OpenBSD is so fast, I had to modify the program slightly to measure itself(flak.tedunangst.com)

91 points by Bogdanp 1 year ago | 66 comments

Doas – dedicated OpenBSD application subexecutor(flak.tedunangst.com)

76 points by osener 3 years ago | 32 comments

Fan Service(flak.tedunangst.com)

133 points by todsacerdoti 1 year ago | 29 comments

Ask HN: Resources for GPU Compilers?

61 points by zvikinoza 2 years ago | 17 comments

Async DNS(flak.tedunangst.com)

111 points by todsacerdoti 10 months ago | 37 comments

Tell HN: GpuOwl/PRPLL, GPU software used to find the largest prime number

63 points by mpreda 2 years ago | 33 comments

Ollama now supports AMD graphics cards(ollama.com)

578 points by tosh 2 years ago | 194 comments

Ollama blog post from March 14, 2024 announcing preview support for running models on AMD graphics cards.

AMD outsells Intel in the datacenter space(tomshardware.com)

426 points by baal80spam 2 years ago | 140 comments

Tom's Hardware news article reporting that AMD's datacenter sales surpassed Intel's for the first time, though both trail Nvidia's AI GPU sales.

How did AMD Ryzen get 50% faster in two years?(lemire.me)

466 points by ibobev 1 month ago | 190 comments

Daniel Lemire blog post examining how AMD Ryzen processors became about 50% faster over two years.

AMD may get across the CUDA moat(hpcwire.com)

498 points by danzheng 3 years ago | 266 comments

HPCwire article by Doug Eadline discussing how AMD's Instinct GPUs and software stack might overcome Nvidia's CUDA software lock-in for GenAI workloads.

Zenbleed(lock.cmpxchg8b.com)

1160 points by loeg 3 years ago | 327 comments

Tavis Ormandy's writeup of Zenbleed, an AMD Zen 2 CPU vulnerability that leaks data from vector registers, covering exploitation, discovery, and mitigation.

AMD acquires Taalas to boost inference performance by etching models in silicon(theregister.com)

902 points by itvision 2 months ago | 681 comments

The Register reports AMD has acquired AI chip startup Taalas, which hardwires model weights into silicon for fast inference, reaching up to 17,000 tokens per second.