Chris Fregly

Arguably the most disturbing prop here @nvidia GTC

@swyx @NVIDIAGTC @NVIDIAAIDev Jensen’s fave slide, apparently!

caption this

25 google slides that summarize the NVIDIA GTC 2026 Keynote: docs.google.com/presentation/d… Love the country-music and OpenClaw themes!! 🤠🦞 @NVIDIAAIDev @NVIDIAAI @nvidia

Hot take from Day 0 at NVIDIA GTC: NVLink (NVIDIA proprietary high-speed GPU-to-GPU interconnect) actually uses InfiniBand under the hood!? 🤯

Was waiting for this… infinity.inc/case-studies/q…

We're celebrating our 100th lecture with @SemiAnalysis_ tomorrow at 9am PST and we'll be talking about arguably the most important OSS benchmark suite out today InferenceX Time flies, thank you to everyone who stuck around for this long 🥲

Just wrapped a conversation on performance optimization and software/hardware co-design across PyTorch, CUDA, and NVIDIA GPUs with @cfregly.

Github Repo: github.com/cfregly/ai-per… O'Reilly Book: amazon.com/Systems-Perfor… Course: academy.nebius.com/ai-engineering… YouTube link: @AIPerformanceEngineering" target="_blank" rel="nofollow noopener">youtube.com/@AIPerformance…

Join us now for the a great set of talks on hashtag#OpenClaw, NVIDIA's NVFP4 numeric format, and a new course on AI Performance Engineering with Riccardo Mereu from Verda and Stan from Nebius! meetup.com/ai-performance…

@steipete @OpenAI @openclaw Hell yeah

I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.🦞 steipete.me/posts/2026/ope…

@DennisHannusch @OpenAIDevs @DennisHannusch Shhh... let's keep Codex our little secret. Let everyone else flood the Anthropic servers with barbaric Ralph Wiggum loops to build simple NextJS apps. Meanwhile, we'll get real shit done with Codex. 👊 @embirico @romainhuet @sama

I started daily driving Codex with gpt-5.3-codex this week.. it's reaaally good. I've gotten used to complex workflows and context management, but Codex just does what I ask. I keep expecting quality to drop deep into a session, but it doesn't. @OpenAIDevs ya'll cooked!

I love when Codex/Claude find my own book and github repo useful for CUDA kernel optimizations... makes the 1-year effort (and $4k in Starbucks lattes ☕️) worth it!

@charles_irl 100% agree. Please move to SF and increase my property value. Leave your family if you have to. You’ll can easily replace them with more-productive and happier AI agents.

you really gotta be in-person in SF to truly join the AI gold rush you know what they say: out of site, out of mine

@typedfemale The same people who chose to optimize for performance over developer convenience, unfortunately. I agree that it's super annoying!! Btw, would love to have you on my monthly AI Performance Meetup if you're interested in speaking? meetup.com/ai-performance…

what if we held whoever made CUDA graphs require fixed memory addresses accountable? what if we used the law to legally detain them, marched them into a courtroom and let a jury of their peers decide their fate?

Can you at least let me enjoy "0%" for the first 60 seconds of my monthly-billing cycle?!

I literally just watched my Cursor/Opus 4.6 (Max) "percentage API usage" increase from 0% to 1% simply by clicking the drop-down to enable Opus 4.6 Max! No tokens were exchanged... none that originated from me, anyway. wtf @cursor_ai ?