Bryce, the CUDA Colonel

Principal Engineer at @NVIDIA working on programming languages. @adspthepodcast co-host. C++ Library Evolution chair emeritus. Frequent flyer. Horology fan.

[Bryce, the CUDA Colonel @blelbach]

The latest revision of @INCITS/@isostandards COBOL comes out this year

The goals of COBOL sound normal today:

- Portable
- Freely available
- Designed by the community

In 1959 it was radical & unprecedented

It was also conceived of & led by women

This is the story of COBOL

[Bryce, the CUDA Colonel @blelbach]

"Nsight Compute invocations per day" is my favorite chart from my agent's report on tool research during last month's autoresearch runs.

[Bryce, the CUDA Colonel @blelbach]

@KayhanB21 Yeah so just give me one big switch to turn all that off

[Bryce, the CUDA Colonel @blelbach]

I tried out Hermes for autoresearch. Constantly hit turn limits, timeouts, etc.

Every harness needs to have a single setting that configures it for long-horizon unattended tasks.

[Bryce, the CUDA Colonel @blelbach]

@gpusteve Optimizing CUDA code.

[Bryce, the CUDA Colonel @blelbach]

I had to disable Claude Code Plan mode and AskUserQuestions - it seems like the only path to keep Opus 4.8 truly non-interactive.

[Bryce, the CUDA Colonel @blelbach]

@trentnelson So much easier to construct those when you just speech to text into it

[Bryce, the CUDA Colonel @blelbach]

My prompting style can best be described as "angry stream-of-consciousness chaos".

[Ryan Shrout @ryanshrout]

Microsoft just announced its most powerful Surface Laptop ever, and it does not run Intel or Qualcomm or AMD. It runs all @nvidia. The new @surface Laptop Ultra is built on the RTX Spark platform.

For a company that anchored Surface on Intel for years and then leaned on Qualcomm for the Copilot+ PC push, choosing NVIDIA for this new flagship category is not unexpected, but definitely a statement. This is the clearest signal yet of where the NVIDIA and Microsoft client alliance is heading.

It is also a truly new twist on the AI PC rather than one in name only. A Blackwell RTX GPU, up to 128GB of unified memory with full CUDA, and 1 petaflop of FP4 let it run models up to 120 billion parameters locally. That is paired with the brightest display Microsoft has shipped, a full set of connections and ports, and the largest haptic touchpad it has put on a Surface.

Microsoft said nothing about pricing, though, and a 128GB unified-memory machine in this memory market will not be cheap. All-day battery is claimed, but all-day means very different things depending on whether you are answering email or rendering, so I am holding judgment until we see real numbers. It is a Copilot+ PC with an NPU (by itself maybe a surprise), but the real AI horsepower here is the GPU. And the usual Windows on Arm skepticism applies until it is tested. (Though NVIDIA and Microsoft are making a renewed push that this is a solved problem.)

Because the Surface Laptop Ultra runs RTX Spark silicon, we already have a decent read on the compute, so the questions that remain are price, battery, and real-world creative and AI-builder performance on shipping units.

Would you put a 128GB NVIDIA-powered Surface on your desk or in your backpack, or is this a niche machine for a specific kind of maker? Curious where you land. I, for one, see the appeal of this new platform and how NVIDIA and Microsoft have started to talk differently about the Windows platform generally. Much to be seen later this fall!

[Satya Nadella @satyanadella]

Our goal is to deliver unmetered intelligence to every home and every desk with Windows.

NVIDIA RTX Spark marks a real breakthrough toward that vision.

Looking forward to sharing more with Jensen, who will be joining us live from Taiwan, at Build this week! https://t.co/O9ttCunAhG

[NVIDIA Newsroom @nvidianewsroom]

The age of AI needs a new kind of CPU.

Introducing NVIDIA Vera.

The CPU for agents, delivering 80% faster agentic task completion compared with x86 CPUs. Vera is built to power the CPU-intensive work behind modern AI factories, from agentic AI and reinforcement learning to data processing and orchestration.

[Bryce, the CUDA Colonel @blelbach]

/Now/ I can finally upgrade my laptop. I swore an oath my next would be all NVIDIA!

[NVIDIA RTX Spark @NVIDIARTXSpark]

Welcome to the NVIDIA RTX Spark channel.

A new superchip for the age of personal AI.

Don't worry, your favorite NVIDIA local AI content continues on right here, just with a new headliner.

Let's get started...

[Andrew Curran @AndrewCurran_]

Anthropic is not a coding company. It is an intelligence company that chose to focus on coding first. As Claude's intelligence scales, it will be applied to every endeavor where human intelligence is useful. Understanding this is the key to understanding the future.

[Nathan Lawrence 🌈 @NathanBLawrence]

If you are still interacting with permission prompts you don’t get it yet.

[sysls @systematicls]

I wrote this ~3 months ago, and since then,

1) Memory has been more or less fully integrated with the frontier models

2) Almost all features that made OpenClaw unique as a harness has been fully absorbed by the frontier models (e.g. schedules, loops, goals, memory, etc.)

3) New, vertical killing features and capabilities are being added every other week

--
All that being said, agentic engineering is still an incredibly high skill affair.

It is now obvious to me that there is a gulf of know-how and tacit knowledge between those that CAN remove humans-out-of-the-loop and actually produce a working product, and the rest of the world insisting that agents are still producing "slop".

[EuroPython @europython]

Join Bryce Adelstein Lelbach (@blelbach) at EuroPython for "GPU Programming in Pure Python". Talk: https://t.co/ThsPwo3nND

[Bryce, the CUDA Colonel @blelbach]

It is confounding to me that this is the case, but sadly I have found this to be true.

[ben hylak @benhylak]

LLM cost-tracking is really hard.

[Charles 🎉 Frye @charles_irl]

Added an easter egg to the GPU Glossary -- a hidden page on what is surely the most important implementation of the CUDA Tile programming model.

https://t.co/8XCPXSwwrs

[Bryce, the CUDA Colonel @blelbach]

Is there an alternative to @TripIt that has a ChatGPT integration or a native chat/agentic interface?

[Ashley Ha @ashleybchae]

Little known fact, i used to be an American Airlines flight attendant✈️

it’s never too late to change the trajectory of your life/career :)

[Ashley Ha @ashleybchae]

During covid, I studied computer programming + math 12 hours a day for almost a year.

Eventually got accepted into UC berkeley & learned enough computer science to land an internship at an EV company for machine learning.

Anything is possible if you work hard enough.