Top all-time · #7160
Doubleword
Company blog · blog.doubleword.aiaisystems
Doubleword is the company blog at blog.doubleword.ai, writing about ai and systems. Since 2026, 11 different Hacker News readers have linked it 23 times, most often “What happens when a GPU reads memory?”. It is still writing: the last post was on 2026-09-22, and you can follow it by feed.
Too few sampled comments (3) to say how often HN recommends it.?
Why HN links it
HN commenters link posts on running large models on AMD MI300X GPUs and on fast SGLang cold starts for multiplexing models on shared GPUs. Comments come mostly from the company, which also mentions its API pricing.
Summarised by Claude from a sample of 3 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Notes on building AI systems
Most-linked posts?
- What happens when a GPU reads memory?3 people · 3 mentions · first linked 2026-08-13
- NVLink, NVSwitch, and All That2 people · 2 mentions · first linked 2026-07-22
- You Could Have Come Up with Kimi Delta Attention2 people · 2 mentions · first linked 2026-07-27
- /fast-sglang-starts2 people · 2 mentions · first linked 2026-06-03
- Moe inference optimizations: 15% lower expert load by request reordering2 people · 2 mentions · first linked 2026-05-20
- What happens when a GPU writes memory2 people · 2 mentions · first linked 2026-08-31
- Unlocking parallel test-time scaling for long-horizon agents2 people · 2 mentions · first linked 2026-09-23
- The gap between open weights LLMs and closed source LLMs2 people · 2 mentions · first linked 2026-06-22
What Hacker News says?
nice! i think the higher HBM on Mi300x is really useful for this kind of thing we did some work on this for 2xMi300x (kindly referenced in the readme) https://blog.doubleword.ai/deepseek-v4-flash-mi300x . https://hotaisle.xyz/quick-start hotaisle is great for getting Mi300x to experiment with
Comment linking /deepseek-v4-flash-mi300x
on “DeepSeek V4 Flash on a Single AMD MI300X”Submitted You Could Have Come Up with Kimi Delta Attention as a story · 87 points
HN discussionhttps://blog.doubleword.ai/fast-sglang-starts yeah, this is part of the idea. If you get cold starts down to seconds or subseconds, then you can run many models multiplexed on the same GPUs
Comment linking /fast-sglang-starts
on “Reverse-engineering Nvidia's CUDA-checkpoint for faster cold starts”Submitted What happens when a GPU writes memory as a story · 76 points
HN discussionHi! Co-founder of Doubleword here - we've hugely increased the number of models that we offer (partly thanks to work that we've done on hotswapping https://blog.doubleword.ai/fast-sglang-starts . We're kind of known for our low prices - our prices (our main usage is for our high throughput API - the async tier) is significantly below average openrouter prices - but cached prices is coming soon which will lower them…
Comment linking /fast-sglang-starts
on “Bringing Up DeepSeek-V4-Flash on AMD MI300X”Submitted The gap between open weights LLMs and closed source LLMs as a story · 26 points
HN discussionSubmitted What happens when a GPU reads memory? as a story · 6 points
HN discussionSubmitted What happens when a GPU writes memory as a story · 5 points
HN discussionSubmitted What happens when a GPU reads memory as a story · 4 points
HN discussionSubmitted The case for disaggregated LLM serving as a story · 4 points
HN discussionSubmitted On-the-fly snapshot compression for elastic inference at scale as a story · 4 points
HN discussionSubmitted Unlocking parallel test-time scaling for long-horizon agents as a story · 3 points
HN discussionSubmitted Reverse-engineering Nvidia's CUDA-checkpoint for faster cold starts as a story · 3 points
HN discussionSubmitted Width vs. Depth: Speculating on the Margin as a story · 3 points
HN discussionSubmitted Unlocking parallel test-time scaling for long-horizon agents as a story · 1 points
HN discussionSubmitted What happens when a GPU reads memory as a story · 1 points
HN discussionSubmitted Guide to the Kimi DeltaNet Family of linear attention as a story · 1 points
HN discussionSubmitted NVLink, NVSwitch, and All That as a story · 1 points
HN discussionSubmitted NVLink, NVSwitch, and All That as a story · 1 points
HN discussionSubmitted Prediction: A Frontier open-source LLM Will Be Released On 3rd December 2026 as a story · 1 points
HN discussionSubmitted MoE expert co-activations: Reordering inputs yields easy throughput gains as a story · 1 points
HN discussionSubmitted Moe inference optimizations: 15% lower expert load by request reordering as a story · 1 points
HN discussionSubmitted QueueSpec – drafting speculation tokens while a request queues as a story · 1 points
HN discussion
