explore | Grilled Cheese

ИИ-агенты превзошли PyTorch: написание более быстрых CUDA-ядер

ИИ-агенты теперь могут писать CUDA-ядра, превосходящие PyTorch, но доказательство реального прироста скорости — более сложная задача. Я протестировал их на NVIDIA DGX Spark и обнаружил, чт…

Telegram ИИ Дайджест
#aiagents #cuda #pytorch

AI Agents Beat PyTorch: Writing Faster CUDA Kernels

AI agents can now write CUDA kernels that outperform PyTorch—but proving those speedups are real is the harder problem. I put them to the test on an NVIDIA DGX Spark and found that benchmark design mat…

Telegram AI Digest
#aiagents #cuda #pytorch

The true cost of breaking free from the CUDA monopoly is rarely measured in raw compute power, but rather is paid in software engineering velocity. #CUDA

ξ
cutile package stack
for #CUDA
in virtual environment
J:\env4xi
with #PyTorch
required by #CUDA
and #cutile

NVIDIA just changed the game: Native Rust on GPU kernels.

Compile Rust directly to PTX with cuda-oxide & cutile-rs. No more runtime data races! But heavy compilation demands true bare-metal power.

Read the deep dive on CUDA Rust & GPU servers:👇
www.migservers.com/blogs/nvidia...

#NVIDIA #CUDA #AI

https://youtu.be/fZvtqqHozwo
🇰🇷 Ks. dr hab. Piotr Maria Natanek podróżując po Korei Południowej opowiada o objawieniach Matki Zbawienia w Naju. #cuda https://youtu.be/fZvtqqHozwo

NVIDIA RTX Spark Debuts on Surface Laptop Ultra

#RTXSpark #SurfaceLaptopUltra #NVIDIA #BlackwellGPU #CUDA

Yesterday my colleagues @maryxek.bsky.social and @mikepapadim.bsky.social showed live demos about using NVIDIA cuFFT and cuBLAS with @tornadovm.org for signal filtering and @jitllm.io!

NVIDIA performance is a jar away from Java.

#Java #CUDA #TornadoVM

CUDA 13.1のGreen ContextsによるGPUリソースの細密制御と性能最適化

CUDA 13.1のGreen Contextsを活用し、単一GPU内でSMやワークキューを明示的に分割してレイテンシを極限まで削減する手法を解説。

#CUDA #NVIDIA #GPU #Performance #Systems

KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation

#CUDA #Triton #LLM #Package

hgpu.org?p=31343

AIM: Agentic Idea Management for Automated Research

#CUDA #LLM

hgpu.org?p=31342

🧮 AI chips keep trading away the precise math science needs.

A Japanese professor's fix: stack rough AI-grade math into one precise FP64 answer.

Now Ozaki Scheme II ships inside NVIDIA's CUDA 13.4. No new hardware needed.

#OzakiScheme #CUDA #NVIDIA #HPC #FP64

Only fools invest in #CUDA #AI from #nVidia #OpenAI #Gemini. #Deepseek is faster, it generates 800 tokens per second VS #MAGA OpenAI 300. #Huawei #China is the way to Go. youtu.be/MImgH4KMtj8?... #Trumptanic #Europe tweakers.net/nieuws/25279... Be smart go with China #Asia #Uk #Australia #Americas

In Nov 2006, a small graphics card team quietly released software that went unnoticed. Six years later, its significance became clear. This is the story of CUDA, the foundation for nearly every AI product, built with remarkable patience for a world not yet ready. #CUDA #AI #tech history

When attacks trip the gate, AMB-SCRAM issues a DMA reset to GPU VRAM, wiping payloads before write-back.

Proven on Qwen2.5-7B.

Receipts: Below

#AISafety #CUDA #MachineLearning #LLM #CyberSecurity