Hackernews posts about FP64
- The AI era is pulling FP64 hardware away from scientific HPC (fortran-lang.discourse.group)
- 15 years of FP64 segmentation, and why the Blackwell Ultra breaks the pattern (nicolasdickenmann.com)
- Nvidia Blackwell Ultra (GB300): -97% INT8/FP64,+50% FP4 Dense,+55% VRAM,+114% At (resources.nvidia.com)
- AMD Hints at Big FP64 Increases in MI430X GPU as Ozaki Underwhelms (www.hpcwire.com)
- FP64 Floating-Point Emulation in INT8 (arxiv.org)
- SC25: Estimating AMD's Upcoming MI430X's FP64 and the Discovery Supercomputer (chipsandcheese.com)
- SC25: Estimating AMD's Upcoming MI430X's FP64 and the Discovery Supercomputer (chipsandcheese.com)
- SC25: Estimating AMD's Upcoming MI430X's FP64 and the Discovery Supercomputer (chipsandcheese.com)
- Bolt Graphics Targets FP64 HPC Workloads with Zeus GPU (www.hpcwire.com)
- 4-bit floating point FP4 (www.johndcook.com)
- Flux runs 2x faster on 10GB VRAM with Nvidia FP4 (blackforestlabs.ai)
- Custom FP4 CUDA Kernel – 129 Tflops on DGX Spark with Pre-Quantized Weight Cache (forums.developer.nvidia.com)
- Sol-RL: FP4 Explore, BF16 Train (nvlabs.github.io)
- Tflops Gap: Why FP4 Moe Kernel Engineering Matters on Blackwell (huggingface.co)
- Huawei unveils Atlas 950 SuperCluster – promises 1 ZettaFLOPS FP4 performance (www.tomshardware.com)
- One Nuclear War Can Ruin the Whole Climate (www.wsj.com)