OpenCAPI-based Image Analysis Pipeline for 18 GB/s kilohertz-framerate X-ray...

on YouTube ↗31m

" summary

A data pipeline ingests the Jungfrau detector stream, decodes per-pixel three-mode ADC values, and writes compressed frames into host memory; a CPU-only implementation tops out at 5 GB/s on CPU requiring four big servers, while an OpenCAPI + FPGA design (started with SNAP, moved to ocxl/oc_accel and HLS C/C++) reaches 18 GB/s with two FPGAs, uses ~35 W per FPGA, scales toward 50 GB/s, and already collects 2000 images in one second to solve a crystal.

AI-written summary · about

" chapters (25)

1. 0:00Introduction
2. 0:25Overview
3. 0:57Xrays
4. 2:29Xray crystallography
5. 5:33Friction lasers
6. 5:57Biological crystallography
7. 7:50Xray detector
8. 8:09Hybrid pixel detectors
9. 9:02New megapixel detector
10. 9:55Dynamic range
11. 10:28Modes
12. 11:36Light Source
13. 12:20Data throughput
14. 12:37Data transfer
15. 13:44throughput
16. 14:20Power9 Architecture
17. 15:32Snap
18. 17:08Energy efficiency
19. 17:40Processing throughput
20. 18:14Memory coherence
21. 19:29Continuous integration pipeline
22. 20:10Results
23. 21:18Challenges
24. 22:11Thank you
25. 24:47Questions