120
edits
Changes
→Run Times
=== Run Times ===
We have decided to include the information about GPU in our final output as we were testing it on different graphics cards.<br/>
[[File:file14.png|700px]][[File:file15resultPixels.png|700px]] <br/><br/>
Looking at the Matrix size of 10,000 x 10,000 the LU decomposition on GPU runs about <b>39 times</b> faster then CPU, which is substantial improvements in the runtimes. <br/><br/>
It is also interesting to note the difference in runtime between LU, QR and Cholesky methods. <br/>
[[File:resultPixels2.png|500px]]
===== QR =====
=====Cholesky =====
Uses a symmetrical matrix that satisfies the relationship A^-1 = A^Tfor ... Which basically means that the data is mirrored over diagonal and therefore takes twice as fast to run as LU
----
<br/><br/> The complete code can be found at Github repo [https://github.com/YuriyKartuzov/cuSOLVER HERE]
