FP8 Is All You Need (Part 1): Debunking Hardware FP64 as the HPC Holy Grail
matt_d
11 points
1 comment
June 08, 2026
Related Discussions
Found 5 related stories in 108.8ms across 10,002 title embeddings via pgvector HNSW
- 4-bit floating point FP4 chmaynard · 44 pts · April 18, 2026 · 60% similar
- The eighth-generation TPU: An architecture deep dive meetpateltech · 67 pts · April 22, 2026 · 52% similar
- BPF support in GCC 16 and beyond tuananh · 16 pts · June 02, 2026 · 49% similar
- Show HN: FPGA soft-core of the Saab Viggen's 1963 airborne computer FormerLabFred · 18 pts · March 20, 2026 · 49% similar
- NanoGPT Slowrun: 10x Data Efficiency with Infinite Compute sdpmas · 122 pts · March 19, 2026 · 49% similar
Discussion Highlights (1 comments)
ebiederm
Assuming this is correct that is a very intriguing result. FP64 emulated with FP8 running faster than the native FP64 implementation.