Integrated photonic 3D tensor processing engine nature.com/articles/s4137…

The global economy is held up by SF lying to NYC about how fast AI is growing and NYC lying to Abu Dhabi about how safe private credit is and Abu Dhabi lying about how safe their oil is

worrying about climate change when C++ struct padding is silently wasting 40% of the world's ram is wild.

Overheard at dinner last night: “If Gulf money is off the table who is going to fund datacenter buildout?”

Why do CPUs have multiple cache levels? by Fabian Giesen. One of the best explanations, and my periodic recommendation to the students. fgiesen.wordpress.com/2016/08/07/why…

Entirely unsurprised that a significant number of people who move to Dubai turn out to be the sort of people who abandon pets when they become a minor inconvenience.

Good memories with TurboC. #nostalgic

Is there a GPU equivalent to what RISC-V represents for CPU ISAs? I mean an open/standardized GPU ISA that hardware vendors could implement independently.

The best argument for Rust in 2026 is not memory safety or performance. It is that AI writes better Rust than it writes C++. The compiler feedback loop is so tight that models self-correct in real time. Every error message is a free training signal. Rust was accidentally designed

I was telling my friend Israel (yeah his name is Israel) "a lot of these FPGAs are going to become ridiculously cheap when they can't figure out how to use them in the data centers." I was right.

Mac Mini + eGPU. Both NVIDIA and AMD supported.

No way. Keir isn’t the best, but definitely better than Rishi.

𝗞-𝗺𝗲𝗮𝗻𝘀 𝗶𝘀 𝘀𝗶𝗺𝗽𝗹𝗲. 𝗠𝗮𝗸𝗶𝗻𝗴 𝗶𝘁 𝗳𝗮𝘀𝘁 𝗼𝗻 𝗚𝗣𝗨𝘀 𝗶𝘀𝗻’𝘁. That’s why we built Flash-KMeans — an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly,

GPUs are slow at AI inference because they hit the memory wall. Cerebras pioneered the SRAM based AI accelerator because GPUs were memory bandwidth constrained.   Let me explain.   There are two types of memory. Memory that can store a lot, but is slow. And memory that is fast,

You can also gain GPU acceleration by treating Soft-KMeans as cross attention! 1. initialize K random vectors in range of values of data, our cluster centers will act as queries 2. let data serve as keys/values 3. compute attention, which is just weighted sum of our data. 4.

In C and C++, the order you write your variables can silently bloat your software's size. A struct with a char, then an int, then another char takes up more physical RAM than an int followed by two chars. The compiler secretly injects invisible, wasted padding bytes just to keep

Watched this 3 days ago, great talk. Almost all optimizations were familiar for a perf guy, but the key takeaway still hits "C++ can reach assembly-class performance for matrix multiplication" (with a free bf16 support)

Super Trouper tausend Watt verschwenden Und die Show fängt an Mikros auf, und dann: Beifall für die Number One Ich seh in Super-Trouper-Lichter, die mich blenden Doch was ich auch tu Ich denk immerzu: Da unten irgendwo bist du“ youtu.be/3MbY2c508_s

Studying engineering at Tsinghua University right now must feel really GOATED. Like you are in the midst of a huge power shift and whatever you study will probably be applied into era defining stuff