CIQ Blog: What's New in Enterprise Linux & AI Infrastructure
We're a team of scientists, Linux geeks, technologists, pilots, designers and more with a mission to make infrastructure something you never have to think about.
Tokens per second won't tell you what an answer costs on a Strix Halo
On a Strix Halo, a model 13x worse per token can be cheaper per answer. Per-answer cost, speed and active parameters across 27 local models
CIQ Stratum Kubernetes Engine: Kubernetes for the AI Era
CIQ Stratum Kubernetes Engine (SKE) installs, joins, and upgrades Kubernetes from one signed file on Enterprise Linux 9 or later. CIQ Stratum Pro adds fleet management.
AI-found CVEs: what PHP2Shell means for your triage
AI turned a parked 2024 PHP bug into remote code execution. Here is what AI-found CVEs change in triage, and why the fix is now the work that counts
One flag took Gemma 4 from 52% to 100% on a Strix Halo
Reasoning off took Gemma 4 from 52% to 100% on short tasks at one-eighth the cost. Plus a prompt change worth 41 points, and models that ignore both
From Samba to post-quantum cryptography: Jeremy Allison’s open source journey
Meet CIQ Distinguished Engineer Jeremy Allison and explore his journey from Samba to FIPS compliance, open source security, and post-quantum cryptography.
Patch PHP2Shell now and prepare for the next AI-found CVE
Six months ago, I wrote that vulnerabilities were going to come in waves. I undersold it. In April and May, Copy Fail, Dirty Frag, Fragnesia and CIFSwitch hit the Linux kernel. Researchers have kept…
Diagnose failed jobs and re-prioritize queued work in place with Fuzzball
Fuzzball names the failed rank in multi-node jobs, re-prioritizes queued work in place, and puts compute policy grants and deletion in the CLI.
Monthly newsletter
No spam. Just the latest releases and tips, interesting articles, and exclusive interviews in your inbox every month.
Read about our privacy policy.
Get more AI speed from your DGX Spark with RLC Pro AI
See how RLC Pro AI did on three Spark Arena tests on a NVIDIA DGX Spark, then set up the same host and rerun every test yourself, step by step.
What a correct answer costs on a Strix Halo depends on how hard the question is
I measured 27 local LLMs on a Strix Halo with a power meter. The cheapest model for easy questions costs 37x more once questions get hard.
Tenable Nessus scans now reflect the CVE fixes CIQ delivers to RLC Pro
New Tenable plugins read CIQ advisories, so Nessus scans of RLC Pro hosts reflect the CVE fixes CIQ delivers, including on LTS releases
Removing roadblocks to discovery: David Godlove’s path from neuroscience to Fuzzball
Meet David Godlove and see how his path through neuroscience, NIH HPC, and Singularity/Apptainer helps CIQ make Fuzzball more accessible to researchers.