• Tacit knowledge/mechanical sympathy

• “Welcome to the Jungle” (CPU design and Moore’s law; cf. “How Many Computers In Your Computer?”)

• “There’s plenty of room at the Top: What will drive computer performance after Moore’s law?”, Leiserson et al 2020 (gains from End-To-End Principle-based systems design¹)

• “Scalability! But at what COST?”, McSherry et al 2015 (when laptops > clusters: the importance of good baselines; “Big Data” doesn’t fit in a laptop); “Command-line Tools can be 235× Faster than your Hadoop Cluster”

• “What it takes to run Stack Overflow”, Nick Craver; “We stand to save $7m over 5 years from our cloud exit”, David Heinemeier Hansson; “Production Twitter on One Machine? 100Gbps NICs and NVMe are fast”

• “Anatomy of a hack: How crackers ransack passwords like qeadzcwrsfxv1331; For Ars, 3 crackers have at 16,000+ hashed passcodes—with 90% success”; “Computing Adler32 Checksums at 41 GB/s”

• “The Science of Brute Force”

• “Bracket pair colorization 10,000× faster”

• “Parsing Gigabytes of JSON per Second”, Langdale & Lemire2019

• 59 GB/s FizzBuzz; “Modern storage is plenty fast. It is the APIs that are bad.”; “10M IOPS, one physical core. #io_uring #linux”; “Achieving 11M IOPS & 66 GB/s IO on a Single ThreadRipper Workstation”

• “The C10K problem”, Kegel 2001_23ya; “Serving Netflix Video at 50 gigabyte/s on FreeBSD”, Gallatin2021/“Serving Netflix Video Traffic at 100 gigabyte/s and Beyond”, Gallatin2022

• “Multicast and the Markets” (network optimization & mechanical sympathy in high-frequency trading)

• “True Stories of Algorithmic Improvement”

• Ultra-low bit-rate audio for tiny podcasts with Codec2 and WaveNet decoders (1 hour = 1MB)

• “The computers are fast, but you don’t know it” (replacing slow pandas with 6,000× faster parallelized C++)

• “Scaling SQLite to 4M QPS on a Single Server (EC2 vs Bare Metal)”, Expensify; “50% faster than 3.7.17” (“… We have achieved this by incorporating hundreds of micro-optimizations. Each micro-optimization might improve the performance by as little as 0.05%.”)

• “Optimization story—quantum mechanics simulation speedup”; “14,000× Speedup a.k.a. Computer Science For the Win”

• “Cutting The Pipe: Achieving Sub-Second Iteration Times”, Frykholm 2012_12ya (optimizing game development)

• “The Billion Row Challenge (1BRC): Step-by-step in Java from 71s to 1.7s” (and 4s in Go & 0.5s in C)

• “GCC Compiler vs Human—119× Faster Assembly 💻🆚🧑‍💻”; “Python, C, Assembly—2,500× Faster Cosine Similarity 📐”; “Summing ASCII encoded integers on Haswell at almost the speed of memcpy” (“320× faster than the following naive C++ solution”)

• Demoscene:

  • “8088 Domination”: “how I created 8088 Domination, which is a program that displays fairly decent full-motion video on a 1981_43ya IBM PC”/part 2

  • “A Mind Is Born” (256 bytes; for comparison, this line of Markdown linking the demo is 194 bytes long, and the video several million bytes)

• Casey Muratori: refterm v2, Witness pathing (summary)

• Algorithmic Progress:

  • “Algorithmic Progress in 6 Domains”, Grace2013

  • “A Time Leap Challenge for SAT Solving”, Fichte et al 2020 (hardware overhang: ~2.5SAT solving since 2000_24ya, about equally due to software & hardware gains, although slightly more software—new software on old hardware beats old on new.)

  • 2019 Unlocking of the LCS35 Challenge

  • “Measuring hardware overhang”, hippke (“with today’s algorithms, computers would have beat the world chess champion already in 1994_30ya on a contemporary desk computer”)

• Memory Locality:

  • “What Every Programmer Should Know About Memory”, Drepper2007

  • “You’re Doing It Wrong: Think you’ve mastered the art of server performance? Think again.”, Poul-Henning Kamp 2010_14ya (on using cache-oblivious B-heaps to optimize Varnish performance 10×)

  • “The LMAX Architecture”, Fowler 2011_13ya (TigerBeetle)

  • “5000× faster CRDTs: An Adventure in Optimization”

  • “In-place Superscalar Samplesort (IPS⁴o): Engineering In-place (Shared-memory) Sorting Algorithms”, Axtmann et al 2020; “Vectorized and performance-portable Quicksort”, Blacher et al 2022

  • “Fast approximate string matching with large edit distances in Big Data (2015)”

  • “ParPaRaw: Massively Parallel Parsing of [CSV] Delimiter-Separated Raw Data”, Stehl & Jacobsen2019

  • “Scaling our Spreadsheet Engine from Thousands to Billions of Cells: From Maps To Arrays”, Lukas Köbis2022

  • “Persistent RNNs: Stashing Recurrent Weights On-Chip”, Diamos et al 2016; “FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness”, Dao et al 2022

  • Algorithms for Modern Hardware

  • Solving Problem 14 with Fast Dynamic Programming

• DL/Deep Reinforcement Learning:

  • Training A3C to solve Atari Pong in <4 minutes on the ARCHER supercomputer through brute parallelism

  • “megastep helps you build 1-million FPS reinforcement learning environments on a single GPU”, Jones

  • “The Mathematics of 2048: Optimal Play with Markov Decision Processes”

  • Bruteforcing Breakout (optimal play by depth-first search of an approximate MDP in an optimized C++ simulator for 6 CPU-years; crazy gameplay—like chess endgame tables, a glimpse of superintelligence)

  • “Sample Factory: Egocentric 3D Control from Pixels at 100,000 FPS with Asynchronous Reinforcement Learning”, Petrenko et al 2020; “Megaverse: Simulating Embodied Agents at One Million Experiences per Second”, Petrenko et al 2021; “Brax: A Differentiable Physics Engine for Large Scale Rigid Body Simulation”, Freeman et al 2021; “Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning”, Makoviychuk et al 2021 (Rudin et al 2021); WarpDrive

  • “Large Batch Simulation for Deep Reinforcement Learning”, Shacklett et al 2021

  • “Scaling Scaling Laws with Board Games”, Jones2021

  • “Mastering Real-Time Strategy Games with Deep Reinforcement Learning: Mere Mortal Edition”, Winter

  • “Scaling down Deep Learning”, Greydanus2020

  • “Podracer architectures for scalable Reinforcement Learning”, Hessel et al 2021 (highly-efficient TPU pod use: eg. solving Pong in <1min at 43 million FPS on a TPU-2048); “Q-DAX: Accelerated Quality-Diversity for Robotics through Massive Parallelism”, Lim et al 2022; “EvoJAX: Hardware-Accelerated Neuroevolution”, Tang et al 2022; “JaxMARL: Multi-Agent RL Environments in JAX”, Rutherford et al 2023

  • “Training the Ant simulating at 1 million steps per second on CUDA (NVIDIA RTX 2080), using Tiny Differentiable Simulator and C++ Augmented Random Search.” (Erwin Coumans2021)

  • “This Snake environment simulates 14M snakes and trains 1M snake steps per second.”

  • “TextWorldExpress: Simulating Text Games at One Million Steps Per Second”, Jansen & Côté2022

  • “Getting the World Record in HATETRIS”, Dave & Filipe2022

  • “NAVIX: Scaling MiniGrid Environments with JAX”

• Latency/UI:

  • “It’s the Latency, Stupid.”

  • “Transatlantic ping faster than sending a pixel to the screen?”

  • “Computer latency: 1977^–₄₀2017”

  • “Keyboard Latency”

  • “Terminal Latency”; improving GNOME terminal

  • “Local-first software: You own your data, in spite of the cloud”/Kleppmann et al 2019

  • “Web Bloat”²

  • “Project Starline: A high-fidelity telepresence system”, Lawrence et al 2021

  • “Use GPT-3 incorrectly: reduce costs 40× and increase speed by 5×”, Pullen2023

  • HFT: “The accidental HFT firm”

  • Shenzhen

• Organizations:

  • WhatsApp (2013_11ya: “50 staff members” for 200m users; 2015: 50 “engineers”, 900m users)

  • Craigslist (2017: ~50 employees)

  • Instagram (2012_12ya FB acquisition: 13 employees; architecture)

  • YouTube (2006_18ya Google acquisition: ~10 employees)

  • Reddit (2011_13ya: 4 engineers; 2021: ~700 total staff, >52m users)

  • Wikimedia Foundation (“Internet hosting” in 2021: $2.4m)

  • Twitter: Musk acquisition cut up to 90% of employees+contractors; widespread confident assertions by ex-Twitter staff that the site would suffer crashes (possibly terminal crashes which could not be recovered from), within a month or two, turned out to be false.

  • See Also: “Some examples of people quickly accomplishing ambitious things together.”

---

1. AI scaling can continue even if semiconductors do not, particularly with self-optimizing stacks; John Carmack notes, apropos of Seymour Cray, that “Hyperscale data centers and even national supercomputers are loosely coupled things today, but if challenges demanded it, there is a world with a zetta[flops] scale, tightly integrated, low latency matrix dissipating a gigawatt in a swimming pool of circulating fluorinert.”↩︎

2. A relevant example is Amazon’s “best practices” for hosting a WordPress blog:

   

   ↩︎

Backlinks

[Backlinks (what links here)]

[ Send Anonymous Feedback ]

[adblock public service announcement]