flops do not matter so much as the whole memory access pattern does, and torch.compile cannot find every optimization. Last stated 9 Jun 2024 · 2 years ago Holds Andrej Karpathy Read this korrent →