@GergelyOrosz: Pretty incredible how turbopuffer are showing off how their new architecture is up to 128x slower than the old architec…

X AI KOLs Following News
vector-database performance-optimization architecture migration search-engine tech-blogging

Summary

Gergely Orosz highlights turbopuffer openly sharing their v3 architecture migration, where the new design was initially up to 128x slower than the old one, with each optimization (like batched reads yielding a ~9x improvement) publicly documented.

Pretty incredible how turbopuffer are showing off how their new architecture is up to 128x slower than the old architecture… before the optimisations they are doing. And they are doing + blogging about each optimization Not seen a migration shared like this before, love it
Original Article
View Cached Full Text

Cached at: 10/03/26, 10:55 AM

Pretty incredible how turbopuffer are showing off how their new architecture is up to 128x slower than the old architecture… before the optimisations they are doing.

And they are doing + blogging about each optimization

Not seen a migration shared like this before, love it

turbopuffer (@turbopuffer): we implemented batched reads in the v3 full-text engine, leading to a ~9x performance improvement. postings are stored as columns, and we were reading them a row at a time

sometimes optimization is not about being clever, but simply avoiding mistakes

Similar Articles

@reprompting: reading about the NVIDIA hopper architecture today https://arxiv.org/pdf/2501.12084

X AI KOLs Timeline

这篇论文对 NVIDIA Hopper GPU 架构进行了多层级微基准测试分析，评估了 L2 分区缓存、第四代张量核心（FP8）、DPX 指令、分布式共享内存（DSM）和张量内存加速器（TMA）等新特性的性能表现，结果显示 TMA 异步编程可实现 1.5 倍矩阵乘法加速，FP8 性能接近 FP16 的两倍，DPX 指令可加速生物信息学算法至少 4.75 倍。