GPU-accelerated Exact K-Nearest Neighobor with Cosine as distance metric
- No external libraries (cuBLAS, cuGRAPH, etc.)
- You can either use Unified Memory or Unmanaged Memory (classic way)
- Copy from existing knn projects will result in immediate reject (better to do less, but do it yourself! :D )
- Submission to the contest must contain:
- Running source-code
- Presentation slides (pdf/pptx)
- Project report (we like graphs, tables, figures! aka. no screenshots of the runtimes from the terminal) explain:
- your reasoning,
- the design that you have employed,
- the options that you evaluated (you could have multiple version of your algorthims for the differnt optimizations that you have developed),
- data structures
- Show your results for all the settings in the main() function
- You can decide to explore reduced or mixed precision arithmetic operations --> only as an addition to the exact solution
- Download this repository locally and upload it as a private repository, invite me to the private one: ian-ofgod
-
17.11.2023 @23:59 Milan
- Subscribe to the contest by sending an email to: ian.didio@polimi.it with your name and possibily your teammate.
- Object: [hpdga-fall23] subscription fullname1 - fullname2
- Group of max 2 people (you can do it alone if you wish)
- If you want to be assinged a teammate just drop me an email and we will se if someone has expressed the same desire.
-
31.12.2023 @23.59 Milan
- send an email with the overall submission (slides + report + source-code)
- naming convetion for the files:
- hpdga-fall23-report-name1surname1-name2surname2
- hpdga-fall23-slides-name1surname1-name2surname2
- hpdga-fall23-code-name1surname1-name2surname2