Feed aggregator

Article URL: https://www.theguardian.com/world/2026/jul/26/typhoon-noul-china-hong-kong

Comments URL: https://news.ycombinator.com/item?id=49057889

Points: 1

# Comments: 0

Article URL: https://www.science.org/content/blog-post/assault-science-funding-continues

Comments URL: https://news.ycombinator.com/item?id=49057887

Points: 1

# Comments: 0

Article URL: https://www.cnn.com/2026/07/25/economy/15-usd-pint-ice-cream-economy

Comments URL: https://news.ycombinator.com/item?id=49057880

Points: 1

# Comments: 0

Article URL: https://www.rickmanelius.com/p/the-new-ai-superpowers-focus-and

Comments URL: https://news.ycombinator.com/item?id=49057877

Points: 1

# Comments: 0

Article URL: https://www.france24.com/en/europe/20260726-romanian-jet-shoots-down-drone-entering-country-airspace-for-third-day-in-a-row

Comments URL: https://news.ycombinator.com/item?id=49057820

Points: 1

# Comments: 0

Article URL: https://www.ft.com/content/822628c5-4f9c-47db-bd27-f3ad7f841700

Comments URL: https://news.ycombinator.com/item?id=49057799

Points: 2

# Comments: 0

Article URL: https://blog.atimin.dev/handling-file-descriptors/

Comments URL: https://news.ycombinator.com/item?id=49057791

Points: 1

# Comments: 0

Article URL: https://underdogfounders.substack.com/p/the-story-of-building-drizzle-orm

Comments URL: https://news.ycombinator.com/item?id=49057781

Points: 1

# Comments: 0

Article URL: https://zenodo.org/records/21541601

Comments URL: https://news.ycombinator.com/item?id=49057774

Points: 1

# Comments: 0

Hey HN, we’re the developers of OpenLake, an open source storage engine for offloading LLM KV caches from GPU memory into a shared tier of RAM and NVMe. We built OpenLake because KV caches are outgrowing GPU memory.

A single 256K token conversation on Gemma 4 31B produces approximately 43GB of KV state, more than half the memory of an 80GB H100. The problem becomes even harder across a cluster: a prefix cached on one GPU host is unavailable when the next request lands on a different GPU, forcing the new GPU to repeat work the fleet has already completed.

Once the KV cache is offloaded, network bandwidth becomes a major constraint on read latency. To move less data across the wire, we built deferred materialization: a custom CUDA kernel that losslessly compresses KV blocks before they leave GPU memory and decompresses them on the GPU after retrieval. In our tests, this achieved:

- 1.72× lossless KV compression. - Approximately 600GB/s decompression throughput on an H100 - 80GB/s of effective KV throughput over a physical 50GB/s link

At 128K context, retrieving cached KV reduces TTFT from 44 seconds to 0.6 seconds, a 66× improvement. Across the complete workload, GPU time reduces from 1,169 seconds to 606 seconds, saving 48.2% of GPU cost.

OpenLake is written in Rust and uses io_uring with one pinned runtime per physical core. We provide connectors for vLLM and SGLang so the cache can be enabled without modifying the inference engine itself.

I would love to hear how others are handling KV reuse across GPU hosts, especially for long contexts, and get to know your thoughts.

Thanks!

GitHub: https://github.com/openlake-project/openlake

Here is our blog: https://cloud.theopenlake.com/blog/taming-the-beast-managing...

Comments URL: https://news.ycombinator.com/item?id=49057767

Points: 10

# Comments: 0

Timeline with depth.

Comments URL: https://news.ycombinator.com/item?id=49057749

Points: 2

# Comments: 0

Article URL: https://www.nextplatform.com/compute/2026/07/24/the-money-amd-is-chasing-with-its-rackscale-ai-system-roadmaps/5278510

Comments URL: https://news.ycombinator.com/item?id=49057722

Points: 1

# Comments: 0

Article URL: https://www.youtube.com/watch?v=kOO31qFmi9A

Comments URL: https://news.ycombinator.com/item?id=49057716

Points: 2

# Comments: 0

Article URL: https://old.reddit.com/r/ClaudeAI/comments/1v6fiyj/you_can_view_a_lot_of_shared_conversations_via/

Comments URL: https://news.ycombinator.com/item?id=49057707

Points: 1

# Comments: 0

Article URL: https://www.ft.com/content/96a33881-27fd-42cf-8cff-4cbc87fc835f

Comments URL: https://news.ycombinator.com/item?id=49057704

Points: 1

# Comments: 0

Article URL: https://github.com/grigio/nixos-noctalia-labwc-flake

Comments URL: https://news.ycombinator.com/item?id=49057688

Points: 3

# Comments: 0

Article URL: https://nezutero.dev/goodbye-microslop-finally-moved-to-codeberg/

Comments URL: https://news.ycombinator.com/item?id=49057686

Points: 1

# Comments: 0

Article URL: https://www.neelocean.com/kahnemans-cabs/

Comments URL: https://news.ycombinator.com/item?id=49057674

Points: 1

# Comments: 0

Article URL: https://www.wsj.com/business/logistics/one-of-the-busiest-shipping-ports-wants-to-go-nuclear-c03f03d2

Comments URL: https://news.ycombinator.com/item?id=49057344

Points: 2

# Comments: 0